Failure prediction method and device for optical network component

Abstract

The invention provides a failure prediction method and device for an optical network component. The method comprises the following steps: acquiring target working performance index data of a to-be-predicted optical network component; taking the target working performance index data as the input of a failure prediction model, and outputting a failure prediction result; based on the failure prediction result, a risk monitoring pool is updated, and the risk monitoring pool records the optical network components with the failure risk and the recording frequency of each optical network component; and based on the recording times of the to-be-predicted optical network component in the risk monitoring pool, determining whether the to-be-predicted optical network component is invalid in a future preset time period. According to the failure prediction method of the optical network component, an extremely low false alarm rate of the normal optical network component can be realized, and the implementation of actual operation and maintenance is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical communication technology, in particular to a failure prediction method and device of an optical network component. BACKGROUND

[0002] The existing optical module mainly uses a threshold rule-based method to predict the failure of a pump laser. The method compares the difference between the theoretical value of the optical power of the device and the actual optical power to realize module failure warning. The method has a very short prediction time and cannot meet the demand for advance operation and maintenance. Moreover, there is no effective intelligent detection method for predicting the failure of a pump laser.

[0003] In order to deal with the problem of service interruption caused by the failure of a pump laser, the internal operation and maintenance personnel of the current operator adopts a method of regular inspection. This method is not only time-consuming and labor-intensive, but also has a low detection rate when the failure of a pump laser is determined by manually checking the corresponding key indicators.

[0004] The manual inspection has a long interval, and the sudden change in indicators before the failure of a pump laser is very short. This will cause many pump lasers to fail within the inspection interval, and the cost-benefit ratio of manual inspection is low.

[0005] The operation and maintenance personnel need to make a replacement decision after prediction. This requires a very low false positive rate for normal optical modules. The current existing technology cannot meet this requirement. SUMMARY

[0006] Embodiments of the present application provide a failure prediction method and device of an optical network component, which can achieve a very low false positive rate for normal optical network components and ensure the implementability of actual operation and maintenance.

[0007] In a first aspect, the present application provides a failure prediction method of an optical network component, comprising: obtaining target working performance indicator data of a to-be-predicted optical network component; taking the target working performance indicator data as an input of a failure prediction model to output a failure prediction result; updating a risk monitoring pool based on the failure prediction result, the risk monitoring pool recording optical network components with a failure risk and a record number of each optical network component; and determining whether the to-be-predicted optical network component will fail within a future preset time period based on the record number of the to-be-predicted optical network component in the risk monitoring pool.

[0008] The failure prediction method of an optical network component provided by the present application first predicts the failure probability of the optical network component through a failure prediction model to realize long-term and stable failure prediction, and then further screens the optical network components with a failure risk through a risk monitoring pool. On the basis of improving the failure prediction accuracy of the optical network component, the false positive rate of the normal optical network component is reduced to meet the requirement of the implementability of actual operation and maintenance on the false positive rate.

[0009] In one possible implementation, one specific implementation of obtaining the working performance indicator data of the optical network component to be predicted is: receiving collected data, the collected data including a plurality of working performance indicator data collected from the optical network component to be predicted at a preset collection period; and pre-processing the collected data to obtain target working performance indicator data.

[0010] For example, the working performance indicator data of the optical network component to be predicted is monitored, collected and stored at a preset period, for example, at a three-minute collection period, at the optical network element side, and then the collected working performance indicator data is periodically sent to the network management side. The network management side receives the collected data, and then the network management side pre-processes the collected data to obtain target working performance indicator data.

[0011] Alternatively, the working performance indicator data of the optical network component to be predicted is monitored, collected and stored at a three-minute period at the optical network element side, and the network management side periodically collects the working performance indicator data of the network element for one hour at an hour interval, and then the network management side pre-processes the collected data to obtain target working performance indicator data.

[0012] In another possible implementation, one specific implementation of pre-processing the collected data to obtain target working performance indicator data is: determining an importance value of each working performance indicator data in the plurality of working performance indicator data, the importance value indicating an influence degree of each working performance indicator data on the failure prediction result; and selecting N working performance indicator data with the highest importance value as the target working performance indicator data.

[0013] For example, the collected working performance indicator data of the optical network component to be predicted is taken as an input of a decision tree or an extreme gradient boosting (XGBoost) model, an importance value of each working performance indicator data on the prediction result is output, and then the top N working performance indicator data with the highest importance value are selected as the target working performance indicator data.

[0014] The target working performance indicator data closer to the failure principle is selected and identified from a plurality of working performance indicator data of the laser, and the target working performance indicator data is used for failure prediction, thereby improving the accuracy of failure prediction.

[0015] In another possible implementation, the optical network component to be predicted is a laser in an optical module; the target working performance index data includes one or more of the following: pre-stage input optical power, pre-stage output optical power, post-stage input optical power, post-stage output optical power, input optical power of an external adjustable optical attenuator, laser drive current, laser back light current, laser refrigeration voltage, adjustable optical attenuator drive voltage, adjustable optical attenuator detection actual voltage, 5V mainboard main power supply voltage, 3.3V mainboard main power supply voltage, and mainboard temperature.

[0016] In another possible implementation, the target working performance index data is obtained by preprocessing the collected data, and the method further includes: performing first processing on the first target working performance index data to obtain a first processing result, the first target working performance index data at least including pre-stage input optical power, and the first processing including normalization processing; performing second processing on the second target working performance index data to obtain a second processing result, the second target working performance index data including pre-stage output optical power, post-stage input optical power, and post-stage output optical power, and the second processing including constraint processing based on expert experience and normalization processing.

[0017] By combining expert experience to perform special normalization processing on part of the working performance index data, the accuracy of the normalization processing is ensured, the inference prediction of the subsequent failure prediction model is facilitated, and the accuracy of the failure prediction is provided.

[0018] In another possible implementation, the failure prediction model at least includes a convolutional neural network (CNN) layer and a long short-term memory (LSTM) network layer. The CNN not only has good high-dimensional spatial feature extraction capability, but also can play a role in data compression and dimension reduction, while the LSTM has stronger capture capability for the relationship between time series data and features before and after the time series data. Through the processing of the CNN layer and the LSTM layer on the input data, the failure prediction accuracy of the failure prediction model is improved.

