Optical signal-to-noise ratio equalization model construction method, system, device and storage medium

Abstract

A method, system, device, and storage medium for constructing an optical signal-to-noise ratio (OSN) equalization model relate to the field of optical communications, and include pre-training a target machine learning model based on an initial training set to obtain an initial model, wherein the initial training set includes an expected OSN ratio and its corresponding WSS parameter set; freezing weight parameters in the initial model and reversely training the frozen model to determine an optimal WSS parameter set; configuring the optimal WSS parameter set to an actual network transmission link to determine a true OSN ratio; calculating a target mean absolute error (MAE) based on the true OSN ratio and the expected OSN ratio; and generating an OSN equalization model when the target MAE is less than or equal to a preset MAE threshold, outputting target WSS parameters that can achieve OSN equalization through the OSN equalization model. This application can effectively improve signal quality during transmission, thereby increasing the maximum capacity of system transmission.

Description

Technical Field

[0001] The present application relates to the field of optical communication technology, and in particular to a method, system, device and storage medium for constructing an optical signal-to-noise ratio equalization model. Background Art

[0002] In multi-band systems, due to stimulated Raman scattering, the power of short-wavelength channels gradually transfers to long-wavelength channels, and this power transfer accumulates continuously within the link. As this cumulative effect continues to intensify, the system's gain flatness will significantly deteriorate, severely impacting overall system performance. Therefore, to meet the rapidly growing demand for communication traffic, extending the wavelength band has become an effective way to increase channel capacity. To ensure signal quality during transmission, optical filters are often used to reduce the impact of ASE (Amplified Spontaneous Emission) on the output power of EDFA (Erbium-Doped Fiber Amplifier), thereby improving system stability and performance. However, since optical filters are optimized for a specific wavelength range, they may not match the EDFA output wavelength, resulting in poor filtering effectiveness.

[0003] Related technologies use machine learning to learn and set the attenuation of a Wavelength Selective Switch (WSS) to adjust the output power of each channel, thereby avoiding power differences between wavelengths during transmission. However, relying solely on channel power adjustment is not sufficient to ensure that signal quality across all channels reaches the desired level. Therefore, effectively improving signal quality during transmission is an urgent problem that needs to be solved. Summary of the Invention

[0004] The present application provides a method, system, device and storage medium for constructing an optical signal-to-noise ratio equalization model, which can effectively improve the signal quality during transmission.

[0005] In a first aspect, an embodiment of the present application provides a method for constructing an optical signal-to-noise ratio equalization model, comprising the following steps:

[0006] Pre-training a target machine learning model based on an initial training set to obtain an initial model, wherein the initial training set includes an expected optical signal-to-noise ratio and a corresponding WSS parameter set;

[0007] Freeze the weight parameters in the initial model and perform reverse training on the frozen model to determine the optimal WSS parameter set;

[0008] Configure the optimal WSS parameter set to the actual network transmission link to determine the true optical signal-to-noise ratio;

[0009] Calculating a target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio;

[0010] When the target mean absolute error is less than or equal to a preset mean absolute error threshold, an optical signal-to-noise ratio equalization model is generated to output target WSS parameters that can achieve optical signal-to-noise ratio equalization through the optical signal-to-noise ratio equalization model.

[0011] In combination with the first aspect, in one embodiment, after the step of calculating the target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio, the method further includes:

[0012] When the target mean absolute error is greater than the mean absolute error threshold, adding the mapping relationship between the true optical signal-to-noise ratio and the optimal WSS parameter set as training data to the initial training set to form a target training set;

[0013] The weight parameters in the frozen model are unfrozen, and the target training set is used as input for retraining to obtain a new model, and the step of freezing the weight parameters in the initial model is performed based on the new model.

