Optical module life prediction method and device

Abstract

The present application provides a method and device for predicting the life of an optical module. The method comprises: performing self-supervised training on the original iTransformer model using unlabeled training samples; modifying the prediction head of the self-supervised trained iTransformer model to a regression head to obtain an improved iTransformer model; and performing supervised training on the improved iTransformer model using labeled training samples to obtain a life prediction model. Through the above training method, the reliance on labeled training samples is significantly reduced, so that only a small number of labeled training samples are required to achieve good training results, ensuring the reliability and accuracy of optical module life prediction.

Description

Technical Field

[0001] The present application relates to the field of network communication technology, and in particular to a method and device for predicting the life of an optical module. Background Art

[0002] Data transmission and reception on network interfaces all require optical modules. Long-term operation of optical modules can cause performance degradation of optical components, leading to link instability. This unstable, "sub-healthy" state can compromise the integrity of data transmission and reception. Traditional O&M methods are unable to identify risks and provide early warnings before optical module aging and failure occur. This sub-healthy state of optical modules degrades the quality of service provided by the network, placing the network in a critical state of intermittent interruptions between "available" and "unavailable," significantly impacting the perceived quality of service.

[0003] Therefore, how to quickly and effectively operate and maintain a large number of optical modules in the network has become a challenge for network operation and maintenance. Summary of the Invention

[0004] To overcome the problems existing in the related art, the present application provides a method and device for predicting the life of an optical module.

[0005] According to a first aspect of an embodiment of the present application, a method for predicting the life of an optical module is provided, the method comprising:

[0006] Use unlabeled training samples to perform self-supervised training on the original iTransformer model;

[0007] The prediction head of the self-supervised iTransformer model is modified to a regression head to obtain an improved iTransformer model;

[0008] Using labeled training samples to perform supervised training on the improved iTransformer model to obtain a lifespan prediction model;

[0009] In which, the life prediction model is used to predict the remaining life of the optical module, the unlabeled training sample includes the feature vector of the optical module within the first time span, and the labeled training sample includes the feature vector of the optical module within the first time span and the remaining life label of the optical module at the last time point within the first time span; the feature vector within the first time span includes feature vectors of multiple time points, and the feature vector of each time point is calculated based on the indicator data of the optical module.

[0010] According to a second aspect of an embodiment of the present application, a device for predicting the life of an optical module is provided, the device comprising:

[0011] The self-supervised training module is used to perform self-supervised training on the original iTransformer model using unlabeled training samples;

[0012] The model modification module is used to modify the prediction head of the iTransformer model that has undergone self-supervised training into a regression head to obtain an improved iTransformer model;

[0013] A supervised training module, configured to perform supervised training on the improved iTransformer model using labeled training samples to obtain a lifespan prediction model;

[0014] In which, the life prediction model is used to predict the remaining life of the optical module, the unlabeled training sample includes the feature vector of the optical module within the first time span, and the labeled training sample includes the feature vector of the optical module within the first time span and the remaining life label of the optical module at the last time point within the first time span; the feature vector within the first time span includes feature vectors of multiple time points, and the feature vector of each time point is calculated based on the indicator data of the optical module.

[0015] According to a third aspect of the embodiments of the present application, there is provided an electronic device, including:

[0016] A memory, one or more processors; the memory is coupled to the processor; wherein computer program code is stored in the memory, the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the method as described above.

[0017] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, comprising computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the method described above.

[0018] According to a fifth aspect of the embodiments of the present application, a computer program product is provided. When the computer program product is run on a computer, the computer is caused to execute the method described above.

[0019] The technical solutions provided by the embodiments of the present application may have the following beneficial effects:

[0020] In this embodiment, the optical module's lifespan is first predicted based on its own indicator data. This not only improves the accuracy of the lifespan prediction, but also avoids maintenance work on different types of optical modules. Furthermore, considering that the temporal distribution of each optical module indicator data is equivalent to a time series, an innovative time series algorithm is proposed to solve the lifespan prediction problem. This fully exploits the long-range dependencies between the various optical module indicators, improving the accuracy and reliability of the lifespan prediction.