[0019] In another possible implementation, the failure prediction result includes a failure probability, and the failure probability indicates a probability that the optical network component to be predicted will fail in a future preset time period; based on the failure prediction result, one specific implementation of updating the risk monitoring pool is: when the failure probability is greater than a preset probability value, it is checked whether the optical network component to be predicted exists in the risk monitoring pool, if the optical network component to be predicted exists, a plus one operation is performed on the record number of the optical network component to be predicted, if the optical network component to be predicted does not exist, the optical network component to be predicted is added to the risk monitoring pool, and the record number of the optical network component to be predicted is set to one; when the failure probability is less than or equal to the preset probability value, it is checked whether the optical network component to be predicted exists in the risk monitoring pool, if the optical network component to be predicted exists, the optical network component to be predicted is removed from the risk monitoring pool.

[0020] In another possible implementation, the determining whether the to-be-predicted optical network component will fail in the future preset time period based on the number of records of the to-be-predicted optical network component in the risk monitoring pool includes: when the number of records of the to-be-predicted optical network component in the risk monitoring pool is greater than or equal to a preset threshold, it is determined that the optical network component will fail in the future preset time period; and when the number of records of the to-be-predicted optical network component in the risk monitoring pool is less than the preset threshold, it is determined that the to-be-predicted optical network component will not fail in the future preset time period.

[0021] By setting the risk monitoring pool, the failure risk of the optical network component is further determined, the failure prediction accuracy of the optical network component is improved, and the misjudgment rate of the normal optical network component is reduced. Compared with directly determining whether the optical network component fails according to the output result of the failure prediction model, the misjudgment rate of the normal optical network component is further reduced by adding the judgment mechanism of the risk monitoring pool, the misjudgment rate of the normal optical network component is extremely low, and the requirement of the actual operation and maintenance on the false alarm rate is met.

[0022] Optionally, the failure prediction result includes a mirror catastrophe failure prediction result and an intrinsic failure prediction result, so that the failure prediction result is more detailed and the failure prediction is more fine-grained.

[0023] In another possible implementation, the input data of the failure prediction model further includes expert experience data and long-term historical statistical data, wherein the expert experience data includes data processed based on expert experience on the target working performance indicator; and the long-term statistical data includes statistical feature data of the target working performance indicator data in a preset historical time period.

[0024] The input data includes more abundant information, and the prediction accuracy of the failure prediction model is improved.

[0025] In another possible implementation, the failure prediction method of the optical network component provided in the present application further includes: when it is determined that the to-be-predicted optical network component will fail in the future preset time period, the to-be-predicted optical network component is reported for alarm. This makes the operation and maintenance personnel know that the optical network component will fail before it actually fails, realizes proactive operation and maintenance, facilitates the equipment operation and maintenance personnel to quickly and accurately identify the potential failure risk, switches the link or replaces the optical network component in advance to ensure that the business is not affected, avoids major complaints, improves the operation and maintenance efficiency, and reduces the operation and maintenance cost.

[0026] In another possible implementation, the failure prediction method of the optical network component provided in the present application further includes displaying the content recorded by the risk monitoring pool, so as to facilitate the operation and maintenance personnel to focus on monitoring the optical network component recorded in the risk monitoring pool or the optical network component with a higher number of records in the risk monitoring pool.

[0027] In a second aspect, the present application also provides a failure prediction device of an optical network component, the device comprising at least an acquisition module, a model prediction module, an update module and a determination module, wherein the acquisition module is configured to acquire target working performance index data of a to-be-predicted optical network component; the model prediction module is configured to take the target working performance index data as an input of a failure prediction model, and output a failure prediction result; the update module is configured to update a risk monitoring pool based on the failure prediction result, the risk monitoring pool recording optical network components with a failure risk and a record number of each optical network component; and the determination module is configured to predict whether the to-be-predicted optical network component will fail in a future preset time period based on the record number of the to-be-predicted optical network component in the risk monitoring pool.

[0028] In another possible implementation, the acquisition module is specifically configured to receive collection data, the collection data comprising a plurality of working performance index data obtained by collecting the to-be-predicted optical network component at a preset collection period; and pre-process the collection data to obtain the target working performance index data.

[0029] In another possible implementation, one specific implementation of pre-processing the collection data to obtain the target working performance index data is as follows: determining an importance value of each item of working performance index data in the plurality of working performance index data, the importance value indicating an influence degree of each item of working performance index data on the failure prediction result; and taking N items of working performance index data with the highest importance value as the target working performance index data.

[0030] In another possible implementation, the to-be-predicted optical network component is a laser in an optical module; and the target working performance index data comprises one or more of a front-stage input optical power, a front-stage output optical power, a rear-stage input optical power, a rear-stage output optical power, an input optical power of an external adjustable optical attenuation, a laser driving current, a laser back light current, a laser refrigeration voltage, a driving voltage of the adjustable optical attenuation, an actual voltage detected by the adjustable optical attenuation, a voltage of a 5V mainboard main power supply, a voltage of a 3.3V mainboard main power supply and a mainboard temperature.

[0031] In another possible implementation, the pre-processing of the collection data to obtain the target working performance index data further comprises: performing first processing on first target working performance index data to obtain a first processing result, the first target working performance index data at least comprising the front-stage input optical power, the first processing comprising normalization processing; and performing second processing on second target working performance index data to obtain a second processing result, the second target working performance index data comprising the front-stage output optical power, the rear-stage input optical power and the rear-stage output optical power, the second processing comprising constraint processing based on expert experience and normalization processing.

[0032] In a possible implementation, the failure prediction result includes a failure probability, the failure probability indicating a probability that the to-be-predicted optical network component will fail in a preset time period in the future; and the updating module is specifically configured to: when the failure probability is greater than a preset probability value, checking whether the to-be-predicted optical network component exists in the risk monitoring pool, and if the to-be-predicted optical network component exists in the risk monitoring pool, performing an increment operation on the record number of the to-be-predicted optical network component, and if the to-be-predicted optical network component does not exist in the risk monitoring pool, adding the to-be-predicted optical network component to the risk monitoring pool and setting the record number of the to-be-predicted optical network component to 1; and when the failure probability is less than or equal to the preset probability value, checking whether the to-be-predicted optical network component exists in the risk monitoring pool, and if the to-be-predicted optical network component exists in the risk monitoring pool, removing the to-be-predicted optical network component from the risk monitoring pool.

[0033] In another possible implementation, the determining module is specifically configured to: when the record number of the to-be-predicted optical network component in the risk monitoring pool is greater than or equal to a preset threshold, determining that the optical network component will fail in a preset time period in the future; and when the record number of the to-be-predicted optical network component in the risk monitoring pool is less than the preset threshold, determining that the to-be-predicted optical network component will not fail in the preset time period in the future.