[0014] In conjunction with the first aspect, in one embodiment, the expected optical signal-to-noise ratio includes a plurality of expected optical signal-to-noise ratios. After the step of generating an optical signal-to-noise ratio equalization model when the target mean absolute error is less than or equal to a preset error threshold, the method further includes:

[0015] The step of freezing the weight parameters in the initial model is performed based on the optical signal-to-noise ratio equalization model corresponding to the next expected optical signal-to-noise ratio and the previous expected optical signal-to-noise ratio until the training of all expected optical signal-to-noise ratios is completed to generate a final optical signal-to-noise ratio equalization model.

[0016] In conjunction with the first aspect, in one embodiment, the reverse training of the frozen model to determine the optimal WSS parameter set includes:

[0017] Re-initialize a target WSS parameter set corresponding to the expected optical signal-to-noise ratio randomly;

[0018] Performing reverse training on the frozen model based on the target WSS parameter set, and adjusting the target WSS parameter set based on the reverse training result;

[0019] The model is then trained backward based on the adjusted WSS parameter set until the mean absolute error between the predicted optical signal-to-noise ratio (OSNR) output by the model and the expected OSNR is minimized.

[0020] The WSS parameter set corresponding to the minimum mean absolute error is taken as the optimal WSS parameter set.

[0021] In a second aspect, an embodiment of the present application provides a system for constructing an optical signal-to-noise ratio equalization model, including:

[0022] a pre-training module for pre-training a target machine learning model based on an initial training set to obtain an initial model, wherein the initial training set includes an expected optical signal-to-noise ratio and its corresponding WSS parameter set;

[0023] The retraining module is used to freeze the weight parameters in the initial model and perform reverse training on the frozen model to determine the optimal WSS parameter set;

[0024] A determination module configured to configure the optimal WSS parameter set to the actual network transmission link to determine the true optical signal-to-noise ratio;

[0025] a calculation module, configured to calculate a target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio;

[0026] A construction module is used to generate an optical signal-to-noise ratio equalization model when the target mean absolute error is less than or equal to a preset mean absolute error threshold, so as to output a target WSS parameter that can achieve optical signal-to-noise ratio equalization through the optical signal-to-noise ratio equalization model.

[0027] In conjunction with the second aspect, in one embodiment, the construction module is further configured to, when the target mean absolute error is greater than the mean absolute error threshold, add the mapping relationship between the true optical signal-to-noise ratio and the optimal WSS parameter set as training data to the initial training set to form a target training set;

[0028] The retraining module is also used to unfreeze the weight parameters in the frozen model, and retrain with the target training set as input to obtain a new model, and perform the step of freezing the weight parameters in the initial model based on the new model.

[0029] With reference to the second aspect, in one embodiment, the expected optical signal-to-noise ratio includes a plurality of expected optical signal-to-noise ratios, and the retraining module is further configured to:

[0030] The step of freezing the weight parameters in the initial model is performed based on the optical signal-to-noise ratio equalization model corresponding to the next expected optical signal-to-noise ratio and the previous expected optical signal-to-noise ratio until the training of all expected optical signal-to-noise ratios is completed to generate a final optical signal-to-noise ratio equalization model.

[0031] In conjunction with the second aspect, in one embodiment, the retraining module is specifically configured to:

[0032] Re-initialize a target WSS parameter set corresponding to the expected optical signal-to-noise ratio randomly;

[0033] Performing reverse training on the frozen model based on the target WSS parameter set, and adjusting the target WSS parameter set based on the reverse training result;

[0034] The model is then trained backward based on the adjusted WSS parameter set until the mean absolute error between the predicted optical signal-to-noise ratio (OSNR) output by the model and the expected OSNR is minimized.

[0035] The WSS parameter set corresponding to the minimum mean absolute error is taken as the optimal WSS parameter set.

[0036] In a third aspect, an embodiment of the present application provides an optical signal-to-noise ratio equalization model construction device, which includes a processor, a memory, and an optical signal-to-noise ratio equalization model construction program stored in the memory and executable by the processor, wherein when the optical signal-to-noise ratio equalization model construction program is executed by the processor, the steps of the optical signal-to-noise ratio equalization model construction method as described above are implemented.

