Abnormality detection method and device for optical communication link in data center network

Abstract

The invention provides an anomaly detection method and equipment for an optical communication link in a data center network, which can be applied to the technical field of optical communication technologies and artificial intelligence data centers. The anomaly detection method for the optical communication link comprises the following steps: acquiring operation data of a transmitting end optical module of the optical communication link at a plurality of moments, wherein the operation data comprises a bias current value, a driving voltage value and a temperature value; based on the sequential relationship among the multiple operation data, transmitting power detection is carried out on the transmitting end optical module to obtain the light transmitting power of the transmitting end optical module at a specified moment, and the specified moment is later than at least one moment; and performing anomaly detection on the optical communication link based on the optical transmitting power and the optical receiving power of the receiving end optical module of the optical communication link at the specified moment to obtain an anomaly detection result.

Description

Technical Field

[0001] This application relates to the fields of optical communication technology and artificial intelligence data center technology, and more specifically, to a method and device for anomaly detection of optical communication links in a data center network. Background Technology

[0002] With the rapid development of artificial intelligence technology, data center networks have widely adopted optical communication-based interconnection solutions. Optical modules are the core physical components for realizing optical communication, used to efficiently convert electrical signals into optical signals, thereby supporting high-speed, long-distance data transmission. The performance of optical modules directly determines the stability and reliability of the optical communication link, thus affecting the transmission rate and stability of the entire network. Therefore, accurate anomaly detection of optical communication links has become a critical aspect of operation and maintenance.

[0003] However, existing anomaly detection schemes for optical communication links suffer from problems such as insufficient reliability and poor detection results. Summary of the Invention

[0004] In view of the above problems, this application provides a method and device for anomaly detection of optical communication links in data center networks. This application also provides an apparatus, device, medium, and program product for anomaly detection of optical communication links in data center networks.

[0005] According to one aspect of this application, an anomaly detection method for optical communication links in a data center network is provided, comprising: acquiring operational data of a transmitting optical module of the optical communication link at multiple time points, the operational data including bias current value, driving voltage value, and temperature value; performing transmit power detection on the transmitting optical module based on the timing relationship between the multiple operational data points to obtain the optical transmit power of the transmitting optical module at a specified time point, the specified time point being later than at least one time point; and performing anomaly detection on the optical communication link based on the optical transmit power and the optical receive power of the receiving optical module of the optical communication link at the specified time point to obtain anomaly detection results.

[0006] Another aspect of this application provides an anomaly detection device for optical communication links in a data center network, comprising: an acquisition module for acquiring operational data of a transmitting optical module of the optical communication link at multiple times, the operational data including bias current value, driving voltage value, and temperature value; a obtaining module for detecting the transmission power of the transmitting optical module based on the timing relationship between the multiple operational data, obtaining the optical transmission power of the transmitting optical module at a specified time, the specified time being later than at least one time; and a detection module for performing anomaly detection on the optical communication link based on the optical transmission power and the optical reception power of the receiving optical module of the optical communication link at the specified time, obtaining anomaly detection results.

[0007] Another aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0008] Another aspect of this application provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0009] Another aspect of this application provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0010] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0011] Figure 1 The illustration shows an application scenario of an anomaly detection method, apparatus, device, medium, and program product for optical communication links in data center networks according to embodiments of this application.

[0012] Figure 2 A flowchart of an anomaly detection method for optical communication links in a data center network according to an embodiment of this application is shown.

[0013] Figure 3A A schematic diagram is shown illustrating feature extraction of multiple operational data using a gated loop unit according to an embodiment of this application.

[0014] Figure 3B A schematic diagram of a gated loop unit according to an embodiment of this application is shown.

[0015] Figure 4 A schematic diagram of the loss value of a gated recurrent network during the training phase according to an embodiment of this application is shown.

[0016] Figure 5 A structural block diagram of an anomaly detection device for optical communication links in a data center network according to an embodiment of this application is shown.

[0017] Figure 6 A block diagram of an electronic device suitable for implementing an anomaly detection method for optical communication links in a data center network, according to an embodiment of this application, is shown. Detailed Implementation

[0018] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0019] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0020] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0021] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0022] In recent years, AI (Artificial Intelligence) content generation technology has developed rapidly and its applications have become increasingly widespread, with the number of parameters and the scale of training data for large-scale language models continuing to rise. This trend has not only significantly increased the complexity and processing performance of models, but also placed higher demands on computing infrastructure and storage resources. The size of current mainstream models and their corresponding computing power requirements far exceed what a single server can handle. Against this backdrop, traditional centralized deployment models are no longer sufficient to meet the needs of practical applications, and the training and inference tasks of large-scale models are gradually shifting to distributed cluster architectures. These distributed cluster architectures rely on multiple high-performance computing nodes (such as GPU (Graphics Processing Unit) servers) to work together to complete complex computational tasks. During the operation of a distributed computing system, each computing node needs to continuously exchange a large amount of data to support key operations such as gradient synchronization, parameter transfer, and intermediate result aggregation in model training and inference. Such communication is typically characterized by high frequency and high bandwidth, so the transmission capacity and stability of the interconnection network between nodes directly determine the overall computing efficiency and task processing speed of the system. Network communication performance has become a key bottleneck restricting the expansion and performance improvement of distributed computing systems. In response, optical communication-based interconnection solutions can effectively reduce latency, improve communication efficiency, and provide solid support for system computing power.

[0023] Data center networks can include Artificial Intelligence Data Centers (AIDCs) based on the aforementioned distributed cluster architecture. AIDCs can provide parallel computing based on clusters of multiple GPUs to handle the massive computations required for AI model training and inference. High-speed data center networks commonly employ interconnection schemes based on optical communication. Optical communication is a communication method that uses light signals to transmit information. It uses light as a carrier, modulates various properties of light to load information, and transmits these information-carrying optical signals in a transmission medium. At the receiving end, the optical signals are converted into electrical signals or other forms of signals to achieve information transmission. For example, in data center scenarios, optical communication can be used for high-speed data transmission between servers within a data center, between racks, and between data centers.