[0034] In another possible implementation, the failure prediction result includes a mirror catastrophe failure prediction result and an intrinsic failure prediction result, so that the failure prediction result is more refined and the failure prediction is more fine-grained.

[0035] In another possible implementation, the input data of the failure prediction model further includes expert experience data and long-term historical statistical data, the expert experience data includes data processed based on expert experience on the target work performance indicator, and the long-term statistical data includes statistical feature data of the target work performance indicator data in a preset historical time period.

[0036] In another possible implementation, the failure prediction apparatus for the optical network component provided in the present application further includes a reporting module, the reporting module being configured to report the to-be-predicted optical network component for alarm when it is determined that the to-be-predicted optical network component will fail in a preset time period in the future.

[0037] In another possible implementation, the failure prediction apparatus for the optical network component provided in the present application further includes a display module, the display module being configured to display the content recorded by the risk monitoring pool.

[0038] In a third aspect, an embodiment of the present application provides a network management device, including a memory and a processor, and the memory stores instructions, when the instructions are executed by the processor, the method in the first aspect is implemented.

[0039] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program, when the computer program is executed by a processor, the method in the first aspect is implemented.

[0040] In a fifth aspect, an embodiment of the present application further provides a computer program or a computer program product, which comprises instructions, when the instructions are executed, causing a computer to execute the method in the first aspect.

[0041] In a sixth aspect, an embodiment of the present application further provides a chip, comprising at least one processor and a communication interface, and the processor is configured to execute the method in the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 An implementation architecture schematic diagram of the failure prediction method of the optical network component provided by the embodiment of the present application is shown;

[0043] Figure 2 A flow schematic diagram of the failure prediction method of the optical network component provided by the embodiment of the present application is shown;

[0044] Figure 3 A reasoning prediction process schematic diagram of the failure prediction model in the embodiment of the present application is shown;

[0045] Figure 4 Three main stages of the training of the failure prediction model are shown;

[0046] Figure 5 A schematic diagram for determining the importance value of the target working performance index is shown;

[0047] Figure 6 A specific implementation flow schematic diagram of the failure prediction method of the optical network component provided by the embodiment of the present application is shown;

[0048] Figure 7 A structure schematic diagram of the failure prediction device of the optical network component provided by the embodiment of the present application is shown;

[0049] Figure 8 A structure schematic diagram of the network management device provided by the embodiment of the present application is shown. DETAILED DESCRIPTION

[0050] The term “and / or” mentioned in the present document is a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that there are three cases of A alone, A and B together, and B alone. The symbol “ / ” in the present document represents an or relationship of the associated objects, for example, A / B represents A or B.

[0051] The terms “first” and “second” and the like in the description and claims of the present document are used to distinguish different objects, and are not used to describe a specific order of the objects. For example, the first processing and the second processing are used to distinguish different processing, and are not used to describe a specific order of the processing.

[0052] In the embodiments of the present application, the word "exemplary" or "for example" is used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the present application is not necessarily to be construed as preferred or advantageous over other embodiments or designs. In fact, a variety of implementations of the present application can be made depending on the specific needs and requirements of the users. The goals of the embodiments of the present application can be achieved by a variety of means, one of which is described herein.

[0053] In the description of the embodiments of the present application, the meaning of "plurality" is two or more, for example, a plurality of processing units means two or more processing units, and the like; a plurality of elements means two or more elements, and the like, unless otherwise specified.

[0054] The failure prediction method and device of the optical network component provided by the embodiments of the present application can be applied to predict the failure of the optical network component. The optical network component can be any optical network component in an optical network communication system, for example, the optical network component includes but is not limited to a laser, a fiber link, a digital signal processor, a driver, an optical detector, a temperature control module, a transimpedance amplifier, an optical interface, an electrical interface, or a fiber amplifier.

[0055] Optionally, the failure prediction method and device of the optical network component provided by the embodiments of the present application can be applied to predict the failure of the pump laser (which can be referred to as a laser) in the optical module. On the premise of ensuring the false positive rate of the normal laser, the laser that is about to fail can be identified and reported in advance, so as to avoid the impact or even interruption of the service.

[0056] The failure prediction method and device of the optical network component provided by the embodiments of the present application are described below by taking the laser as an example.

[0057] With the wide application of the optical transmission network (OTN) network, the demand for hardware devices such as optical modules also increases exponentially. The pump laser is an important device of the optical module. The failure of the pump laser will directly lead to the failure of the optical module, and then will affect the service. The related operation and maintenance personnel need to perform regular inspection, which is not only time-consuming and laborious, but also has a very low detection rate by manual inspection, and has a high failure risk.

[0058] The related art mainly uses a threshold rule-based method for the technical solution of predicting the failure of a pump laser. First, the laser has a preset theoretical normal working value after leaving the factory. Second, for a large amount of historical failure module data, the light power and other parameter data near the failure can be deeply analyzed and statistically analyzed to obtain a failure threshold range. Finally, the threshold range is combined to determine the final threshold for determining the failure of the laser. After the failure threshold is determined, in daily work, the related working parameters of the laser are periodically collected, and then the related working parameters are compared with the failure threshold to determine whether the laser exceeds the threshold. If the laser exceeds the threshold, it is marked as having a failure risk and an alarm is given.

[0059] However, the scheme has a very short prediction time, which cannot meet the needs of advance maintenance. In addition, the false positive rate of the scheme for normal optical modules is high, and maintenance prediction is required for replacement decisions, which has a very high requirement for the false positive rate of normal optical modules. The scheme cannot meet the requirements of normal maintenance for the false positive rate of normal optical modules.

[0060] To this end, the embodiments of the present application provide a failure prediction method and device for an optical network component. A deep learning failure prediction model based on historical feature data (such as light power, current, temperature, etc.) before the failure of a pump laser and combined with expert knowledge is established to achieve a more accurate failure prediction of the pump laser. In order to effectively reduce the false positive rate of normal lasers, a risk monitoring pool is additionally set up after the failure prediction model is predicted to further identify and confirm the lasers with a failure risk, thereby achieving a very low false positive rate of normal lasers, further improving the accuracy of laser failure prediction, facilitating equipment maintenance personnel to quickly and accurately identify potential failure risks, switching links or replacing optical modules in advance to ensure that the business is not affected, avoiding major complaints, improving maintenance efficiency, and reducing maintenance costs.