[0021] Finally, in view of the difficulty in obtaining a large number of labeled training samples, and the fact that the training method and output results of the original iTransformer model are not suitable for solving the life prediction problem, the embodiment of the present application implements model training in the following manner: using two training steps, first using a large number of unlabeled training samples to perform self-supervised training on the original iTransformer model, then modifying the model's prediction head to a regression head to obtain an improved iTransformer model, and finally using a small number of labeled training samples to perform supervised training on the improved iTransformer model to obtain a life prediction model. Through the above training method, the dependence on labeled training samples is significantly reduced, so that only a small number of labeled training samples are needed to achieve good training results, ensuring the reliability and accuracy of optical module life prediction.

[0022] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0024] Figure 1 A schematic diagram of a first flow chart of the optical module life prediction method provided in an embodiment of the present application;

[0025] Figure 2 A second flow chart of the optical module life prediction method provided in an embodiment of the present application;

[0026] Figure 3 A third flow chart of the optical module life prediction method provided in an embodiment of the present application;

[0027] Figure 4 A fourth flow chart of the optical module life prediction method provided in an embodiment of the present application;

[0028] Figure 5 A fifth flow chart of the optical module life prediction method provided in an embodiment of the present application;

[0029] Figure 6 A sixth flow chart of the optical module life prediction method provided in an embodiment of the present application;

[0030] Figure 7 This is a functional diagram of the optical module life prediction device provided in an embodiment of the present application;

[0031] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application are described below in conjunction with the accompanying drawings in the embodiments of the present application. In the description of the embodiments of the present application, the terms used in the following embodiments are only for the purpose of describing specific embodiments and are not intended to limit the present application.

[0033] Current optical module lifespan prediction solutions don't consider actual performance data for individual optical modules, relying solely on statistical data from similar modules. Consequently, their accuracy is low and their usefulness is limited. Furthermore, the algorithms require timely updates to the training dataset and the constant addition of support for new optical module types, resulting in high maintenance costs.

[0034] To address the above issues, this application provides a method and device for predicting the lifespan of an optical module. This method actively monitors the health of optical modules and predicts their remaining lifespan (i.e., the time until a failure occurs). This helps network administrators identify potential link failures before system performance is impacted, thereby proactively identifying and eliminating potential faults. Using the remaining lifespan alert, network administrators can switch services to backup links or replace suspected components, allowing system repairs without interrupting service.

[0035] Specifically, this application achieves the goal of optical module life prediction based on the improved iTransformer model, and through two innovative training steps, reduces the demand for large-scale fault optical module data, effectively saves training time, and improves training results.

[0036] It should be pointed out here that the failure modes of optical modules can be roughly divided into two types: natural aging and internal failure. The latter is a sudden internal failure that is unpredictable. This application mainly focuses on the former, which is caused by laser aging.

[0037] The embodiments of the present application are described in detail below.

[0038] The present application embodiment provides a method for predicting the life of an optical module. Figure 1 As shown, the method may include the following steps:

[0039] Step 110: Perform self-supervised training on the original iTransformer model using unlabeled training samples.

[0040] Step 120: Modify the prediction head of the iTransformer model that has undergone self-supervised training into a regression head to obtain an improved iTransformer model;

[0041] Step 130: Perform supervised training on the improved iTransformer model using labeled training samples to obtain a lifespan prediction model;

[0042] In which, the life prediction model is used to predict the remaining life of the optical module, the unlabeled training sample includes the feature vector of the optical module within the first time span, and the labeled training sample includes the feature vector of the optical module within the first time span and the remaining life label of the optical module at the last time point within the first time span; the feature vector within the first time span includes feature vectors of multiple time points, and the feature vector of each time point is calculated based on the indicator data of the optical module.