[0024] Optical transceiver modules (OTMs) are core components of optical communication systems, responsible for converting between optical and electrical signals to support high-speed, long-distance data transmission. The performance of an OTM directly determines the stability and reliability of the optical communication link, thus affecting the overall network transmission rate and stability. However, in large-scale deployment and operation environments, due to the complexity of the OTM's own operating mechanism and the external physical environment, failures are difficult to eliminate. Common failure modes include progressive performance degradation, complete module failure, and communication link interruption. When a failure occurs, it often significantly increases inter-node communication latency, triggers numerous data packet retransmissions, and may even lead to a complete network connection interruption. Especially during critical phases of model training or distributed inference, OTM failures can have severe impacts: not only can they cause computational task interruptions or failures, wasting significant computing resources, but they can also delay the overall task progress and reduce training and inference efficiency. More importantly, the interruption of critical services can also cause significant economic losses and operational risks.

[0025] Therefore, accurately detecting anomalies in the operating status of optical modules and providing rapid early warning and fault handling in case of performance degradation or failure has become a core challenge in ensuring the stable and efficient operation of large-scale AI computing clusters. Overcoming this technical challenge is of great significance for improving the reliability and resource utilization of distributed computing systems. However, anomaly detection schemes for optical communication links still face many challenges.

[0026] For example, Digital Diagnostic Monitoring (DDM) technology can be used to monitor the operating parameters of optical modules (such as transmit power and receive power) to determine the connectivity and quality status of the optical communication link, and to provide fault warnings and location. However, this DDM-based monitoring method has inherent defects; its reliability heavily depends on the accuracy of the transmit optical power value reported by the optical module. The transmit optical power monitored by DDM technology is not a true value directly measured by physical sensors, but a fitted or estimated value calculated by the microcontroller inside the optical module based on its built-in fitting function (usually calibrated by the optical module manufacturer before shipment). However, this fitting function is difficult to adapt to temperature drift errors and component aging drift errors that exist in actual applications. Due to the existence of these errors, the reliability of conclusions drawn from judging the health status of the entire optical communication link based on a potentially inaccurate "transmit power" value is questionable. This could lead to two serious consequences: first, "missed reports," meaning that the actual performance of the optical module has deteriorated, but the system fails to issue a timely warning, ultimately causing a link interruption; second, "false alarms," ​​meaning that inaccurate readings result in incorrect alarms, increasing the complexity and cost of operation and maintenance.

[0027] It should be noted that although optical power meters, as instruments specifically designed to measure optical power, can accurately measure the optical power of optical modules, in actual industrial scenarios (especially large-scale online scenarios such as AI clusters), it is impractical to use optical power meters to measure the optical power of a large number of optical modules over a period of time. This would also affect the normal operation of the system due to service interruption caused by fiber disconnection. Therefore, the industry generally adopts DDM technology to monitor the transmit power of optical modules.

[0028] For example, deep learning or machine learning models can be used to identify and predict failure modes of optical modules, thereby achieving anomaly detection. However, model training heavily relies on a large number of high-quality failure samples. In absolute terms, the failure rate of optical modules is typically very low due to their high reliability and durability design, making it extremely difficult to collect a sufficient number of failure samples in reality. This poses a substantial challenge to model training and optimization, limiting the model's generalization ability and predictive accuracy. Furthermore, constructing and labeling a sufficiently large number of failure data samples is also costly, all of which contribute to the difficulties in model training and optimization.

[0029] In view of this, embodiments of this application provide an anomaly detection method for optical communication links in data center networks. The method acquires operational data of the transmitting optical module of the optical communication link at multiple time points, including bias current, driving voltage, and temperature values. Based on the timing relationship between the multiple operational data points, the transmitting optical module's transmit power is detected to obtain the optical transmit power of the transmitting optical module at a specified time point, where the specified time point is later than at least one time point. Based on the optical transmit power and the optical receive power of the receiving optical module of the optical communication link at the specified time point, anomaly detection is performed on the optical communication link to obtain an anomaly detection result.

[0030] Figure 1 The illustration shows an application scenario diagram of an anomaly detection method and apparatus for optical communication links in a data center network according to embodiments of this application.

[0031] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0032] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0033] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0034] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0035] It should be noted that the anomaly detection method for optical communication links in data center networks provided in this application embodiment can generally be executed by server 105. Correspondingly, the anomaly detection device for optical communication links in data center networks provided in this application embodiment can generally be located in server 105. The anomaly detection method for optical communication links in data center networks provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the anomaly detection device for optical communication links in data center networks provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0036] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0037] Figure 2 A flowchart of an anomaly detection method for optical communication links in a data center network according to an embodiment of this application is shown.

[0038] like Figure 2As shown, the anomaly detection method for optical communication links in the data center network includes operations S210 to S230.

[0039] In operation S210, the operating data of the transmitting optical module of the optical communication link at multiple times is acquired. The operating data includes bias current value, drive voltage value and temperature value.

[0040] In operation S220, based on the timing relationship between multiple operating data, the transmit power of the transmitting optical module is detected to obtain the optical transmit power of the transmitting optical module at a specified time, where the specified time is later than at least one time.

[0041] In operation S230, based on the optical transmission power and the optical reception power of the receiving optical module of the optical communication link at a specified time, anomaly detection is performed on the optical communication link, and the anomaly detection result is obtained.

[0042] Optical communication links in data center networks include transmitting optical modules and receiving optical modules.

