Method and apparatus for fault location and severity estimation in optical transmission systems and media

Abstract

The application provides a fault positioning and severity estimation method, equipment and medium for an optical transmission system, the method comprising: artificially introducing a fault in an optical transmission experimental system and collecting multimodal data generated by a digital signal processing link; constructing a prompt word template that fuses experimental conditions and fault knowledge; based on the template and the multimodal data, fine-tuning an open-source multimodal large language model using a low-rank adaptive algorithm to obtain a fault diagnosis model; and finally, comprehensively analyzing multimodal data of an actual system using the model to automatically output an estimation result of a fault type and severity. The application realizes automated, accurate and interpretable diagnosis of faults in an optical transmission system, effectively improving operation and maintenance efficiency and system reliability.

Description

Technical Field

[0001] This invention relates to the field of optical transmission technology, and more specifically, to a method, equipment, and medium for fault location and severity estimation in optical transmission systems. Background Technology

[0002] As the backbone of modern communication networks, the stability and reliability of optical transmission systems directly impact the operational efficiency of the entire information society. With the explosive growth in data traffic demand, the complex system environment makes them more susceptible to various faults, such as modulator phase deviation, signal power imbalance, and receiver sampling synchronization issues. These faults not only degrade signal quality but can also cause communication outages and significant economic losses in severe cases. Therefore, rapid and accurate fault location and severity assessment of optical transmission systems are crucial for efficient operation and maintenance and ensuring network reliability.

[0003] Currently, fault diagnosis for optical transmission systems primarily relies on digital signal processing (DSP) technology. By analyzing the signal recovered from the DSP at the receiver, various images such as constellation diagrams and eye diagrams can be generated, and these can be combined with quantitative data such as bit error rate and signal-to-noise ratio to assess the system status. However, existing diagnostic methods have significant limitations: First, the diagnostic process is highly dependent on the prior knowledge and experience of domain experts. Engineers need to manually observe the deformation and offset characteristics of the constellation diagram and perform correlation analysis with specific fault modes. This is not only inefficient but also highly subjective, and prone to bias in judging complex or compound faults. Second, existing automated methods lack sufficient intelligence. Some studies attempt to use traditional machine learning or single deep neural networks to analyze specific types of data (such as images or numerical values), but these methods are usually "black box" models, lacking interpretability and struggling to achieve deep fusion and joint reasoning of cross-modal (image and data) information. This results in inaccurate estimation of fault severity and makes it difficult to comprehensively and systematically understand the impact of faults on system performance.

[0004] In recent years, multimodal large language models have demonstrated powerful capabilities in text and image understanding, semantic association, and logical reasoning, providing a new paradigm for the intelligent analysis of complex systems. However, directly applying them to the highly specialized field of optical transmission system fault diagnosis still faces challenges. General-purpose large models lack expertise in the optical communication domain and cannot understand the complex mapping relationships between specific fault modes and DSP image / data features. Therefore, how to effectively inject domain knowledge into the model and construct an adapted diagnostic process to achieve automated, accurate, and interpretable analysis of multimodal fault information in optical transmission systems has become a pressing technical challenge for improving intelligent operation and maintenance. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method, device, and medium for fault location and severity estimation in optical transmission systems.

[0006] A method for fault location and severity estimation in an optical transmission system according to the present invention includes the following steps: Step S1: Build an optical transmission experimental system, which includes a digital signal processing module; Step S2: Introduce a preset fault into the optical transmission experimental system and collect multimodal data generated by the digital signal processing module under the fault condition; the multimodal data includes at least image data and numerical data; Step S3: Construct a prompt word template, which includes at least experimental condition parameters, a description of the correspondence between faults and DSP data / image deformation, and standardized output response format requirements; Step S4: Based on the collected multimodal data and corresponding fault labels, the pre-trained multimodal large language model is fine-tuned using the prompt word template to obtain a fault diagnosis model; Step S5: Input the multimodal data generated by the digital signal processing module of the optical transmission system to be diagnosed into the fault diagnosis model, and output the estimation results of the fault type and severity in the optical transmission system to be diagnosed.

[0007] Preferably, step S2 includes: Step S2.1: Optimize the performance parameters of the optical transmission experimental system to bring the system to a fault-free optimal state; Step S2.2: In the optical transmission experimental system, one or more preset fault types are randomly introduced multiple times. The fault type and severity of each introduced fault are recorded as fault labels, and the experiment is run to collect the corresponding multimodal data.