[0061] Figure 1 An implementation architecture diagram of the failure prediction method for the optical network component provided by the embodiments of the present application is shown. As shown in FIG. 1, the implementation architecture diagram of the failure prediction method for the optical network component provided by the embodiments of the present application includes a data collection module 101, a deep learning model 102, a risk monitoring pool 103, and a risk alarm module 104. Figure 1As shown, the optical network element is provided with a collection module for collecting key performance indicator (KPI) data of the laser inside the optical module related to failure prediction. The calculation module in the network management system applies the failure prediction method of the optical network component provided in the embodiments of the present application to regularly obtain the working performance indicator data collected by the collection module, process the working performance indicator data, use the working performance indicator data to predict failure, for example, input the working performance indicator data into a failure prediction model to obtain a failure prediction result, and then use the failure prediction result to update the risk monitoring pool. By focusing on monitoring the laser entering the risk monitoring pool, the laser meeting the reporting conditions (such as the laser being added to the risk monitoring pool for n consecutive times) is finally reported in the form of a failure alarm, and the northbound interface reports to the upper layer operations support system (OSS). After reporting the failure alarm, the relevant operation and maintenance personnel can take effective operation and maintenance management measures (such as replacing the optical module) in a timely manner for the laser with failure risk to avoid affecting or even interrupting the service, thereby ensuring the service operation reliability of the optical network system.

[0062] Figure 2 A flowchart of a failure prediction method of an optical network component provided in the embodiments of the present application is shown. The failure prediction device of the optical network component applying the method can be integrated on the network management system shown to be implemented. The network management system can be deployed on a physical server, for example, the network management system is deployed on a server or a server cluster composed of multiple servers; the network management system can also be deployed in the cloud, that is, the network management service is provided in the form of cloud service, and the network management service includes providing a failure prediction service. As shown, Figure 1 Figure 2 As shown, the failure prediction method of the optical network component provided in the present application at least includes steps S201 to S204.

[0063] In step S201, target working performance indicator data of a laser to be predicted is obtained.

[0064] ​In the embodiments of the present application, the collector is arranged at the optical network element side to collect the working performance index data of the laser in the optical module. For example, the collector collects the working performance index data of the laser according to a preset collection period, such as collecting the working performance index data of the laser once every 3 minutes. Then the collector regularly sends the collected working performance index data to the network management system (i.e. the failure prediction device in the network management system), and the network management system receives the collection data sent by the collector to obtain the working performance index data of the laser. For example, the collector sends the collected data to the network management system every hour, and the network management system receives the collection data to obtain the working performance index data of the laser, so that the optical network component failure prediction device uses the working performance index data to perform failure prediction, that is, the optical network component failure prediction device performs failure prediction on the laser once every hour to further determine whether the optical module will fail. Of course, in some other examples, the network management system can also actively acquire the collection data collected by the collector regularly, such as sending a collection data acquisition request to the collector every hour, and the collector sends the collected collection data to the network management system in response to the request. The network management system receives the collection data sent by the collector to obtain the working performance index data of the laser.

[0065] It can be understood that the collection period of the collector and the sending period of the collector to the network management system for sending the collection data can be set according to actual needs, and the above is only an example and does not constitute a limitation on the embodiments of the present application.

[0066] Optionally, the collector can be integrated in the optical module to collect the working performance index data of the laser. Alternatively, the collector can also be separately arranged and connected with the laser in the optical module to collect the working performance index data of the laser.

[0067] The working performance index data collected by the collector can be all working performance index data reflecting the working state of the laser. After the collector sends the collected working performance index data to the network management system, the network management system receives the working performance index data and then selects target working performance index data required for failure prediction from the working performance index data. The target working performance index data is working performance index data more relevant to failure prediction, which can also be referred to as working performance index data with a higher importance value for the failure prediction result. For example, the target working performance index data includes one or more of the front-stage input optical power, the front-stage output optical power, the rear-stage input optical power, the rear-stage output optical power, the input optical power of the external adjustable optical attenuation, the laser drive current, the laser back light current, the laser refrigeration voltage, the drive voltage of the adjustable optical attenuation, the actual voltage detected by the adjustable optical attenuation, the voltage of the 5V main board main power supply, the voltage of the 3.3V main board main power supply, and the main board temperature.

[0068] An example way of determining the importance value of the working performance index data of the laser can be: inputting each working performance index data of the laser into a decision tree model or an XGBoost model to obtain the importance value of each working performance index data of the laser, then sorting the importance values from large to small, and taking the top N working performance index data as the target working performance index data.

[0069] By screening, the target working performance index data closer to the failure principle is identified from the collected multiple working performance index data, and the target working performance index data is used for failure prediction, thereby improving the accuracy of failure prediction.

[0070] In another example, the collector can also directly collect the target working performance index data of the laser, and then send the collected target working performance index data to the network management system. The network management system receives the target working performance index data to obtain the target working performance index data of the laser.

[0071] In step S202, the target working performance index data is taken as the input of the failure prediction model, and the failure prediction result is output.

[0072] In order to facilitate the inference prediction of the failure prediction model, the target work performance index data is preprocessed before being input into the failure prediction model. Optionally, the preprocessing includes processing of missing values and abnormal values; removing data with non-continuous data sampling time from the collected target work performance index data of the laser; uniformly normalizing part of the work performance index data according to expert experience, and performing special normalization processing on part of the work performance index data in combination with expert experience, and the like. For example, according to expert experience, it is judged that the actual values of the front-stage output optical power, the rear-stage input optical power and the rear-stage output optical power in the target work performance index data of the laser may exceed the range of the theoretical maximum value and the minimum value, and need to be constrained, otherwise it will affect the result of the normalization processing, and further affect the inference prediction of the subsequent failure prediction model. For example, the theoretical value range (which can also be referred to as the constraint value range) of the front-stage output optical power is 1100 to 2100, with a unit of 0.01 dBm; when the collected value of the front-stage output optical power exceeds the value range, it is constrained, for example, the collected value of the front-stage output optical power is 2150, and after constraint, the value of the front-stage output optical power is 2100, and the collected value of the front-stage output optical power is 910, and after constraint, the value of the front-stage output optical power is 1100. Similarly, the rear-stage input optical power and the rear-stage output optical power are constrained and processed, and optionally, the constraint value range of the rear-stage input optical power is -2300 to 1500, with a unit of 0.01 dBm, and the constraint value range of the rear-stage output optical power is -90 to 2300, with a unit of 0.01 dBm. After the front-stage output optical power, the rear-stage input optical power and the rear-stage output optical power are constrained and processed, they are uniformly normalized together with other collected target work performance index data.

[0073] Since different work performance index data have different dimensions and dimension units, such a situation will affect the result of data analysis, that is, the inference prediction of the failure prediction model. After the normalization processing of different work performance index data, the different work performance index data to be processed are limited within a certain range (such as [0, 1] or [-1, 1]), so as to eliminate the adverse effects caused by singular sample data, and facilitate the inference prediction of the failure prediction model.