[0043] In the embodiment of the present application, the feature vector of the optical module at any time point is obtained by performing feature engineering on the index data of the optical module. The embodiment of the present application periodically collects the index data of the optical module.

[0044] For example, the indicator data of the optical module can be continuously collected through network devices such as switches. Common indicator data includes data based on interface statistics, such as voltage, and data based on channel statistics, such as temperature, bias current, receive power, and transmit power.

[0045] Assume there is a 4-channel optical module, its indicator data includes:

[0046] Voltage + 4 channel x (Temp, Current Bias, Rx Power, Tx power)

[0047] Furthermore, optical module port data is also highly valuable for lifespan prediction. Therefore, as a preferred implementation, this embodiment further collects some port-based statistical data from the optical module, such as port error packet data and FEC (Forward Error Correction) data. Specifically, the port error packet data may include the number of port error packets, and the FEC data may include the number of error packets corrected by FEC and the number of error packets that could not be corrected by FEC.

[0048] Therefore, combined with the data based on port statistics, the indicator data of the optical module includes:

[0049] Voltage + #Channel x (Temp, Current Bias, Rx Power, Tx power) + in_error_packets + FEC_Correctable + FEC_UnCorrectable

[0050] For example, if optical module indicator data is collected every six hours, indicator data for four time points can be collected per day. Assuming the first time span is three months (90 days), the first time span includes 360 time points, and the number of feature vectors within the first time span is 360.

[0051] As a specific implementation method, after collecting various indicator data of the optical module, for each indicator data, the embodiment of the present application obtains the characteristic vector of the optical module by calculating its characteristic value presented within a period of time, such as gradient, mean, mean square deviation STD, minimum value Min, maximum value Max, etc.

[0052] Therefore, the feature vector of an optical module at any time point can be expressed as follows:

[0053] [Voltage + #Channel x (Temp, Current Bias, Rx Power, Tx power) + in_error_packets + FEC_Correctable + FEC_UnCorrectable] x [Gradient, mean, std,min, max]

[0054] In the above formula, in_error_packets represents the number of error packets on the port, FEC_Correctable represents the number of error packets that can be automatically corrected by FEC technology, and FEC_unCorrectable represents the number of error packets that cannot be automatically corrected by FEC technology.

[0055] Illustratively, the embodiment of the present application calculates the characteristic vector of the optical module based on the index data of the optical module within the second time span, and the second time span can be set to 3 days.

[0056] It is understandable that in practical applications, the dimension of the feature vector needs to be set in advance. Considering that the maximum number of channels of the optical module is 4, the embodiment of the present application sets the dimension of the feature vector of the optical module to 100 for example.

[0057] It's worth noting that in real-world applications, some optical modules may only have one channel, and some may not support FEC data. In these cases, the corresponding positions in the feature vector can be set to zero. This ensures that the dimensions of the feature vector for each optical module are constant at each time point.

[0058] In short, one can Figure 2 Generate the feature vector of the optical module at any time point in the following manner:

[0059] Step 210: For the current time point, obtain indicator data of the optical module in a second time span closest to the current time point, where the current time point is the last time point in the second time span;

[0060] The above-mentioned current time point may be any time point, but within the second time span, the current time point is the last time point within the second time span.

[0061] Step 220: Calculate the characteristic values ​​of each indicator data in the second time span;

[0062] The above-mentioned characteristic value includes at least one of the change rate, mean value, mean square error, minimum value, and maximum value.

[0063] Step 230: Generate a feature vector of the optical module at the current time point based on the feature value.

[0064] Exemplarily, a feature vector of the optical module at the current time point is generated according to a preset feature dimension size, such as 100.

[0065] As mentioned above, the feature vectors within the first time span are 360 ​​feature vectors. Since the indicator data of the optical module is collected periodically, each dimension of the 360 ​​feature vectors generated based on the indicator data can be regarded as a time series of a variable. Assuming that the dimension of the feature vector is 100, these 360 ​​feature vectors can be regarded as a time series of 100 variables. In other words, the model input can be regarded as a time series of 100 variables. Therefore, the embodiment of the present application considers the use of a time series algorithm to achieve the life prediction goal. Specifically, the embodiment of the present application achieves the life prediction goal based on the improved iTransformer model.