[0008] Preferably, the preset fault type includes at least one of the following: phase deviation of the coherent drive modulator, four-channel power deviation of the arbitrary waveform generator, sampling deviation in the digital signal processing at the receiving end, and transmission delay difference.

[0009] Preferably, step S2 further includes: Step S2.3: Based on the acquired multimodal data, calculate the feature values ​​of the constellation diagram generated by the digital signal processing module. The feature values ​​include at least the ellipticity and rotation angle used to describe the constellation diagram changes. The feature values ​​are associated with and stored in conjunction with the multimodal data and the fault labels. The eigenvalue calculation method includes: representing the received N constellation points as complex numbers, constructing their covariance matrix, and performing eigenvalue decomposition on the covariance matrix to obtain two eigenvalues. and the corresponding eigenvectors; where ellipticity is defined as The rotation angle is the angle between the eigenvector corresponding to the principal eigenvalue and the horizontal axis.

[0010] Preferably, in step S3, the experimental condition parameters include at least the transmission distance and the modulation format.

[0011] Preferably, in step S4, a low-rank adaptive algorithm is used to fine-tune the multimodal large language model. The fine-tuning method includes: introducing a trainable low-rank decomposition matrix as a bypass to the weight matrix of the original large language model, freezing the original parameters and optimizing only the low-rank matrix, thereby significantly reducing the number of trainable parameters and reducing computational and storage overhead.

[0012] Preferably, in step S4, when fine-tuning the multimodal large language model, the input information includes: the task description generated according to the prompt word template, the multimodal data, and the constellation graph feature values.

[0013] An electronic device according to the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the optical transmission system fault location and severity estimation method.

[0014] According to the present invention, a computer-readable storage medium is provided thereon storing a computer program that, when executed by a processor, implements the steps of the optical transmission system fault location and severity estimation method.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. To address the shortcomings of existing methods that rely on human experience and are prone to bias in judging complex faults, this invention fine-tunes a multimodal large language model, deeply integrating image and numerical data generated by digital signal processing (DSP) and performing cross-modal joint inference. This model can learn and associate deep mapping relationships between complex fault modes and multimodal features, thereby achieving more accurate localization and severity estimation of single and complex faults, significantly reducing the false positive rate and improving the reliability of diagnostic results.

[0016] 2. To address the inefficiency of traditional manual analysis, this invention constructs standardized prompt word templates and diagnostic processes. The trained model can automatically receive and parse multimodal inputs generated by the DSP, replacing manual analysis of image deformation and data changes, and directly outputting structured fault diagnosis conclusions. This significantly reduces reliance on domain expert experience, shortens fault diagnosis time, and achieves an intelligent upgrade of the operation and maintenance process.

[0017] 3. To address the poor interpretability of existing "black box" automated models, this invention leverages the inherent natural language generation capabilities of large language models to output diagnostic conclusions through standardized response templates. The model not only provides the fault type and severity, but its analysis process is also more logical and traceable, making the diagnostic basis more transparent. This helps maintenance personnel understand and trust the model's judgment, facilitating subsequent decision-making and handling. Attached Figure Description

[0018] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A schematic diagram of an optical transmission system with digital signal processing added to an embodiment of the present invention; Figure 2 This is a schematic diagram of image deformation caused by different system failures in an embodiment of the present invention; Figure 3 This is a training flowchart in an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0020] This invention provides a method for fault location and severity estimation in optical transmission systems based on a fine-tuned multimodal large model, such as... Figure 3 As shown, the specific steps include: Step S1: Build an optical transmission experimental system that combines digital signal processing (DSP).

[0021] First, such as Figure 1 As shown, a typical optical transmission experimental system is built. The system mainly includes a transmitter, an optical fiber transmission link, and a receiver.

[0022] The transmitting end includes key components such as an arbitrary waveform generator (AWG), a transmissive laser (Tx Laser), and a coherent drive modulator (CDM) for generating and modulating optical signals. The optical signal is transmitted through a specific length of optical fiber to the receiving end. The receiving end includes components such as a local oscillator laser (Lo Laser), an oscilloscope, and a low-pass filter.