[0043] For the optical communication link under test, the operating data of the transmitting optical module in the link at multiple times can be obtained. The operating data includes the bias current value ( ), driving voltage value ( ) and temperature value ( Wherein, the bias current value is the bias current of the laser in the transmitter optical module, the driving voltage value is the driving voltage of the transmitter optical module, and the temperature value is the operating temperature of the transmitter optical module.

[0044] The optical transmission power of an optical module is a state value that changes continuously over time. The bias current, drive voltage, and temperature of the optical module at a certain moment can characterize the operating state of the optical module at that moment. Therefore, based on the timing relationship between multiple operating data, the transmission power of the transmitting optical module can be detected to obtain the optical transmission power of the transmitting optical module at a specified moment. ).

[0045] Anomaly detection can be performed on the optical communication link based on the optical transmission power of the transmitting optical module and the optical reception power of the receiving optical module at a specified time, and the anomaly detection result can be obtained. By comparing the optical transmission power of the transmitting optical module and the reception power of the receiving optical module in real time, it is possible to automatically and accurately determine whether the optical communication link is abnormal.

[0046] According to embodiments of this application, the optical transmission power of the transmitting optical module at a specified time can be determined based on the temporal relationship between operational data at multiple time points. This allows for anomaly detection of the optical communication link based on the transmission power and the optical reception power of the receiving optical module at that specified time, yielding anomaly detection results. Therefore, by performing real-time anomaly detection on optical communication links in data center networks, early identification and sensitive, accurate warnings of optical communication link anomalies can be achieved, effectively ensuring the stable and reliable operation of large-scale AI clusters.

[0047] It is understandable that, compared to anomaly detection methods based on DDM technology to monitor optical transmission power, the anomaly detection method for optical communication links provided in this application (hereinafter referred to as "this method") determines the optical transmission power used for anomaly detection based on the temporal relationship between multiple operational data. This overcomes the shortcomings of poor accuracy and low reliability in calculating transmitted optical power by relying on the built-in fitting function of the optical module, ensuring the reliability of the optical transmission power itself. Furthermore, compared to anomaly detection methods that use deep learning models to identify optical module fault modes, this method achieves anomaly detection based on the optical transmission power of the transmitting optical module and the optical receiving power of the receiving end. This eliminates the reliance on fault samples and manually labeled data, significantly reducing implementation costs and complexity.

[0048] According to an embodiment of this application, the transmission power detection of the transmitting optical module based on the temporal relationship between multiple operating data to obtain the optical transmission power of the transmitting optical module at a specified time includes: using the gated loop unit of the gated loop network to extract features from multiple operating data according to the temporal relationship to obtain the operating state features corresponding to each of the multiple times; and determining the optical transmission power based on the operating state features corresponding to the specified time.

[0049] In one embodiment, operational data of the transmitting optical module at multiple points within a specified time period can be acquired. These multiple points may include, for example, the current time and multiple historical times. The current time can be used as the specified time, and the transmit power of the transmitting optical module can be detected based on the temporal relationship between the multiple operational data points to obtain the optical transmit power of the transmitting optical module at the current time. For example, the operational data at each of the multiple times can be arranged chronologically to obtain an operational data sequence. For example, the operational data sequence can be input into a preset temporal prediction model to output the optical transmit power of the transmitting optical module at the current time. The preset temporal prediction model can achieve high-precision fitting and prediction of the optical transmit power.

[0050] In one embodiment, the preset timing prediction model may include a gated recurrent network. The gated recurrent units of the gated recurrent network can be used to extract features from multiple operational data points based on temporal relationships, obtaining operational state features corresponding to multiple time points. For any given time point, the operational state features can characterize the operational state of the transmitting optical module at that time. For example, the current time can be used as a specified time point, and the optical transmission power of the transmitting optical module at the current time can be determined based on the operational state features corresponding to the current time point.

[0051] The Gated Recurrent Unit (GRU) controls the flow of information by introducing two gating mechanisms: a reset gate and an update gate. This effectively captures both short-term and long-term dependencies in time-series data, solving the vanishing and exploding gradient problems found in traditional recurrent neural networks. By leveraging gating mechanisms for fine-grained control of information flow, GRU can improve the modeling ability and prediction accuracy of complex time-series relationships while maintaining computational efficiency.

[0052] Understandably, by utilizing a gated loop unit to extract features from multiple operational data points based on temporal relationships, short-term and long-term dependencies between these data points can be effectively captured. This enables adaptive memorization and forgetting of information, allowing the system to learn the operational state characteristics and change patterns of the transmitting optical module at multiple moments. Consequently, high-precision detection of optical transmission power at a specified moment (such as the current moment) can be achieved, significantly improving the fitting accuracy of transmission power under normal conditions.

[0053] According to an embodiment of this application, for any given moment of running data, the reset gate of the gated loop unit controls the proportion of discarded associated running state features to the total associated running state features by controlling the degree of influence of associated running state features among multiple running state features on the running data, so as to obtain candidate running state features for that moment; wherein, the moment corresponding to the associated running state feature is earlier than the moment of the running data, and the running state feature corresponding to the running data is determined based on the candidate running state features.

[0054] According to an embodiment of this application, for any given moment of running data, the update gate of the gated loop unit controls the proportion of each of the candidate running state features and the associated running state features in the running state features at that moment by fusing the candidate running state features and the associated running state features, so as to obtain the running state features at that moment.

[0055] For any given moment of runtime data, the moment corresponding to the associated runtime state feature is earlier than the moment of the runtime data. The associated runtime state feature can be understood as representing the runtime state at a historical moment; for example, the runtime state feature at the previous moment can be used as the associated runtime state feature.

[0056] Figure 3A A schematic diagram is shown illustrating feature extraction of multiple operational data using a gated loop unit according to an embodiment of this application.