[0066] iTransformer is a multivariate time series prediction model algorithm with a strong ability to capture long-range dependencies. In terms of model architecture, iTransformer uses MPL Projection (Multi-layer Perceptual Mapping Layer) to output prediction results. However, this algorithm is a prediction algorithm and cannot be used directly for classification, and its training method uses a self-supervised approach. These are not suitable for the lifespan prediction goal of the present embodiment.

[0067] Therefore, the present embodiment makes the following modifications to the iTransformer model's network structure: the final prediction header (MPL Projection) is modified to a regression header. The difference between the prediction header and the regression header is that the former maps the feature vectors extracted by the backbone network to the output format required for the prediction task, generally a numerical sequence; the latter is specifically used for regression tasks, mapping feature vectors to continuous numerical outputs. For example, in the present embodiment, the model output is a predicted numerical value for the remaining lifespan.

[0068] In addition, for the improved iTransformer model, the embodiment of the present application adopts a supervised training method.

[0069] As a specific implementation method, the present application embodiment is as follows Figure 3 The steps shown are for supervised training:

[0070] Step 310: construct a target loss function, and use the target loss function to calculate the mean square error between the life prediction value and the remaining life label, wherein the life prediction value is predicted by the improved iTransformer model based on the labeled training samples;

[0071] Step 320: Update the weight parameters of the improved iTransformer model by gradient backpropagation according to the mean square error calculated by the target loss function.

[0072] It should be noted that deep learning algorithms generally need to learn a large amount of training data to achieve excellent practical results. However, the data of faulty optical modules is very valuable, and it is very difficult to collect a large amount of such data. Therefore, the embodiment of the present application proposes two training steps:

[0073] First, the original iTransformer model is trained in a self-supervised manner using unlabeled training samples. This step does not require data from faulty optical modules or data labeling, allowing the network to effectively learn the temporal characteristics of multivariate time series and discover correlations between different variables.

[0074] Next, building on the previous step, the final prediction head of the network model was modified into a regression head, resulting in an improved iTransformer model. This improved iTransformer model was then trained in a supervised manner using labeled training samples to obtain a lifespan prediction model.

[0075] This training method can reduce the demand for large-scale fault optical module data, effectively save training time, and improve training results.

[0076] For example, a large amount of basic training data (unlabeled training samples) is first collected and produced for offline self-supervised training. Then, a small amount of faulty optical module data is used to produce labeled training samples for offline supervised training to obtain a life prediction model. During model deployment, network operation and maintenance generates new faulty optical module data, and incremental training data (labeled training samples) are produced in real time based on this data for online tuning to improve the model's adaptability to actual scenarios and prediction accuracy.

[0077] The following describes the process of collecting optical module indicator data.

[0078] Typically, the frequency of collecting optical module indicator data is high, usually set to once every 30 seconds, once every 5 minutes, etc. Considering that optical modules age gradually and develop slowly, there is no need to monitor these indicator data very frequently in the context of life prediction.

[0079] Therefore, illustratively, the embodiment of the present application collects the index data of the optical module once every 6 hours, collects the data of the optical module at 4 time points every day, and collects the index data of the optical module at 188*4=752 time points in half a year.

[0080] To achieve life prediction, the embodiment of the present application cyclically maintains the indicator data of the optical module at each time point within the latest third time span, and the third time span must be guaranteed to be greater than half a year.

[0081] It's worth noting that the emphasis on a period greater than six months is based on statistical and practical experience, indicating that optical modules with a remaining lifespan of more than six months generally perform well. By cyclically maintaining the optical module's indicator data for the past six months, the embodiments of this application can observe the long-term trends in these indicators, allowing for a more accurate assessment of the optical module's health. For example, if the optical module's transmitted optical power gradually decreases or its extinction ratio gradually decreases over six months, this could be an early sign of impending failure.