[0074] The specific way of the normalization processing operation is not limited in the embodiments of the present application, for example, the normalization operation can be min-max normalization, z-score standardization, log function normalization and arctangent function normalization, and the like, and a suitable normalization processing operation method can be selected according to actual needs.

[0075] Then, the preprocessed target working performance index data is input into the failure prediction model to perform failure prediction, and a failure prediction result of the laser is obtained. The failure prediction result includes a failure probability, which indicates a probability that the laser will fail in a future preset time period (e.g., one week in the future). By comparing the failure probability with a threshold value γ, it can be preliminarily determined whether the laser will fail in the future preset time period.

[0076] In one example, to further increase the accuracy of the inference prediction of the failure prediction model, the input data of the failure prediction model further includes expert experience data and long-term historical statistical data, where the expert experience data includes preprocessed data of the target working performance index based on expert experience; and the long-term statistical data includes statistical feature data of the target working performance index data in a preset historical time period. For example, specific working index data in the target working performance index data is processed based on expert experience, so that the failure characteristics of the laser are more reflected, such as differential mean processing of the front-stage input optical power, variance processing of the laser drive current, etc. The long-term statistical data is statistical feature data of the working performance index data of the laser in a longer historical period, such as the working performance index data of the laser in the past month, and then the specific working performance index data is processed, including mean, variance, first-order difference, second-order difference, etc. statistical feature processing, to obtain long-term historical statistical data of the target working performance index data.

[0077] After the expert experience data and the long-term historical statistical data of the laser are processed, they are spliced with the preprocessed target input working performance index data, and the spliced data is input into the failure prediction model to obtain a failure prediction result.

[0078] It should be noted that the preprocessed target working performance index data, the expert experience data, and the long-term historical statistical data are all feature vectors after processing, and the same working vector index is spliced with the preprocessed target working performance index data, the expert experience data, and the long-term historical statistical data. For example, the target working performance index data includes 15 working performance index data of the laser, such as front-stage input optical power, front-stage output optical power, rear-stage input optical power, rear-stage output optical power, input optical power of an external adjustable optical attenuator, laser drive current, laser back light current, laser refrigeration voltage, adjustable optical attenuator drive voltage, adjustable optical attenuator detection actual voltage, 5V mainboard main power voltage, 3.3V mainboard main power voltage, and mainboard temperature. The same working performance index is spliced with the preprocessed target working performance index data, the expert experience data, and the long-term historical statistical data, and after splicing processing, it is still a feature vector corresponding to 15 working performance indexes.

[0079] Figure 3is a schematic diagram of a reasoning prediction process of a failure prediction model in an embodiment of the present application. As shown in Figure 3 The data obtained by splicing the preprocessed target work performance index data, expert experience data, and long-term historical statistical data is input into the failure prediction model for reasoning prediction, and a probability distribution is output, including the probability of normal laser, the probability of intrinsic failure, and the probability of CMD failure. Comparison with the threshold value γ can preliminarily determine whether the laser will have intrinsic failure or CMD failure in the future preset time period, and realize more fine-grained failure prediction.

[0080] It is easy to understand that intrinsic failure is also called lattice failure. The lattice is a multilayer semiconductor material structure grown layer by layer. Intrinsic defects such as uneven growth thickness may occur in the growth process due to differences in growth process or material itself. CMD failure is caused by the failure of the desiccant in the new air compressor (CDA) equipment of the manufacturer, which leads to air humidity, pipe corrosion, and the presence of small copper and nickel components in the compressed air, which may contaminate the pump chip light emitting surface after pipeline transmission.

[0081] Referring back to Figure 3 The failure prediction model is a CNN-LSTM fusion model, including a CNN layer, an LSTM layer, and a full connection layer. The CNN has good high-dimensional space feature extraction capability, and can also play a role in data compression and dimension reduction. The LSTM has stronger capture capability for the relationship between time series data before and after time and between features. Through the processing of the input data by the CNN layer and the LSTM layer, the failure prediction accuracy of the failure prediction model is improved. For example, the input window length of the failure prediction model is one week, and the output window length is also one week, that is, the work performance index data of the past one week is used to predict whether the laser in the optical module will fail in the future one week, and whether the optical module will fail in the future one week is predicted.

[0082] Of course, the structure of the above failure prediction model is only an example that can be realized, and does not constitute a limitation on the implementation of the present application. The failure prediction model can also be other neural network structures, such as a failure prediction model including a CNN layer and a full connection layer, or a failure prediction model including an LSTM layer and a full connection layer, etc.

[0083] The training implementation scheme of the failure prediction model is introduced below.

[0084] The training of the failure prediction model adopts an offline training scheme, Figure 4 The three main stages of training of the failure prediction model are shown in FIG. 4, including Figure 4As shown, the training of the failure prediction model mainly includes three stages: a data cleaning and preprocessing stage, a feature understanding and feature engineering stage, and an AI model training stage. Among them, the data cleaning and preprocessing stage is used to clean and preprocess the collected working performance indicator data of the laser, which is more suitable for training of the AI model (i.e. the failure prediction model). The feature understanding and feature engineering stage is used to select the working performance indicator data more related to failure prediction from the collected working performance indicator data of the laser as the features of the laser, so that the prediction of the trained failure prediction model is more accurate. The AI model training stage is used to train the AI model by using the data processed in the data cleaning and preprocessing stage and the feature understanding and feature engineering stage as training data. The trained AI model is deployed in the network management system as a failure prediction model to predict the failure of the laser.

[0085] For example, an implementation of the data cleaning and preprocessing stage can be: processing the missing values and abnormal values in the collected working performance indicator data of the laser; removing the data with non-continuous data sampling time in the collected working performance indicator data of the laser; uniformly normalizing part of the working performance indicator data according to expert experience, and specially normalizing part of the working performance indicator data in combination with expert experience; the expert labels the failure time of the failed optical module.

[0086] The feature understanding and feature engineering stage is mainly to find out the target working performance indicator data most related to failure prediction, for example, as shown in the following table: Figure 5 As shown, the original collected working performance indicator data reflecting all items of the laser are taken as the input of the decision tree (or XGBoost) model, to obtain the importance value of each working performance indicator data as a feature for prediction, and then perform feature screening to select the top N features as the input of the prediction model.