[0057] like Figure 3A As shown, multiple time points can include, for example, time points t1 to t10. The operating data of the transmitting optical module from time point t1 to t10 includes x1 to x10. For each time point, the operating data includes bias current value, drive voltage value, and temperature value. Multiple operating data points can be arranged in chronological order (e.g., x1, x2, ..., x10) and input into a gated recurrent network. The GRU unit of the gated recurrent network can extract features from the operating data x1 to x10 sequentially according to the chronological order, obtaining the respective operating state features from time point t1 to t10, and finally outputting the optical emission frequency at a specified time (e.g., time t10).

[0058] For example, at time t1, the initial features can be preset (usually chosen as all-zero vectors). Figure 3A (Not shown in the image) is used as the associated running feature. The GRU unit takes the preset initialization feature and x1 as input and outputs the running state feature at time t1. Figure 3A (See h1). For example, for time t2, the operating state characteristics at time t1 can be used as associated operating characteristics. The GRU unit takes the operating state characteristics at time t1 and x2 as inputs and outputs the operating state characteristics at time t2. Similarly, for time t3, the operating state characteristics at time t2 can be used as associated operating characteristics. The GRU unit takes the operating state characteristics at time t2 and x3 as inputs and outputs the operating state characteristics at time t3. Finally, the GRU unit can determine the optical emission frequency at time t10 based on the operating state characteristics at time t10.

[0059] For any given moment, the associated runtime features not only retain past memory information but can also be updated or reset through two gating mechanisms to control the final output runtime features at that moment. This design allows the network to maintain long-term contextual information in sequential tasks while adapting to data changes. In the GRU unit, the reset gate helps capture short-term dependencies in the time series, while the update gate helps capture long-term dependencies. These two gating mechanisms work together to control the flow of information, specifically:

[0060] The reset gate determines how much past information needs to be forgotten. The reset gate value is a value between 0 and 1, adjusted by a sigmoid activation function. When the reset gate is close to 0, the network "forgets" more past information; when it's close to 1, it retains more. This helps the network dynamically retain important historical information while forgetting unimportant information when processing sequential data. For example, for time t2, the reset gate controls the proportion of discarded time t1 running state features relative to the complete time t1 running state features by controlling the influence of the running state features at time t1 on the running data at time t2, thus obtaining candidate running state features for time t2.

[0061] The update gate controls the extent to which the network should consider new information (new input information) at the current time step and old information (past states) from the previous time step. The update gate value is also a value between 0 and 1, derived through the sigmoid activation function. The update gate helps the network determine how much previous information should be retained and how much new information should be introduced in the current state, thus maintaining the network's flexibility when processing different types of sequence data. For example, for time t2, the update gate controls the proportion of each of the candidate running state features at time t2 and the running state features at time t1 within the total running state features at time t2, thus obtaining the running state features at time t2.

[0062] The combination of reset and update gates makes GRU more effective and powerful when handling tasks that require remembering long-term dependencies, such as time series data. These two gating mechanisms work together to dynamically adjust the degree of retention of past information and the ratio of integration of new input information in the runtime state features at each time step. As a result, GRU can better capture the temporal relationships between multiple runtime data and mitigate the problems of vanishing and exploding gradients.

[0063] According to an embodiment of this application, the multiple operational data include operational data for N time periods arranged chronologically, where N is a positive integer greater than 1. For the nth time period, the (n-1)th operational state feature at the (n-1)th time period is used as an associated operational state feature. The nth candidate operational state feature at the nth time period is determined based on the following operations, where n = 2, ..., N: Based on the weight of the reset gate, reset gate information is determined according to the nth operational data at the nth time period and the (n-1)th operational state feature. The (n-1)th operational state feature is filtered using the reset gate information, and the nth candidate operational state feature at the nth time period is determined based on the nth operational data and the filtered (n-1)th operational state feature. The reset gate information is used to control the proportion of the filtered (n-1)th operational state feature to the total (n-1)th operational state feature.

[0064] According to an embodiment of this application, for time n, the nth running state feature at time n is determined based on the following operations, where n = 2, ..., N: Based on the weight of the update gate, update gate information is determined according to the nth running data and the (n-1)th running state feature. The (n-1)th running state feature and the nth candidate running state feature are fused using the update gate information to obtain the nth running state feature at time n; the update gate information is used to control the proportion of the (n-1)th running state feature and the nth candidate running state feature in the nth running state feature; when n = N, the nth running state feature at time N is determined as the running state feature at the specified time.

[0065] Figure 3B A schematic diagram of a gated loop unit according to an embodiment of this application is shown.

[0066] Taking time n as denoted as time t and time n-1 as denoted as time t-1 as an example, Figure 3B The diagram illustrates the specific implementation principle of the GRU unit at time t. The GRU unit is jointly controlled by two logic gates: a reset gate and an update gate. Through these two gating mechanisms, the GRU unit can dynamically adjust the information flow to fully capture and accurately model the temporal relationships between multiple running data.

[0067] exist Figure 3B In this context, "reset gate" refers to the gate that resets the data, and "update gate" refers to the gate that updates the data. This represents a non-linear activation function (sigmoid activation function). Its output value range is . This represents a non-linear activation function (tanh activation function). Its output value range is .

[0068] like Figure 3B As shown, at time t, the input of a GRU cell includes the hidden state at time t-1. (That is, the running state characteristics at time t-1, used as associated state characteristics) and the current input data at time t. (That is, the running data at time t). Let be the hidden state output by the GRU unit at time t-1. At time t, the output of the GRU unit includes the hidden state at time t. (That is, the characteristics of the operating state at time t).