[0082] In other words, if an optical module's indicators remain stable over six months, its performance is reliable and can continue to be used. Conversely, if the indicators fluctuate or decrease significantly, the probability of failure is high. From a cost-effectiveness perspective, storing data for six months can provide sufficient effective information for lifespan prediction and fault diagnosis without excessively increasing storage costs.

[0083] For example, the third time span may be set to 300 days. In this embodiment of the present application, the indicator data of the optical module at each time point in the last 300 days is cyclically maintained, that is, the indicator data at the last 1200 time points is maintained.

[0084] The following describes the training process of the lifespan prediction model.

[0085] Exemplarily, the model training process may include two stages: an offline training stage and an online training stage.

[0086] In the offline training phase, Figure 1 As shown in the figure, the offline training process includes: using unlabeled training samples to perform self-supervised training on the original iTransformer model; modifying the prediction head of the self-supervised trained iTransformer model to a regression head to obtain an improved iTransformer model; and using labeled training samples to perform supervised training on the improved iTransformer model to obtain a lifespan prediction model.

[0087] After that, the life prediction model is deployed to the operating environment and enters the online training phase. Figure 4 As shown in Figure 2, the online training process includes:

[0088] Step 410: cyclically maintain the index data of the optical module within the latest third time span;

[0089] Step 420: If a fault is detected in the optical module, a new labeled training sample is generated based on the cyclically maintained indicator data and the time of the fault.

[0090] Step 430: Use the newly added labeled training samples to perform online tuning on the life prediction model.

[0091] Specifically, the network operations and maintenance system monitors the status of each optical module. Optical module failures typically correspond to a sharp deterioration in health indicators, such as a sharp increase in packet error rate and abnormal key indicator verification, which can impact related services. Upon detecting an optical module failure, step 420 is executed to extract recent indicator data for the faulty optical module, along with the time of the failure, to create labeled training samples.

[0092] For example, we cyclically maintain the indicator data of the optical module at the most recent 1200 time points. The goal is to use the indicator data of the optical module in the most recent 3 months (90 days) to predict its remaining lifespan. 90 days of indicator data corresponds to 90 x 4 = 360 time points. Therefore, the specific method for creating training samples is as follows:

[0093] The indicator data for the faulty optical module at 1200 time points is grouped into groups of 360 consecutive time points. The corresponding remaining lifetime label for each group is the number of days from the last time point in the group to the time of failure. This results in 840 groups: 1200 - 360 = 840. The remaining lifetime label for the first group is 840 / 4 = 210, indicating a remaining lifetime of 210 days. The remaining lifetime label for the 600th group is 240 / 4 = 60, indicating a remaining lifetime of 60 days. And so on. This yields a total of 840 labeled training samples, each consisting of 360 100-dimensional feature vectors and a remaining lifetime label.

[0094] The life prediction process based on the life prediction module is explained below.

[0095] It is understandable that after the life prediction model is deployed in the operating environment, the life prediction model can be used to predict the remaining life of the optical module, that is, the time when the optical module will fail can be predicted. Specifically, the prediction process is as follows Figure 5 As shown:

[0096] Step 510: When a life prediction instruction is received or the system time reaches a preset time node, a feature vector of the optical module in the most recent first time span is generated based on the cyclically maintained indicator data, and input into the life prediction model to obtain the remaining life of the optical module.

[0097] Specifically, during network operation and maintenance, the remaining life of the optical module is predicted according to a preset period, for example, once a day.

[0098] For example, the third time span is set to 300 days, and the first time span is set to 3 months. Based on the cyclically maintained indicator data for the last 300 days, feature vectors are generated for each time point within the last three months, that is, feature vectors for the last 360 time points. These feature vectors are input into the lifespan prediction model to determine the remaining lifespan of the optical module. This predicted value provides effective reference and support for network operations and maintenance.