[0087] After the operation of the above steps, the top N target working performance indicator data are obtained, as described above. Then, the target working performance indicator data of the laser are processed to obtain expert experience data and long-term historical statistical data of the laser, and then the preprocessed target working performance indicator data, expert experience data and long-term historical statistical data are spliced to obtain training samples. In combination with the failure time labeling of the failed optical module by the expert, the label of the training sample is obtained.

[0088] In the AI model training stage, the training sample is taken as an input of the to-be-trained failure prediction model, and an output failure prediction result is obtained. A loss value is obtained by using the failure prediction result and the label, and the weight parameters of the to-be-trained failure prediction model are adjusted to minimize the loss value. After multiple rounds of training iterations, when a training condition is met, the trained failure prediction model is obtained. The training condition can be that the loss function converges, or that a preset training round is reached.

[0089] Finally, the trained failure prediction model is deployed in the network management system as a failure prediction model for predicting the failure of the laser.

[0090] In step S203, the risk monitoring pool is updated based on the failure prediction result.

[0091] In the embodiment of the present application, the failure probability value obtained by the failure prediction model is compared with the set threshold value γ. If the probability value is less than or equal to the threshold value γ, the risk monitoring pool is further checked. If the laser is already in the risk monitoring pool, it is removed. If the probability score is greater than the threshold value γ, the risk monitoring pool is also further checked. If the laser is not in the risk monitoring pool, it is added to the risk monitoring pool, and the parameter tag is set to 1, which means that the laser has exceeded the threshold value γ for a continuous number of times. If the laser is already in the risk monitoring pool, the operation of tag = tag + 1 is performed. That is, the risk monitoring pool records the laser (which can be represented by the optical module id) that is preliminarily predicted to have a failure risk by the failure prediction model and the number of times the laser is continuously added to the risk monitoring pool.

[0092] The risk monitoring pool further judges the failure prediction result output by the failure prediction model, greatly reduces the misjudgment rate of misjudging a normal laser as a failure laser, and meets the requirement of actual operation and implementation on the misjudgment rate.

[0093] In step S204, whether the to-be-predicted laser will fail in a future preset time period is determined based on the recorded number of times of the to-be-predicted laser in the risk monitoring pool.

[0094] If the number of times recorded in the risk monitoring pool of the laser is greater than or equal to a preset threshold value, it is determined that the laser will fail in the future preset time period. If the number of times recorded in the risk monitoring pool of the laser is less than the preset threshold value, it is determined that the laser component will not fail in the future preset time period.

[0095] For example, if the laser to be predicted is preliminarily predicted to fail in the next week by the failure prediction model, it is found that the laser is recorded in the risk monitoring pool, and the record number of the laser is 4. The record number of the laser is incremented by 1, and the record number of the laser becomes 5. The failure judgment failure threshold of the laser is compared with the failure threshold 5, and the result is equal to the failure threshold 5. Therefore, it is determined that the laser will fail in the next week.

[0096] By setting the risk monitoring pool, the failure risk of the laser is further determined, the failure prediction accuracy of the laser is improved, and the misjudgment rate of the normal laser is reduced. Compared with directly determining whether the laser fails according to the output result of the failure prediction model, the misjudgment rate of the normal laser is further reduced by adding the judgment mechanism of the risk monitoring pool, the misjudgment rate of the normal laser is extremely low, and the requirement of actual operation and maintenance on the misjudgment rate is met.

[0097] In some other examples, the failure prediction method of the optical network component provided by the embodiments of the present application further includes, when it is determined that the laser to be predicted will fail in a future preset time period, the laser to be predicted is reported for alarm. The operator is informed that the laser will fail before the laser fails, active operation is realized, the potential failure risk is quickly and accurately identified by the equipment operator, the link is switched or the optical module in which the laser is located is replaced in advance to ensure that the business is not affected, major complaints are avoided, the operation and maintenance efficiency is improved, and the operation and maintenance cost is reduced.

[0098] In another possible implementation, the failure prediction method of the optical network component provided by the present application further includes displaying the content recorded in the risk monitoring pool, so as to facilitate the operator to focus on monitoring the optical module in which the laser recorded in the risk monitoring pool is located, or focus on monitoring the optical module in which the laser with a higher record number in the risk monitoring pool is located.

[0099] For example, at a certain time, the operator needs to patrol the optical network system optical module. At this time, the operator can log in to the interactive interface of the network management system, click the risk monitoring pool control, and display the content recorded in the risk monitoring pool to the operator in response to the operation. The operator focuses on monitoring the optical module with potential failure risk recorded in the risk monitoring pool during the patrol, so as to realize focused monitoring, reduce the workload of the operator, and more easily identify the optical module with failure risk.

[0100] Figure 6 A specific implementation flowchart of the failure prediction method of the optical network component provided by the embodiments of the present application is shown. As shown in FIG. 4, the failure prediction method of the optical network component provided by the embodiments of the present application includes the following steps. Figure 6As shown, in the normal operation process of the optical network system, first, the optical module itself will monitor and collect and store the internal laser related working performance index data with a three-minute cycle, and the network management system will collect the working performance index data of the network element for one hour every hour. After collection, the failure prediction model is used for prediction, that is, the laser in each optical module is detected once an hour. The failure probability score obtained by the failure prediction model is compared with the set threshold value γ (for example, γ can be 0.8). If the probability score is less than or equal to the threshold value γ, the risk monitoring pool is further viewed. If the optical module is in the risk monitoring pool, it is removed. If the probability score is greater than the threshold value γ, the risk monitoring pool is further viewed. If the optical module is not in the risk monitoring pool, it is added to the risk monitoring pool, and the parameter tag is set to 1, which means that the number of times that the optical module continuously exceeds the threshold value γ. If the optical module is already in the risk monitoring pool, the operation of tag = tag + 1 is performed, and then compared with the set threshold value M (for example, M can be 5, and the value of M can be set by the operation and maintenance personnel according to actual needs). When the number of times (that is, tag) that the predicted probability score value of the optical module continuously exceeds the threshold value γ is greater than or equal to the threshold value M, the optical module information is reported in the form of a failure hidden danger alarm. When the operation and maintenance personnel see the alarm report, they will perform operations such as on-site replacement to avoid problems such as business interruption caused by optical module laser failure. The failure prediction method of the optical network component provided in the embodiment of the present application can achieve a very low false positive rate of normal lasers, ensuring the implementability of actual operation and maintenance. At the same time, in the scheme provided in the embodiment of the present application, there is a failure risk hidden pool, which can be monitored, and the optical module most likely to fail can be reported in the form of a failure hidden danger alarm through the network management side through the judgment mechanism, realizing proactive operation and maintenance.