[0037] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which is stored a program for constructing an optical signal-to-noise ratio equalization model. When the program for constructing an optical signal-to-noise ratio equalization model is executed by a processor, the steps of the aforementioned method for constructing an optical signal-to-noise ratio equalization model are implemented.

[0038] The beneficial effects of the technical solutions provided in the embodiments of the present application include:

[0039] A target machine learning model is pre-trained using an initial training set including an expected optical signal-to-noise ratio (OSN) and its corresponding WSS parameter set to obtain an initial model. The weight parameters in the initial model are then frozen and reversely trained to determine an optimal WSS parameter set. The optimal WSS parameter set is then deployed in an actual network transmission link to determine the actual OSN ratio, and a target mean absolute error (MAE) is calculated based on the actual OSN ratio and the expected OSN ratio. When the target MAE is less than or equal to a MAE threshold, an OSN equalization model is generated, and the OSN equalization model outputs target WSS parameters that can achieve OSN equalization. Specifically, when the target WSS parameters are applied to the actual network transmission link, OSN equalization is achieved in the link, ensuring that the signal quality of all channels reaches the desired level. This effectively improves the signal quality during transmission, allowing the system to transmit data at a higher rate and over a longer distance, thereby significantly increasing the maximum transmission capacity of the system. Furthermore, the present application uses OSN as a signal quality adjustment indicator, which is more reliable and authentic than traditional power indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1This is a flow chart of an embodiment of the method for constructing an optical signal-to-noise ratio equalization model of the present application;

[0041] Figure 2 This is a schematic diagram of the structure of a multi-stage amplification and equalization device involved in the embodiment of the present application;

[0042] Figure 3 This is a schematic diagram of the structure of the machine learning model involved in the embodiment of the present application;

[0043] Figure 4 A schematic diagram of the functional modules of an embodiment of the optical signal-to-noise ratio equalization model construction system of the present application;

[0044] Figure 5 Schematic diagram of the hardware structure of the optical signal-to-noise ratio equalization model construction device involved in the embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0047] In a first aspect, an embodiment of the present application provides a method for constructing an optical signal-to-noise ratio equalization model.

[0048] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of an embodiment of the method for constructing an optical signal-to-noise ratio equalization model of this application. Figure 1 As shown, the optical signal-to-noise ratio equalization model construction method includes:

[0049] Step S10: Pre-training the target machine learning model based on the initial training set to obtain an initial model, wherein the initial training set includes the expected optical signal-to-noise ratio and its corresponding WSS parameter set.

[0050] For example, in this embodiment, Figure 2The multi-stage amplification and equalization device shown achieves optical signal-to-noise ratio (OSNR) balanced control. The multi-stage amplification and equalization device includes a wavelength selective switch (WSS) corresponding to the wavelength band, multiple sections of continuous optical fiber, multiple erbium-doped fiber amplifiers (EDFAs), and an optical spectrum analyzer (OSA) for collecting the optical power spectrum of the EDFA output. It is understood that the M EDFA amplifiers are used to compensate for signal attenuation during optical fiber link transmission. The WSS is used to control the attenuation input of each channel to ensure that the OSNR of each channel remains at a similar level. Furthermore, the first EDFA after the WSS is used to control the power input to the fiber.

[0051] It should be understood that this embodiment can roughly capture channel characteristic information by constructing an initial model (Initial_model), and then continuously adjust the model parameters through multiple online iterations to gradually improve the accuracy of the model until the model performance meets the preset OSNR target.

[0052] Among them, a smaller initial data set can be used to pre-train the target machine learning model to roughly capture the channel feature information, and then generate an initial model, that is, the initial model obtained by pre-training can roughly capture the channel feature information. It should be noted that the target machine learning model can be an artificial neural network (ANN) or a deep neural network (DNN). Of course, it can also be other neural networks and deep learning models. The specific model type can be determined according to actual needs and is not limited here. For example, the subsequent embodiments preferably use Figure 3 The ANN shown is used as the target machine learning model, and the target machine learning model is a fully connected network consisting of an input layer, a hidden layer and an output layer. The input layer length can be preferably set to N, the hidden layer length can be preferably set to L*N, and the output layer length can be preferably set to N, where N represents the number of channels and is a positive integer, and L represents the number of hidden layers and is a positive integer.