[0099] A specific embodiment is provided below to further illustrate the optical module life prediction method of the embodiment of the present application.

[0100] Illustratively, in this embodiment, the first time span is set to 3 months, the second time span is set to 3 days, and the third time span is set to 300 days; the indicator data of the optical module is collected every 6 hours, and each collection time is called a time point.

[0101] The overall process is as follows Figure 6 As shown:

[0102] First, the model input and model output are determined based on the lifetime prediction requirements: the model input is the feature vector of the optical module within three months, and the model output is the predicted value of the remaining lifetime of the optical module.

[0103] Determine the calculation method of the characteristic vector: At any time point, obtain the indicator data of the optical module for the three days before the time point, calculate the characteristic values ​​of each indicator data during these three days, and obtain the characteristic vector of the optical module at the time point.

[0104] Select an algorithm for lifespan prediction: Select the iTransformer model. To achieve the lifespan prediction goal, modify its prediction head to a regression head and adopt a supervised training method.

[0105] Create unlabeled training samples and perform self-supervised training: Collect the optical module's metrics data and generate feature vectors for each time point using the method described above. Use the feature vectors from every three months as an unlabeled training sample. Generate a large number of unlabeled training samples and use them to perform self-supervised training on the original iTransformer model.

[0106] Modify the model: Modify the prediction head of the self-supervised iTransformer model to a regression head to obtain an improved iTransformer model.

[0107] Create labeled training samples: Collect indicator data and the time of the faulty optical module. Generate feature vectors for the optical module at each time point using the method described above. The feature vector and remaining life label for every three months serve as a labeled training sample. The remaining life label is calculated as the difference between the last time point within three months and the time of the fault.

[0108] Supervised training: The improved iTransformer model is trained using labeled training samples to obtain a RUL prediction model. Specifically, a target loss function is constructed to calculate the mean squared error between the predicted lifespan and the RUL label. The mean squared error of the labeled training samples is calculated based on the target loss function. The weight parameters of the improved iTransformer model are updated through gradient backpropagation to obtain the RUL prediction model.

[0109] Deploy the remaining life prediction model to the operating environment and cyclically maintain the indicator data of the optical module within 300 days.

[0110] When the preset life prediction time node is reached, or when a life prediction command is received, life prediction is implemented in the following manner: based on the indicator data of cyclic maintenance, the feature vector of the optical module in the last three months is generated and input into the life prediction model to obtain the remaining life of the optical module.

[0111] When an optical module failure is detected, new labeled training samples are created and online tuning is performed in the following manner: Based on the indicator data of the faulty optical module and the time of the failure, new labeled training samples are created in accordance with the aforementioned method for creating labeled training samples, and the life prediction model is tuned online using the new labeled training samples.

[0112] In summary, feature engineering is performed on the optical module indicator data at each time point. The characteristics of each indicator over a period of time, such as the rate of change (gradient), mean, mean square error, minimum value, and maximum value, are calculated to obtain a 100-dimensional feature vector. The indicator data of the optical module is maintained cyclically for at least six months, with data collected four times a day (one data point every six hours). For six months, 188*4=752 data points need to be maintained, and a maximum of 1200 data points are maintained (cyclically). If a fault is detected in an optical module, 1200 data points and the time of the fault are extracted, and labeled training samples corresponding to the optical module are generated based on these data. These labeled training samples are used to train the algorithm model. This training process can be offline or online during network operation and maintenance. The algorithm uses an improved regression model based on the iTransformer architecture. During network operation and maintenance, the remaining life of the optical module can be regularly predicted. For example, the indicator data of each optical module at the last 360 time points can be used daily to predict the remaining life of the optical module. Through these measures, the temporal and spatial correlation of the optical module's various indicator data is effectively utilized, and the model algorithm is continuously optimized by combining offline and online training to improve the accuracy of remaining life prediction.