[0101] The failure prediction method of the optical network component provided in the embodiment of the present application can be particularly applied to accurately predict the failure of the pump laser in the optical module, such as intrinsic failure and CMD failure, to achieve a very low false positive rate of normal optical modules and ensure the implementability of actual operation and maintenance. The failure prediction method of the optical network component provided in the embodiment of the present application can also be applied to accurately predict the failure of the pump laser in the optical module, such as intrinsic failure and CMD failure, to achieve a very low false positive rate of normal optical modules and ensure the implementability of actual operation and maintenance.

[0102] Based on the same concept as the foregoing embodiment of the failure prediction method of the optical network component, the embodiment of the present application also provides a failure prediction device 700 of an optical network component. The failure prediction device 700 of the optical network component can be integrated in a network management system to achieve accurate failure prediction of the optical network component and a very low false positive rate of normal optical network components, ensuring the implementability of actual operation and maintenance. The failure prediction device 700 of the optical network component includes a device for Figures 2-6The units or modules of each step in the failure prediction method for optical network components shown.

[0103] Figure 7 This is a schematic diagram of the structure of a failure prediction device for an optical network component provided in an embodiment of this application. Figure 7 As shown, the optical network component failure prediction device 700 includes at least an acquisition module 701, a model prediction module 702, an update module 703, and a determination module 704. The acquisition module 701 acquires target performance index data of the optical network component to be predicted; the model prediction module 702 uses the target performance index data as input to a failure prediction model and outputs a failure prediction result; the update module 703 updates a risk monitoring pool based on the failure prediction result, and the risk monitoring pool records optical network components with failure risk and the number of times each optical network component is recorded; the determination module 704 predicts whether the optical network component to be predicted will fail within a preset time period in the future based on the number of records of the optical network component to be predicted in the risk monitoring pool.

[0104] In another possible implementation, the acquisition module 701 is specifically used to receive the collected data, which includes multiple performance index data of the optical network component to be predicted, collected at a preset acquisition period; and to preprocess the collected data to obtain the target performance index data.

[0105] In another possible implementation, a specific way to preprocess the collected data to obtain the target performance index data is as follows: determine the importance value of each performance index data among multiple performance index data, the importance value indicating the degree of influence of each performance index data on the failure prediction result; and take the N performance index data with the highest importance value as the target performance index data.

[0106] In another possible implementation, the optical network component to be predicted is a laser in an optical module; the target operating performance data includes one or more of the following: laser pre-stage input optical power, pre-stage output optical power, post-stage input optical power, post-stage output optical power, input optical power of external adjustable optical decay, laser drive current, laser backlight current, laser cooling voltage, adjustable optical decay drive voltage, actual voltage of adjustable optical decay detection, 5V mainboard power supply voltage, 3.3V mainboard power supply voltage, and mainboard temperature.

[0107] In another possible implementation, the pre-processing of the collected data to obtain the target working performance indicator data further includes: performing first processing on the first target working performance indicator data to obtain a first processing result, the first target working performance indicator data at least including the front-stage input optical power, and the first processing including normalization processing; and performing second processing on the second target working performance indicator data to obtain a second processing result, the second target working performance indicator data including the front-stage output optical power, the rear-stage input optical power, and the rear-stage output optical power, and the second processing including constraint processing based on expert experience and normalization processing. The failure prediction model at least includes a CNN layer and an LSTM layer.

[0108] In one possible implementation, the failure prediction result includes a failure probability, and the failure probability indicates a probability that the to-be-predicted optical network component will fail in a preset time period in the future; and the updating module 703 is specifically configured to: when the failure probability is greater than a preset probability value, check whether the to-be-predicted optical network component exists in the risk monitoring pool, if the to-be-predicted optical network component exists in the risk monitoring pool, perform an increment operation on the record number of the to-be-predicted optical network component, if the to-be-predicted optical network component does not exist in the risk monitoring pool, add the to-be-predicted optical network component to the risk monitoring pool, and set the record number of the to-be-predicted optical network component to 1; and when the failure probability is less than or equal to the preset probability value, check whether the to-be-predicted optical network component exists in the risk monitoring pool, if the to-be-predicted optical network component exists in the risk monitoring pool, remove the to-be-predicted optical network component from the risk monitoring pool.

[0109] In another possible implementation, the determining module 704 is specifically configured to: when the record number of the to-be-predicted optical network component in the risk monitoring pool is greater than or equal to a preset threshold, determine that the optical network component will fail in a preset time period in the future; and when the record number of the to-be-predicted optical network component in the risk monitoring pool is less than the preset threshold, determine that the to-be-predicted optical network component will not fail in the preset time period in the future.

[0110] In another possible implementation, the failure prediction result includes a mirror catastrophe failure prediction result and an intrinsic failure prediction result, so that the failure prediction result is more detailed and the failure prediction is more fine-grained.

[0111] In another possible implementation, the input data of the failure prediction model further includes expert experience data and long-term historical statistical data, wherein the expert experience data includes data processed based on expert experience on the target working performance indicator; and the long-term statistical data includes statistical feature data of the target working performance indicator data in a preset historical time period.

[0112] In another possible implementation, the failure prediction apparatus 700 for the optical network component provided in the present application further includes a reporting module 705, which is configured to: when it is determined that the to-be-predicted optical network component will fail in a preset time period in the future, report the to-be-predicted optical network component for alarm.

[0113] In another possible implementation, the failure prediction device 700 for optical network components provided in this application further includes a display module 706, which is used to display the contents recorded by the risk monitoring pool.

[0114] The failure prediction device 700 for optical network components provided according to the embodiments of this application can correspond to the execution of the methods described in the embodiments of this application, and the above and other operations and / or functions of each module in the failure prediction device 700 for optical network components are respectively for implementing Figures 2-6 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0115] This application embodiment also provides a network management device, including at least one processor, a memory, and a communication interface, wherein the processor is used to execute... Figures 2-6 The method described.

[0116] Figure 8 This is a schematic diagram of the structure of a network management device provided in an embodiment of this application.

[0117] like Figure 8 As shown, the network management device 800 includes at least one processor 801, a memory 802, and a communication interface 803. The processor 801, memory 802, and communication interface 803 are communicatively connected, which can be achieved via a wired (e.g., bus) or wireless connection. The communication interface 803 is used to send and / or receive data from other devices, including target performance indicators of optical network components. The memory 802 stores computer instructions, which the processor 801 executes to perform the methods described in the preceding embodiments, thereby achieving accurate failure prediction of optical network components and an extremely low false alarm rate for normal optical network components, ensuring the feasibility of actual operation and maintenance.