[0069] like Figure 3B As shown above, the input at time t can be determined based on the weights of the reset gate. and the hidden state at time t-1 Confirm door information reset The specific formula is expressed as follows:

[0070]

[0071] in, and As weight, For bias. Represents the input at time t Impact on the reset door Represents the hidden state at time t-1 The impact on the reset door.

[0072] Afterwards, the door information can be reset. right Perform filtering, and based on and after screening Determine the candidate hidden state at time t. (That is, the candidate running characteristics at time t), the specific formula is expressed as follows:

[0073]

[0074] in, and As weight, For bias. Indicates resetting the door selection Important information in the middle (i.e., the filtered information) ), This indicates element-wise multiplication. For example, The smaller the value, the better. The smaller the value, the more information from time t-1 needs to be discarded. Conversely, The larger the value, the better. The larger the value, the less information from time t-1 needs to be discarded. The information from time t-1 that is not discarded is used to combine with the input at time t to generate candidate hidden states at time t. It is understandable that by controlling... The value of t-1 can determine whether and to what extent information at time t-1 is discarded. Thus, the reset gate information can be used to control the degree to which the hidden state at time t-1 is discarded, thereby helping to capture short-term dependencies in multiple running data.

[0075] like Figure 3B As shown below, the input at time t can be determined based on the weights of the update gate. and the hidden state at time t-1 Confirm door information update The specific formula is expressed as follows:

[0076]

[0077] in, and Indicates weight, Indicates bias. Represents the input at time t Impact on the update gate Represents the hidden state at time t-1 Impact on the update gate.

[0078] Afterwards, the updated door information can be used. right and We perform weighted fusion to obtain the hidden state at time t. (That is, the operating state characteristics at time t), the specific formula is expressed as follows:

[0079]

[0080] in, Indicates the extent to which it is retained. To generate , Indicates the extent to which it is introduced. To generate It is understandable that by controlling... The value of can determine and Each in The proportion of the hidden state in the data is such that updating the gate information can be used to control the fusion ratio of the candidate hidden state at time t and the hidden state at time t-1, thereby helping to capture long-term dependencies in multiple running data.

[0081] By using the reset gate and update gate in tandem, the degree of retention of associated operational states and the fusion ratio of operational data in the operational state features at any given time can be dynamically adjusted, thereby effectively capturing long-term and short-term dependencies in time series data. The update gate determines the hidden state at time t-1. How much information should be transmitted to time t? The reset door controls... The amount of information that should be ignored or reset in order to provide runtime data at time t. Make room. After the GRU unit iterates, the final output is the optical transmission power of the transmitter optical module at a specified time. Assuming the specified time is time t, the final output is the hidden state at time t. .

[0082] According to an embodiment of this application, the gated recurrent network is obtained based on the following operations: acquiring sample operation data of the sample optical module at multiple sample times in the working state, the sample operation data including sample bias current, sample drive voltage and sample temperature; inputting the multiple sample operation data into the candidate gated recurrent network to obtain the predicted transmit power value of the sample optical module at a specified sample time; the specified sample time is later than at least one sample time; training the candidate gated recurrent network based on the predicted transmit power value and the real label corresponding to the predicted transmit power value to obtain the gated recurrent network; wherein, the real label corresponding to the predicted transmit power value is: the optical received power of the peer optical module that transmits data with the sample optical module at the specified sample time.

[0083] The sample optical module can be understood as the transmitting optical module in the sample optical communication link, and the peer optical module that transmits data with the sample optical module can be understood as the receiving optical module in the sample optical communication link.

[0084] The gated recurrent network of this application embodiment can be trained based on sample operation data of sample optical modules in working state at multiple sample times. In addition, during the training phase, the optical received power of the peer optical module at a specified time is used as the real label. This avoids the dependence on fault samples and manually labeled data, greatly reducing the implementation cost and complexity. It allows a high-precision gated recurrent network to accurately fit the real dynamic characteristics of the optical module's transmit power by only collecting a large amount of sample operation data of sample optical communication links in normal state.

[0085] Furthermore, considering the influencing factors of optical module transmission power and the actual model training effect, this embodiment of the application selects sample bias current, sample driving voltage, and sample temperature as sample running data to train the gated recurrent network, as detailed below:

[0086] Optical module manufacturers incorporate an output power fitting function within the optical module. According to this function, the output optical power is affected by the bias current, and different temperatures result in different optical output power fitting curves. For example, even with the same bias current and modulation current, different temperatures will lead to different emitted optical power outputs. Although the device parameters of the optical module will age during long-term deployment, and the temperature may differ from the manufacturer's preset temperature value, this indirectly proves that the optical module's emitted power is closely related to the bias current and temperature.

[0087] Therefore, in this embodiment, bias current and temperature are chosen as features for training the neural network. As for whether the driving voltage should be used as a feature for training the neural network, based on bias current and temperature, a practical network training comparison experiment was conducted for two cases: with driving voltage and without driving voltage.

[0088] Figure 4 A schematic diagram of the loss value of a gated recurrent network during the training phase according to an embodiment of this application is shown.

[0089] Figure 4 (a) shows a comparison of loss values ​​under different conditions during the overall training process. Figure 4 (b) is a magnified view of the later stage of the training process in (a).

[0090] like Figure 4 As shown in (a), during the overall training process, the solid line representing "training loss of network with driving voltage" and the dashed line representing "test loss of network with driving voltage" are always below the dotted-dash line representing "training loss of network without driving voltage" and the dotted-dash line representing "test loss of network without driving voltage," and the convergence of the solid and dashed lines is smoother and less volatile. In contrast,... Figure 4 In (b), this difference is even more pronounced. The solid and dashed lines corresponding to the network with driving voltage have converged to a lower loss level and continue to decrease slowly, while the dotted and dashed lines corresponding to the network without driving voltage fluctuate within a higher loss range.