[0053] In this embodiment, the initial data set includes the expected OSNR and its corresponding WSS parameter set, which includes multiple WSS parameters, and the number of WSS parameters is the same as the number of channels. For example, if the number of channels is 51, the WSS parameter set includes 51 WSS parameters, where the WSS parameters include but are not limited to attenuation values. In addition, each WSS parameter corresponds to an expected OSNR. However, in order to achieve OSNR balance, this embodiment sets the expected OSNRs corresponding to the WSS parameters in the same WSS parameter set to be the same. In other words, although 51 WSS parameters correspond to 51 expected OSNRs, the 51 expected OSNRs are all the same, that is, one WSS parameter set corresponds to one expected OSNR.

[0054] Therefore, see Figure 3 As shown, in this embodiment, the WSS parameters in the initial training set are input to the hidden layer through the input layer for training, and the OSNR is output through the output layer until the training is completed to generate the initial model.

[0055] Step S20: Freeze the weight parameters in the initial model and perform reverse training on the frozen model to determine the optimal WSS parameter set.

[0056] For example, in this embodiment, the internal weight parameters in the initial model are frozen first, and then the frozen initial model is reversely trained to minimize the mean absolute error (MAE) between the ANN output and the expected OSNR, and the input WSS parameters are continuously updated to find the optimal input (i.e., the optimal WSS parameter set) based on the mean absolute error (MAE).

[0057] Furthermore, in one embodiment, freezing the weight parameters in the initial model and performing reverse training on the frozen model to determine the optimal WSS parameter set includes:

[0058] Re-initialize a target WSS parameter set corresponding to the expected optical signal-to-noise ratio randomly;

[0059] Performing reverse training on the frozen model based on the target WSS parameter set, and adjusting the target WSS parameter set based on the reverse training result;

[0060] The model is then trained backward based on the adjusted WSS parameter set until the mean absolute error between the predicted optical signal-to-noise ratio (OSNR) output by the model and the expected OSNR is minimized.

[0061] The WSS parameter set corresponding to the minimum mean absolute error is taken as the optimal WSS parameter set.

[0062] For example, in this embodiment, before training the initial model, its internal weight parameters are first frozen, and then a target WSS parameter set is randomly initialized as input to perform reverse training on the frozen model; then, based on the reverse training results, the WSS parameters in the target WSS parameter set are continuously updated and adjusted, and the model is continuously reversely trained according to the adjusted WSS parameter set until the mean absolute error (MAE) between the predicted OSNR output by the model and the expected OSNR is minimized; finally, the WSS parameter set that minimizes the mean absolute error (MAE) is used as the optimal input.

[0063] Step S30: configuring the optimal WSS parameter set to the actual network transmission link to determine the real optical signal-to-noise ratio.

[0064] For example, in this embodiment, after the optimal WSS parameter set is determined, the WSS parameters in the optimal WSS parameter set can be low-pass filtered respectively, and the WSS parameters after low-pass filtering can be configured to the corresponding actual network transmission link to measure the actual OSNR corresponding to the optimal WSS parameter set.

[0065] Step S40: Calculating a target mean absolute error according to the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio.

[0066] Exemplarily, in this embodiment, after the actual OSNR corresponding to the optimal WSS parameter set is measured, a target mean absolute error (MAE) is calculated based on the actual OSNR and the expected OSNR to further measure the prediction effect of the model.

[0067] Step S50: When the target mean absolute error is less than or equal to a preset mean absolute error threshold, an optical signal to noise ratio equalization model is generated to output a target WSS parameter that can achieve optical signal to noise ratio equalization through the optical signal to noise ratio equalization model.