[0118] It should be understood that, in the embodiments of this application, the processor 801 may be a central processing unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0119] The memory 802 can include read-only memory and random access memory, and provide instructions and data to the processor 801. The memory 802 can also include non-volatile random access memory.

[0120] The memory 802 can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or flash memory. The volatile memory can be random access memory (RAM) used as an external cache. By way of example, but not by way of limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct memory bus random access memory (DRAM) (DRAM).

[0121] It should be understood that the network management device 800 according to the embodiments of the present application can perform the method shown in the embodiments of the present application, and the detailed description of the method implemented is described above. For the sake of brevity, it will not be repeated here. Figures 2-6 The method shown in the embodiments of the present application can be performed, and the detailed description of the method implemented is described above. For the sake of brevity, it will not be repeated here.

[0122] The embodiments of the present application provide a computer readable storage medium, which stores computer programs, when the computer instructions are executed by a processor, the above-mentioned method is implemented.

[0123] The embodiments of the present application provide a chip, which includes at least one processor and an interface, the at least one processor determines program instructions or data through the interface; the at least one processor is used to execute the program instructions to implement the above-mentioned method.

[0124] Embodiments of the present application provide a computer program or computer program product comprising instructions which, when executed, cause a computer to perform the above-mentioned method.

[0125] Those skilled in the art should further understand that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in the above description in general terms. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0126] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be implemented in hardware, software executed by a processor, or a combination of both. The software modules can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art.

[0127] The above detailed description of the specific implementation is further detailed for the purpose of the present application, technical solutions and beneficial effects. It should be understood that the above description is only a specific implementation of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting the failure of an optical network component, characterized in that, include: Obtain target performance data for the optical network components to be predicted; The target performance index data is used as input to the failure prediction model, and the failure prediction result is output. Based on the failure prediction results, the risk monitoring pool is updated. The risk monitoring pool records optical network components with failure risks and the number of times each optical network component is recorded. Based on the number of records of the optical network component to be predicted in the risk monitoring pool, it is determined whether the optical network component to be predicted will fail within a preset time period in the future.

2. The method according to claim 1, characterized in that, The acquisition of the performance index data of the optical network component to be predicted includes: Receive collected data, which includes multiple performance indicators obtained by collecting data from the optical network component to be predicted at a preset collection period; The collected data is preprocessed to obtain the target performance index data.

3. The method according to claim 2, characterized in that, The preprocessing of the collected data to obtain the target performance index data includes: Determine the importance value of each of the multiple performance index data, whereby the importance value indicates the degree of influence of each performance index data on the failure prediction result; The N performance indicators with the highest importance values ​​are used as the target performance indicators.

4. The method according to any one of claims 1-3, characterized in that, The optical network component to be predicted is a laser in an optical module; The target performance indicators include one or more of the following: the laser's pre-stage input optical power, pre-stage output optical power, post-stage input optical power, post-stage output optical power, external adjustable optical decay input optical power, laser drive current, laser backlight current, laser cooling voltage, adjustable optical decay drive voltage, actual voltage of adjustable optical decay detection, 5V mainboard power supply voltage, 3.3V mainboard power supply voltage, and mainboard temperature.

5. The method according to claim 4, characterized in that, The step of preprocessing the collected data to obtain the target performance index data further includes: The first target performance index data is subjected to a first processing to obtain a first processing result. The first target performance index data includes at least the front-end input optical power. The first processing includes normalization processing. The second target performance index data is subjected to a second processing to obtain a second processing result. The second target performance index data includes the output optical power of the pre-stage, the input optical power of the post-stage, and the output optical power of the post-stage. The second processing includes constraint processing based on expert experience and normalization processing.

6. The method according to any one of claims 1-5, characterized in that, The failure prediction model includes at least a convolutional neural network layer and a long short memory recursive network layer.

7. The method according to any one of claims 1-7, characterized in that, The failure prediction result includes a failure probability, which indicates the probability that the optical network component to be predicted will fail within a preset time period in the future. The step of updating the risk monitoring pool based on the failure prediction results includes: When the failure probability is greater than the preset probability value, check whether the optical network component to be predicted exists in the risk monitoring pool. If it exists, increment the number of records for the optical network component to be predicted by 1. If it does not exist, add the optical network component to be predicted to the risk monitoring pool and set the number of records for the optical network component to be predicted to 1. When the failure probability is less than or equal to the preset probability value, check whether the optical network component to be predicted exists in the risk monitoring pool. If it exists, remove the optical network component to be predicted from the risk monitoring pool.

8. The method according to any one of claims 1-7, characterized in that, The step of determining whether the optical network component to be predicted will fail within a preset time period based on the number of records of the component in the risk monitoring pool includes: If the number of records of the optical network component to be predicted in the risk monitoring pool is greater than or equal to a preset threshold, it is determined that the optical network component will fail within a preset time period in the future. If the number of records for the optical network component to be predicted in the risk monitoring pool is less than the preset threshold, then it is determined that the optical network component to be predicted will not fail within a preset time period in the future.

9. The method according to any one of claims 1-8, characterized in that, The failure prediction results include mirror catastrophic failure prediction results and intrinsic failure prediction results.

10. The method according to any one of claims 1-9, characterized in that, The input data for the failure prediction model also includes expert experience data and long-term historical statistical data, among which The expert experience data includes data processed based on expert experience for the target work performance indicators; The long-term statistical data includes statistical feature data of the target work performance index data within a preset historical time period.

11. The method according to any one of claims 1-10, characterized in that, Also includes: If it is determined that the optical network component to be predicted will fail within a preset time period in the future, an alarm will be reported for the optical network component to be predicted.

12. The method according to any one of claims 1-11, characterized in that, Also includes: Display the contents recorded by the risk monitoring pool.

13. A failure prediction device for an optical network component, characterized in that, include: The acquisition module is used to acquire target performance index data of the optical network components to be predicted; The model prediction module is used to take the target performance index data as input to the failure prediction model and output the failure prediction result. An update module is used to update the risk monitoring pool based on the failure prediction results. The risk monitoring pool records optical network components with failure risks and the number of times each optical network component is recorded. The determination module is used to predict whether the optical network component to be predicted will fail within a preset time period in the future, based on the number of times the optical network component to be predicted is recorded in the risk monitoring pool.

14. A network management device, comprising a memory and a processor, characterized in that, The memory stores instructions that, when executed by a processor, cause the method described in any one of claims 1-12 to be implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it causes the method as described in any one of claims 1-12 to be implemented.

16. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, causes the method as described in any one of claims 1-12 to be implemented.

Images