[0091] This shows that, under the same model structure and training conditions, simply adding the "driving voltage" input feature can simultaneously reduce training error and testing error and improve convergence stability. It can be considered that introducing the driving voltage as the network's input feature significantly improves the performance of the neural network. Therefore, choosing the driving voltage as the input feature is necessary and effective.

[0092] According to an embodiment of this application, before acquiring the operating data of the transmitting optical module of the optical communication link at multiple times, the method further includes: acquiring the original operating data of the transmitting optical module at each of the multiple times, the original operating data including the original bias current, the original driving voltage and the original temperature; determining the preset upper limit alarm threshold and the preset lower limit alarm threshold corresponding to the original operating data of the transmitting optical module as the normalized upper limit value and the normalized lower limit value for the original operating data, and performing normalization processing on the multiple original operating data based on the normalized upper limit value and the normalized lower limit value to obtain multiple operating data.

[0093] In one embodiment, methods such as Telemetry (a remote data acquisition technology), SSH (Secure Shell), API (Application Programming Interface), and SNMP (Simple Network Management Protocol) can be used to acquire the raw operating data of the transmitting optical module at multiple points in time. This raw operating data includes the raw bias current, raw drive voltage, and raw temperature.

[0094] Given that optical modules from different manufacturers have varying high and low threshold values ​​for their operating data, to eliminate the impact of these different thresholds, preset upper and lower alarm thresholds corresponding to the original operating data of the transmitting optical module can be determined as normalized upper and lower limits for the original operating data. The original operating data is then normalized accordingly to obtain the final operating data. Specifically, the bias current, drive voltage, and temperature of the optical module each have their corresponding preset upper and lower alarm thresholds.

[0095] For example, the normalization formula is as follows:

[0096]

[0097] Where x represents the original operating data, such as the original bias current, original drive voltage, and original temperature. This indicates the preset upper limit alarm threshold corresponding to the original operating data. This represents the preset lower limit alarm threshold corresponding to the original data.

[0098] According to an embodiment of this application, based on the optical transmission power and the optical reception power of the receiving optical module of the optical communication link at a specified time, anomaly detection is performed on the optical communication link to obtain the anomaly detection result, including: when the link distance between the transmitting optical module and the receiving optical module is less than or equal to a preset distance threshold, comparing the difference between the optical transmission power and the optical reception power with a preset power threshold; when the difference is greater than the preset threshold, determining that the anomaly detection result indicates an anomaly in the optical communication link; when the difference is less than or equal to the preset threshold, determining that the anomaly detection result indicates that the optical communication link is normal.

[0099] Optical modules, as key components in optical fiber communication, incur losses when interconnecting to form optical communication links. These losses primarily fall into two categories: linear loss and insertion loss. Linear loss stems from the attenuation characteristics of the optical fiber itself, while insertion loss is caused by components such as connectors and interfaces and can generally be considered a constant. However, in data center networks, especially in high-performance network architectures within AIGC (Artificial Intelligence Generated Content) scenarios, the link distance between the transmitting and receiving optical modules is typically less than or equal to a preset distance threshold (e.g., within a few meters). Therefore, the impact of linear loss is negligible. Based on this premise, under normal optical module interconnection conditions, the received power of the receiving optical module can be considered as the actual output optical power of the transmitting optical module. The difference between the transmitted and received optical power can then be compared with a preset power threshold to verify whether the deviation between the actual link loss and the ideal loss predicted by the model exceeds the normal fluctuation range. This allows for automatic detection of any anomalies in the optical communication link, enabling early fault warnings.

[0100] For example, if the difference is less than or equal to a preset threshold, the anomaly detection result indicates that the optical communication link is normal. Conversely, if the difference is greater than the preset threshold, the anomaly detection result indicates that the optical communication link is abnormal. An abnormal optical communication link can include at least one of the following: an abnormality in the transmitting optical module, an abnormality in the receiving optical module, or an abnormality in the optical fiber.

[0101] According to an embodiment of this application, a data center network includes multiple interconnected node pairs that achieve optical communication via optical communication links; one of the interconnected node pairs is equipped with a transmitting optical module, and the other of the interconnected node pairs is equipped with a receiving optical module; one end of an optical fiber is connected to the transmitting port of the transmitting optical module, and the other end of the optical fiber is connected to the receiving port of the receiving optical module; the method further includes: if the anomaly detection result indicates that the optical communication link is abnormal, determining that the interconnected node pair corresponding to the optical communication link has a communication failure.

[0102] Data center network interconnection nodes can include two interconnection nodes that communicate via optical communication links (e.g., server and switch, or switch and switch).

[0103] In one embodiment, the interconnected node pair includes, for example, a server and a switch. For instance, the server has a transmitting optical module, and the switch has a receiving optical module; one end of the optical fiber is connected to the transmitting port of the transmitting optical module, and the other end of the optical fiber is connected to the receiving port of the receiving optical module. As another example, the switch has a transmitting optical module, the server has a receiving optical module, and one end of the optical fiber is connected to the transmitting port of the transmitting optical module, and the other end of the optical fiber is connected to the receiving port of the receiving optical module.

[0104] In one embodiment, if the anomaly detection result indicates an anomaly in the optical communication link, it can be determined that the interconnected node pair corresponding to the optical communication link has a communication fault, thereby realizing early warning of communication faults in interconnected node pairs in the data center network.

[0105] Figure 5 A structural block diagram of an anomaly detection device for optical communication links in a data center network according to an embodiment of this application is shown.

[0106] like Figure 5 As shown, the device 500 includes an acquisition module 510, a obtaining module 520, and a detection module 530.

[0107] The acquisition module 510 is used to acquire the operating data of the transmitting optical module of the optical communication link at multiple times. The operating data includes bias current value, driving voltage value and temperature value.