[0023] Both the transmitting and receiving ends are equipped with DSP modules. These DSP modules are responsible for executing a series of standard algorithms, including clock recovery, carrier phase estimation, equalization, and demapping, to recover the transmitted signal. This invention collects and analyzes the multimodal data generated by this DSP module during and after processing, including but not limited to constellation diagrams, eye diagrams, and other image data, as well as various parameters calculated in the intermediate steps.

[0024] Step S2: Introduce faults and collect multimodal data.

[0025] To ensure accurate labeled training data, faults are actively introduced and recorded in a controlled experimental environment; this includes the following sub-steps: Step S2.1: System Pre-tuning. Before introducing a fault, all parameters of the entire optical transmission system are finely tuned to ensure the system is in a fault-free, optimal performance baseline state. This is to ensure that when a fault is subsequently added, the observed performance degradation can be accurately attributed to the specific fault added, avoiding interference with the accuracy of fault labeling due to a suboptimal system state.

[0026] Step S2.2: Fault Injection and Data Acquisition. Under baseline conditions, one or more common faults in the optical transmission system are artificially and randomly introduced multiple times. Typical fault types include, but are not limited to: 1) I / Q phase deviation of the coherent drive modulator (CDM); 2) Power amplitude deviation of the four drive signals (Xi, Xq, Yi, Yq) of the arbitrary waveform generator (AWG); 3) Sampling offset in digital signal processing at the receiving end; 4) Differential group delay (transmission delay difference) introduced by the transmission link.

[0027] In each experiment, the type of fault injected and its severity (e.g., the specific degree of phase deviation, the dB value of power deviation, etc.) are precisely recorded to form a "fault label". After the system is running, the multimodal data output by the DSP stage is collected and saved, mainly the recovered signal constellation diagram (image mode) and a series of internal performance parameters (data mode).

[0028] Step S2.3: Eigenvalue calculation and integration.

[0029] To further quantify the impact of faults on signal quality and enhance the understandable features of the model, the acquired constellation diagrams were analyzed. Eigenvalues ​​for each constellation diagram were calculated, primarily including ellipticity and rotation angle. Ellipticity describes the degree to which the constellation point distribution deviates from an ideal circle, while the rotation angle describes the rotational offset of the entire constellation diagram around its center. These eigenvalues ​​can effectively and concisely describe, for example,... Figure 2The diagram illustrates the different constellation pattern changes caused by various faults (e.g., circles turning into ellipses, overall rotation, etc.). Finally, the original image data, numerical data, calculated feature values, and corresponding fault labels are correlated to form a complete training sample.

[0030] Step S3: Construct a prompt word template.

[0031] A prompt word template is constructed that fully describes the experimental conditions and task background, specifying key parameters such as the transmission distance and modulation format used in the current experiment. It also details how different types of faults correspond to image distortions and data changes caused by digital signal processing. Finally, the model is standardized to generate a fault prediction response template. In a preferred embodiment, the template includes the following parts: 1. System background description: Describe the key parameters of the current experiment or the system to be diagnosed, for example: "The current optical transmission system uses the DP-16QAM modulation format and the transmission distance is 80km." 2. Fault-Feature Mapping Knowledge: Briefly describe in text form the changes in DSP output that may be caused by different faults. For example: "When there is a phase deviation in the CDM, the constellation diagram may rotate as a whole; when there is an IQ power deviation in the AWG, the constellation diagram may be compressed or stretched on the I-axis or Q-axis, exhibiting elliptical deformation." This knowledge is provided to the model as prior information.

[0032] 3. Task Instructions and Output Specifications: Clearly define the task of the instruction model as analyzing the provided images and data, and strictly specify its output format. For example: "Please analyze the provided constellation diagram features and system data, determine the type of fault, and reply in the format of 'Fault Type: [Type]; Severity: [Level]'." Step S4: Fine-tune the multimodal large language model.

[0033] Deploy an open-source multimodal large language model on a local server as a base model and fine-tune it using the low-rank adaptive (LoRA) algorithm. The advantage of this algorithm is that it does not require a lot of computing power and can reduce training costs.