[0068] For example, it should be noted that the specific value setting of the mean absolute error threshold can be determined according to actual needs and is not limited here. In this embodiment, if the target mean absolute error (MAE) is less than or equal to the mean absolute error threshold, it means that the prediction effect of the current model is good, so as to achieve the expected performance target, and the iteration can be stopped to complete the model training, thereby generating an optical signal-to-noise ratio equalization model; and the optical signal-to-noise ratio equalization model can output the target WSS parameter that can achieve optical signal-to-noise ratio equalization. That is, when the target WSS parameter is applied to a real network transmission link, optical signal-to-noise ratio equalization in the link can be achieved to ensure that the signal quality of all channels reaches the expected level, that is, effectively improve the signal quality during the transmission process, and then allow the system to transmit data at a higher rate and longer distance, thereby significantly improving the maximum capacity of the system transmission; in addition, the present application uses optical signal-to-noise ratio as a signal quality adjustment indicator, which is more reliable and authentic than the traditional power indicator.

[0069] Furthermore, in one embodiment, after the step of calculating the target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio, the method further includes:

[0070] When the target mean absolute error is greater than the mean absolute error threshold, adding the mapping relationship between the true optical signal-to-noise ratio and the optimal WSS parameter set as training data to the initial training set to form a target training set;

[0071] The weight parameters in the frozen model are unfrozen, and the target training set is used as input for retraining to obtain a new model, and the step of freezing the weight parameters in the initial model is performed based on the new model.

[0072] Exemplarily, in this embodiment, if the target mean absolute error (MAE) exceeds the mean absolute error threshold, indicating that the prediction effect of the model has not yet reached the expected performance target, the mapping relationship between the measured OSNR and the optimal WSS parameter set is added to the initial training set as training data to form a target training set; the weight parameters inside the model are unfrozen and re-trained online through the target training set to generate a new model (Online_model), and the new model is used to continue optimization from step S20 until the target mean absolute error (MAE) drops below the mean absolute error threshold. At this time, the iteration is stopped to complete the online training of the model to generate an online iterative model.

[0073] In this embodiment, data collection and model training can be completed in a loop iteration, so the goal can be achieved in less than 50 iterations at most; while the generation of a traditional offline model requires the collection of a large amount of input and output data to construct a training data set, and the model training can only begin after the collection of all training data is completed. If it wants to achieve an effect similar to that of the online iterative model, it requires at least hundreds of groups of data; it can be seen that the data of the training data set in this embodiment is reduced from hundreds of groups to dozens of groups, which can significantly reduce the cost of data set construction and training time cost.

[0074] In summary, this embodiment continuously adjusts model parameters during the iterative training process, continuously adds more effective data to the model, and establishes dynamic feedback until the model meets the preset performance goals; through such a setting, not only can the power drift generated during the transmission process be balanced, but also the time and training costs can be reduced.

[0075] Furthermore, in one embodiment, the expected optical signal-to-noise ratio includes a plurality of expected optical signal-to-noise ratios. After the step of generating the optical signal-to-noise ratio equalization model when the target mean absolute error is less than or equal to a preset error threshold, the step further includes:

[0076] The step of freezing the weight parameters in the initial model is performed based on the optical signal-to-noise ratio equalization model corresponding to the next expected optical signal-to-noise ratio and the previous expected optical signal-to-noise ratio until the training of all expected optical signal-to-noise ratios is completed to generate a final optical signal-to-noise ratio equalization model.

[0077] For example, in this embodiment, the online training of the model is achieved through a series of expected OSNR target values. After completing the training of the previous expected OSNR target value, the system can automatically switch to the next newly set expected OSNR target value to continue training.

[0078] Specifically, after achieving the initial expected OSNR target value training, this embodiment will perform subsequent training for a series of different expected OSNR target values ​​based on the trained model. It will be understood that once the initial expected OSNR target value is successfully achieved, the model will be reconstructed based on this, and other expected OSNR target values ​​will be more effectively trained based on the reconstructed model. That is, after completing the training for the previous expected OSNR target value, the trained model will be used as the training object for the next expected OSNR target value. In other words, the internal weight parameters of the trained model will be frozen first, and then the frozen model will be continuously optimized based on the next expected OSNR target value.