[0113] Based on the same inventive concept, this application also provides an optical module life prediction device, such as Figure 7 As shown, the device includes:

[0114] A self-supervised training module 710 is used to perform self-supervised training on the original iTransformer model using unlabeled training samples;

[0115] A model modification module 720 is used to modify the prediction head of the iTransformer model that has undergone self-supervised training into a regression head to obtain an improved iTransformer model;

[0116] A supervised training module 730 is used to perform supervised training on the improved iTransformer model using labeled training samples to obtain a lifespan prediction model;

[0117] In which, the life prediction model is used to predict the remaining life of the optical module, the unlabeled training sample includes the feature vector of the optical module within the first time span, and the labeled training sample includes the feature vector of the optical module within the first time span and the remaining life label of the optical module at the last time point within the first time span; the feature vector within the first time span includes feature vectors of multiple time points, and the feature vector of each time point is calculated based on the indicator data of the optical module.

[0118] As a specific implementation, the apparatus further includes a feature vector generation module, which generates a feature vector of the optical module at a current time point in the following manner:

[0119] Obtain indicator data of the optical module within a second time span closest to the current time point; the indicator data include data based on interface statistics, data based on channel statistics, and data based on port statistics; the current time point is the last time point in the second time span; calculate the characteristic values ​​of each item of the indicator data in the second time span, the characteristic values ​​including at least one of the rate of change, mean, mean square error, minimum value, and maximum value; based on the characteristic values, generate a characteristic vector of the optical module at the current time point.

[0120] As a specific implementation, the supervised training module 730 performs supervised training on the improved iTransformer model in the following manner:

[0121] Constructing a target loss function and using the target loss function to calculate the mean square error between the life prediction value and the remaining life label, wherein the life prediction value is predicted by the improved iTransformer model based on the labeled training samples; updating the weight parameters of the improved iTransformer model by gradient backpropagation based on the mean square error calculated by the target loss function.

[0122] As a specific embodiment, the device further includes an online tuning module, which performs online tuning on the life prediction model in the following manner:

[0123] Cyclic maintenance is performed on the indicator data of the optical module within a recent third time span; the third time span is greater than the first time span; if a failure of the optical module is detected, new labeled training samples are generated based on the cyclically maintained indicator data and the time when the failure occurred; and the life prediction model is online tuned using the new labeled training samples.

[0124] As a specific embodiment, the device further includes a lifespan prediction module, which is configured to:

[0125] Upon receiving the life prediction instruction, generating a feature vector of the optical module within a most recent first time span according to the cyclically maintained indicator data, and inputting the feature vector into the life prediction model to obtain the remaining life of the optical module; and / or,

[0126] When the system time reaches the preset life prediction time node, the feature vector of the optical module in the most recent first time span is generated according to the cyclically maintained indicator data, and is input into the life prediction model to obtain the remaining life of the optical module.

[0127] An embodiment of the present application provides an electronic device that may include a memory and one or more processors. The memory is configured to store computer program code, which includes computer instructions. When the processor executes the computer instructions, the electronic device may perform the functions or steps of the above-described method embodiments.

[0128] The structure of the electronic device can refer to Figure 8 The structure of the electronic device 100 is shown.

[0129] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0130] An embodiment of the present application further provides a computer-readable storage medium, which includes computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes each function or step of the above method embodiment.

[0131] The computer-readable storage medium includes but is not limited to any one of the following: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0132] The embodiment of the present application further provides a computer program product, which, when executed on a computer, enables the computer to execute the functions or steps of the above method embodiment.

[0133] The electronic device, computer-readable storage medium, and computer program product provided in the embodiments of the present application are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects of the corresponding methods provided above, and will not be repeated here.

[0134] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes or replacements. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for predicting the life of an optical module, characterized in that: The method comprises: Use unlabeled training samples to perform self-supervised training on the original iTransformer model; The prediction head of the self-supervised iTransformer model is modified to a regression head to obtain an improved iTransformer model; Using labeled training samples to perform supervised training on the improved iTransformer model to obtain a lifespan prediction model; In which, the life prediction model is used to predict the remaining life of the optical module, the unlabeled training sample includes the feature vector of the optical module within the first time span, and the labeled training sample includes the feature vector of the optical module within the first time span and the remaining life label of the optical module at the last time point within the first time span; the feature vector within the first time span includes feature vectors of multiple time points, and the feature vector of each time point is calculated based on the indicator data of the optical module.