[0034] For each training sample prepared in step S2, perform the following operation: replace the background parameters in the prompt word template constructed in step S3 with the actual parameters of the current sample to form a specific task description text. Then, use this description text, the corresponding constellation image, DSP numerical data, and the calculated constellation feature values ​​(ellipticity, rotation angle) as input to the model. The training objective of the model is to make its output as consistent as possible with the real "fault labels" in the samples. Through iterative training with a large number of samples, the model can learn the complex nonlinear mapping relationship between the multimodal characteristics of the optical transmission system and the fault type and severity, ultimately obtaining a dedicated fault diagnosis model for optical transmission systems.

[0035] Step S5: Perform fault diagnosis.

[0036] For a new, unknown optical transmission system, real-time multimodal data (constellation diagram and related parameters) generated by its receiver DSP module during operation is collected, and constellation diagram eigenvalues ​​are calculated. Actual system parameters are entered into the prompt word template and input along with the real-time data into the fault diagnosis model trained in step S4. The model will automatically perform cross-modal analysis and inference, and output diagnostic results in a preset format, clearly indicating the possible fault types in the system (e.g., "Xq-channel power deviation of the AWG") and estimating their severity (e.g., "moderate").

[0037] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for fault location and severity estimation in an optical transmission system, characterized in that, include: Step S1: Build an optical transmission experimental system, which includes a digital signal processing module; Step S2: Introduce a preset fault into the optical transmission experimental system and collect multimodal data generated by the digital signal processing module under the fault condition; the multimodal data includes at least image data and numerical data; Step S3: Construct a prompt word template, which includes at least experimental condition parameters, a description of the correspondence between faults and DSP data / image deformation, and standardized output response format requirements; Step S4: Based on the collected multimodal data and corresponding fault labels, the pre-trained multimodal large language model is fine-tuned using the prompt word template to obtain a fault diagnosis model; Step S5: Input the multimodal data generated by the digital signal processing module of the optical transmission system to be diagnosed into the fault diagnosis model, and output the estimation results of the fault type and severity in the optical transmission system to be diagnosed.

2. The method for fault location and severity estimation of an optical transmission system according to claim 1, characterized in that, Step S2 includes: Step S2.1: Optimize the performance parameters of the optical transmission experimental system to bring the system to a fault-free optimal state; Step S2.2: In the optical transmission experimental system, one or more preset fault types are randomly introduced multiple times. The fault type and severity of each introduced fault are recorded as fault labels, and the experiment is run to collect the corresponding multimodal data.

3. The method for fault location and severity estimation of an optical transmission system according to claim 2, characterized in that, The preset fault types include at least one of the following: phase deviation of the coherent drive modulator, four-channel power deviation of the arbitrary waveform generator, sampling deviation in the digital signal processing at the receiver, and transmission delay difference.

4. The method for fault location and severity estimation of an optical transmission system according to claim 2 or 3, characterized in that, Step S2 further includes: Step S2.3: Based on the acquired multimodal data, calculate the feature values ​​of the constellation diagram generated by the digital signal processing module. The feature values ​​include at least the ellipticity and rotation angle used to describe the constellation diagram changes. The feature values ​​are associated with and stored in conjunction with the multimodal data and the fault labels. The eigenvalue calculation method includes: representing the received N constellation points as complex numbers, constructing their covariance matrix, and performing eigenvalue decomposition on the covariance matrix to obtain two eigenvalues. and the corresponding eigenvectors; where ellipticity is defined as The rotation angle is the angle between the eigenvector corresponding to the principal eigenvalue and the horizontal axis.

5. The method for fault location and severity estimation of an optical transmission system according to claim 1, characterized in that, In step S3, the experimental condition parameters include at least the transmission distance and modulation format.

6. The method for fault location and severity estimation of an optical transmission system according to claim 1, characterized in that, In step S4, a low-rank adaptive algorithm is used to fine-tune the multimodal large language model. The fine-tuning method includes: introducing a trainable low-rank decomposition matrix as a bypass to the weight matrix of the original large language model, freezing the original parameters and optimizing only the low-rank matrix, thereby significantly reducing the number of trainable parameters and reducing computational and storage overhead.

7. The method for fault location and severity estimation of an optical transmission system according to claim 1, characterized in that, In step S4, when fine-tuning the multimodal large language model, the input information includes: the task description generated based on the prompt word template, the multimodal data, and the constellation graph feature values.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the optical transmission system fault location and severity estimation method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the optical transmission system fault location and severity estimation method as described in any one of claims 1 to 7.

Images