[0108] The module 520 is used to detect the transmit power of the transmitting optical module based on the timing relationship between multiple operating data, and obtain the optical transmit power of the transmitting optical module at a specified time, wherein the specified time is later than at least one time.

[0109] The detection module 530 is used to perform anomaly detection on the optical communication link based on the optical transmission power and the optical reception power of the receiving optical module of the optical communication link at a specified time, and obtain the anomaly detection result.

[0110] According to an embodiment of this application, the obtaining module 520 may include a first obtaining submodule and a first determining submodule.

[0111] The first submodule is used to extract features from multiple running data based on temporal relationships using the gated loop unit of the gated loop network, thereby obtaining the running state features corresponding to each time point.

[0112] The first determination submodule is used to determine the optical emission power based on the operating state characteristics corresponding to a specified time.

[0113] Specifically, for any given moment of the running data, the reset gate of the gated loop unit controls the proportion of discarded associated running state features to the total associated running state features by controlling the degree of influence of associated running state features among multiple running state features on the running data, so as to obtain candidate running state features for that moment; wherein, the moment corresponding to the associated running state feature is earlier than the moment of the running data, and the running state feature corresponding to the running data is determined based on the candidate running state features.

[0114] According to an embodiment of this application, for any given moment of running data, the update gate of the gated loop unit controls the proportion of each of the candidate running state features and the associated running state features in the running state features at that moment by fusing the candidate running state features and the associated running state features, so as to obtain the running state features at that moment.

[0115] According to an embodiment of this application, the multiple operational data include operational data for N time periods arranged in chronological order, where N is a positive integer greater than 1. For time n, the (n-1)th operational state feature at time n-1 is used as an associated operational state feature, and the nth candidate operational state feature at time n is determined based on the following operations, n=2,…,N: Based on the weight of the reset gate, reset gate information is determined according to the nth operational data and the (n-1)th operational state feature at time n; the (n-1)th operational state feature is filtered using the reset gate information, and the nth candidate operational state feature at time n is determined based on the nth operational data and the filtered (n-1)th operational state feature; the reset gate information is used to control the proportion of the filtered (n-1)th operational state feature to the total (n-1)th operational state feature.

[0116] According to an embodiment of this application, for time n, the nth running state feature at time n is determined based on the following operations, n=2,…,N: Based on the weight of the update gate, the update gate information is determined according to the nth running data and the (n-1)th running state feature; the update gate information is used to fuse the (n-1)th running state feature and the nth candidate running state feature to obtain the nth running state feature at time n; the update gate information is used to control the proportion of the (n-1)th running state feature and the nth candidate running state feature in the nth running state feature.

[0117] In the case of n=N, the Nth running state feature at time N is determined as the running state feature at the specified time.

[0118] According to an embodiment of this application, the gated recurrent network is obtained based on the following operations: acquiring sample operation data of the sample optical module at multiple sample times in the working state, the sample operation data including sample bias current, sample drive voltage and sample temperature; inputting the multiple sample operation data into the candidate gated recurrent network to obtain the predicted transmit power value of the sample optical module at a specified sample time; the specified sample time is later than at least one sample time; training the candidate gated recurrent network based on the predicted transmit power value and the real label corresponding to the predicted transmit power value to obtain the gated recurrent network; wherein, the real label corresponding to the predicted transmit power value is: the optical received power of the peer optical module that transmits data with the sample optical module at the specified sample time.

[0119] According to an embodiment of this application, the apparatus 500 further includes a preprocessing module.

[0120] The preprocessing module is used to acquire the original operating data of the transmitting optical module at multiple times. The original operating data includes the original bias current, the original driving voltage, and the original temperature. The preset upper limit alarm threshold and preset lower limit alarm threshold corresponding to the original operating data of the transmitting optical module are determined as the normalized upper limit value and normalized lower limit value for the original operating data. Based on the normalized upper limit value and normalized lower limit value, the multiple original operating data are normalized respectively to obtain multiple operating data.

[0121] According to an embodiment of this application, the detection module 530 includes a comparison submodule, a second determination submodule, and a third determination submodule.

[0122] The comparison submodule is used to compare the difference between the optical transmit power and the optical receive power with a preset power threshold when the link distance between the transmitting optical module and the receiving optical module is less than or equal to a preset distance threshold.

[0123] The second determination submodule is used to determine the anomaly detection result to characterize the optical communication link anomaly when the difference is greater than a preset threshold.

[0124] The third determination submodule is used to determine whether the abnormal detection result indicates that the optical communication link is normal when the difference is less than or equal to a preset threshold.

[0125] According to an embodiment of this application, a data center network includes multiple interconnected node pairs that realize optical communication through optical communication links; one of the interconnected node pairs is equipped with a transmitting optical module, and the other of the interconnected node pairs is equipped with a receiving optical module. One end of the optical fiber is connected to the transmitting port of the transmitting optical module, and the other end of the optical fiber is connected to the receiving port of the receiving optical module.

[0126] The device 500 also includes a determination module, which is used to determine that there is a communication failure in the interconnection node pair corresponding to the optical communication link when the anomaly detection result indicates that the optical communication link is abnormal.

[0127] According to embodiments of this application, any plurality of modules among the acquisition module 510, the obtaining module 520, and the detection module 530 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this application, at least one of the acquisition module 510, the obtaining module 520, and the detection module 530 can be at least partially implemented as a hardware circuit, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented by any other reasonable means of integrating or packaging the circuit, or implemented in any one of the three methods of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, at least one of the acquisition module 510, the obtaining module 520, and the detection module 530 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0128] Figure 6 A block diagram of an electronic device suitable for implementing an anomaly detection method for optical communication links in a data center network, according to an embodiment of this application, is shown.