2. The method according to claim 1, characterized in that The method further includes generating a feature vector of the optical module at the current time point by: Obtaining indicator data of the optical module within a second time span closest to the current time point; the indicator data includes data based on interface statistics, data based on channel statistics, and data based on port statistics; the current time point is the last time point in the second time span; Calculating characteristic values ​​of each indicator data in the second time span, wherein the characteristic values ​​include at least one of a rate of change, a mean, a mean square error, a minimum value, and a maximum value; Based on the characteristic value, a characteristic vector of the optical module at the current time point is generated.

3. The method according to claim 1, characterized in that The method specifically performs supervised training on the improved iTransformer model in the following manner: Constructing a target loss function, and calculating the mean square error between a lifespan prediction value and a remaining lifespan label using the target loss function, wherein the lifespan prediction value is predicted by the improved iTransformer model based on the labeled training samples; According to the mean square error calculated by the target loss function, the weight parameters of the improved iTransformer model are updated by gradient backpropagation.

4. The method according to claim 1, wherein The method further includes online tuning of the life prediction model by: cyclically maintaining indicator data of the optical module within a recent third time span; the third time span is greater than the first time span; If an optical module failure is detected, new labeled training samples are generated based on the cyclic maintenance indicator data and the time of the failure. The life prediction model is online tuned using the newly added labeled training samples.

5. The method according to claim 1, wherein The method further comprises: Upon receiving the life prediction instruction, generating a feature vector of the optical module within a most recent first time span according to the cyclically maintained indicator data, and inputting the feature vector into the life prediction model to obtain the remaining life of the optical module; and / or, When the system time reaches the preset life prediction time node, the feature vector of the optical module in the most recent first time span is generated according to the cyclically maintained indicator data, and is input into the life prediction model to obtain the remaining life of the optical module.

6. An optical module life prediction device, characterized in that: The device comprises: The self-supervised training module is used to perform self-supervised training on the original iTransformer model using unlabeled training samples; The model modification module is used to modify the prediction head of the iTransformer model that has undergone self-supervised training into a regression head to obtain an improved iTransformer model; A supervised training module, configured to perform supervised training on the improved iTransformer model using labeled training samples to obtain a lifespan prediction model; In which, the life prediction model is used to predict the remaining life of the optical module, the unlabeled training sample includes the feature vector of the optical module within the first time span, and the labeled training sample includes the feature vector of the optical module within the first time span and the remaining life label of the optical module at the last time point within the first time span; the feature vector within the first time span includes feature vectors of multiple time points, and the feature vector of each time point is calculated based on the indicator data of the optical module.

7. The device according to claim 6, characterized in that The device further includes a feature vector generating module, which generates a feature vector of the optical module at a current time point by: Obtaining indicator data of the optical module within a second time span closest to the current time point; the indicator data includes data based on interface statistics, data based on channel statistics, and data based on port statistics; the current time point is the last time point in the second time span; Calculating characteristic values ​​of each indicator data in the second time span, wherein the characteristic values ​​include at least one of a rate of change, a mean, a mean square error, a minimum value, and a maximum value; Based on the characteristic value, a characteristic vector of the optical module at the current time point is generated.

8. An electronic device, characterized in that: include: A memory, one or more processors; the memory is coupled to the processor; wherein computer program code is stored in the memory, the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 5.

9. A computer-readable storage medium comprising computer instructions, characterized in that: When the computer instructions are executed on an electronic device, the electronic device is caused to execute the method according to any one of claims 1 to 5.

10. A computer program product, characterized in that When the computer program product is run on a computer, the computer is caused to perform the method according to any one of claims 1 to 5.

Images