[0129] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a ROM 602 (Read-Only Memory) or a program loaded from a storage portion 608 into a RAM 603 (Random Access Memory). The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0130] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0131] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0132] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0133] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0134] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0135] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0136] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0137] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0138] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0139] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0140] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0141] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A method for anomaly detection of optical communication links in a data center network, the method comprising: The optical communication link in the data center network comprises a transmitting optical module and a receiving optical module, and the method comprises: obtaining running data of the transmitting optical module of the optical communication link at multiple time instants, the running data comprising bias current values, driving voltage values and temperature values; based on the time sequence relationship between the multiple running data, performing transmitting power detection on the transmitting optical module to obtain optical transmitting power of the transmitting optical module at a specified time instant, the specified time instant being later than at least one of the time instants; based on the optical transmitting power and optical receiving power of the receiving optical module of the optical communication link at the specified time instant, performing anomaly detection on the optical communication link to obtain an anomaly detection result.

2. The method of claim 1, wherein, The method comprises: using a gating loop unit of a gated recurrent network to perform feature extraction on the multiple running data according to the time sequence relationship to obtain running state features corresponding to the multiple time instants respectively; and determining the optical transmitting power based on a running state feature corresponding to the specified time instant; wherein, for running data of any time instant, a reset gate of the gating loop unit controls the proportion of associated running state features discarded in the associated running state features by controlling the influence degree of the associated running state features on the running data, to obtain a candidate running state feature of the time instant; wherein, the time instant corresponding to the associated running state features is earlier than the time instant of the running data, and a running state feature corresponding to the running data is determined according to the candidate running state feature.

3. The method of claim 2, wherein, The multiple running data comprise running data of N time instants arranged in time sequence, N being a positive integer greater than 1; for the nth time instant, the (n-1)th running state feature of the (n-1)th time instant is taken as the associated running state feature, and the (n)th candidate running state feature of the nth time instant is determined based on the following operations, n=2,…,N: based on the weight of the reset gate, determining reset gate information according to the (n)th running data of the nth time instant and the (n-1)th running state feature; using the reset gate information to filter the (n-1)th running state feature, and determining the (n)th candidate running state feature of the nth time instant based on the (n)th running data and the filtered (n-1)th running state feature; the reset gate information is used to control the proportion of the filtered (n-1)th running state feature in the (n-1)th running state feature.

4. The method of claim 3, wherein, For running data of any time instant, the update gate of the gating loop unit controls the proportion of the candidate running state feature and the associated running state feature in the running state feature of the time instant by fusing the candidate running state feature and the associated running state feature, to obtain the running state feature of the time instant.

5. The method of claim 4, wherein, for the nth time instant, the (n)th running state feature of the nth time instant is determined based on the following operations, n=2,…,N: determine, based on the updated gate weight, the nth running data and the (n-1)th running state feature, updated gate information; fuse the (n-1)th running state feature and the nth candidate running state feature based on the updated gate information to obtain the nth running state feature at the nth time point; the updated gate information is used to control the proportion of the (n-1)th running state feature and the nth candidate running state feature in the nth running state feature; in the case of n=N, the Nth running state feature at the Nth time point is determined as the running state feature at the specified time point.

6. The method of claim 1, wherein, Before obtaining the running data of the transmitting optical module of the optical communication link at multiple time points, the method further comprises: obtaining the original running data of the transmitting optical module at multiple time points respectively, the original running data including original bias current, original driving voltage and original temperature; determining the preset upper limit alarm threshold and the preset lower limit alarm threshold corresponding to the original running data of the transmitting optical module as the normalized upper limit value and the normalized lower limit value for the original running data, and performing normalization processing on multiple original running data respectively based on the normalized upper limit value and the normalized lower limit value to obtain multiple running data.

7. The method of claim 1, wherein, The abnormal detection of the optical communication link based on the optical transmitting power and the optical receiving power of the receiving optical module of the optical communication link at the specified time point includes: in the case that the link distance between the transmitting optical module and the receiving optical module is less than or equal to a preset distance threshold, comparing the difference between the optical transmitting power and the optical receiving power with a preset power threshold; in the case that the difference is greater than the preset threshold, determining that the abnormal detection result represents that the optical communication link is abnormal; in the case that the difference is less than or equal to the preset threshold, determining that the abnormal detection result represents that the optical communication link is normal.

8. The method according to any one of claims 1-7, characterized in that, The data center network includes multiple interconnected node pairs that realize optical communication through the optical communication link; one of the interconnected node pairs is deployed with the transmitting optical module, and the other of the interconnected node pairs is deployed with the receiving optical module, one end of the optical fiber is connected with the transmitting port of the transmitting optical module, and the other end of the optical fiber is connected with the receiving port of the receiving optical module; the method further comprises: in the case that the abnormal detection result represents that the optical communication link is abnormal, determining that the interconnected node pair corresponding to the optical communication link has a communication fault.

9. The method according to any one of claims 2-5, characterized in that, The gated recurrent network is obtained based on the following operations: obtaining sample running data of the sample optical module at multiple sample time points in the working state, the sample running data including sample bias current, sample driving voltage and sample temperature; inputting multiple sample running data into a candidate gated recurrent network to obtain a predicted value of the transmitting power of the sample optical module at a sample specified time point; the sample specified time point is later than at least one of the sample time points; The candidate gated recurrent network is trained according to the transmission power prediction value and a true label corresponding to the transmission power prediction value to obtain the gated recurrent network; wherein the true label corresponding to the transmission power prediction value is: an optical receiving power of an opposite end optical module of the sample optical module for data transmission at a specified moment of the sample. 10.An electronic device, comprising: one or more processors; memory for storing one or more computer programs, characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-9.

Images