[0079] It can be seen that after completing the initial expected OSNR target, this embodiment reconstructs the model based on this to train other expected OSNR targets, thereby significantly reducing the number of training rounds. It is also shown that the optimized model can accelerate the convergence speed of the mean absolute error (MAE), effectively promote the optimization of subsequent targets, and thus demonstrate the high efficiency and good transferability of the model.

[0080] In summary, this embodiment utilizes online iterative training of a machine learning model to equalize the OSNR of multiple EDFA amplifiers and obtain a target OSNR curve that accurately reflects signal quality. Therefore, using OSNR as a signal quality adjustment indicator directly reflects the power ratio between the optical signal and noise, accurately ensuring signal quality and effectively achieving OSNR equalization of the optical signal, resulting in better signal quality. This allows the system to transmit data at higher rates and longer distances, significantly increasing the maximum transmission capacity of the system. Compared to traditional power indicators, OSNR is more reliable and authentic, and can be applied to OSNR equalization control in multiple bands, making it an effective method for increasing system transmission capacity. Furthermore, compared to traditional offline training, which requires the collection and training of large datasets, the online iterative training process provided by this embodiment significantly reduces the amount of training data and time, namely, the dataset construction cost and training time cost. Furthermore, it directly optimizes near the target, which allows for faster target achievement and effectively reduces complexity, resulting in lower cost control and good portability.

[0081] In addition, after achieving the training of one target, this embodiment can be used as a basis for training other targets, thereby achieving faster and better results. This shows that the model has good portability and can adapt to complex and dynamic OSNR targets, making it show broad potential in practical applications and is expected to be applied to full-band channel gain optimization.

[0082] In a second aspect, an embodiment of the present application further provides an optical signal-to-noise ratio equalization model construction system.

[0083] In one embodiment, referring to Figure 4 , Figure 4 This is a functional module diagram of an embodiment of the optical signal-to-noise ratio equalization model construction system of this application. Figure 4 As shown, the optical signal-to-noise ratio equalization model construction system includes:

[0084] a pre-training module for pre-training a target machine learning model based on an initial training set to obtain an initial model, wherein the initial training set includes an expected optical signal-to-noise ratio and its corresponding WSS parameter set;

[0085] The retraining module is used to freeze the weight parameters in the initial model and perform reverse training on the frozen model to determine the optimal WSS parameter set;

[0086] A determination module configured to configure the optimal WSS parameter set to the actual network transmission link to determine the true optical signal-to-noise ratio;

[0087] a calculation module, configured to calculate a target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio;

[0088] A construction module is used to generate an optical signal-to-noise ratio equalization model when the target mean absolute error is less than or equal to a preset mean absolute error threshold, so as to output a target WSS parameter that can achieve optical signal-to-noise ratio equalization through the optical signal-to-noise ratio equalization model.

[0089] Furthermore, in one embodiment, the construction module is also used to add the mapping relationship between the true optical signal-to-noise ratio and the optimal WSS parameter set as training data to the initial training set to form a target training set when the target mean absolute error is greater than the mean absolute error threshold; the retraining module is also used to unfreeze the weight parameters in the frozen model, and retrain with the target training set as input to obtain a new model, and perform the step of freezing the weight parameters in the initial model based on the new model.

[0090] Furthermore, in one embodiment, the expected optical signal-to-noise ratio includes multiple optical signal-to-noise ratios, and the retraining module is further configured to: perform the step of freezing the weight parameters in the initial model based on the optical signal-to-noise ratio equalization model corresponding to the next expected optical signal-to-noise ratio and the previous expected optical signal-to-noise ratio, until the training of all expected optical signal-to-noise ratios is completed, thereby generating a final optical signal-to-noise ratio equalization model.

[0091] Furthermore, in one embodiment, the retraining module is specifically configured to:

[0092] Re-initialize a target WSS parameter set corresponding to the expected optical signal-to-noise ratio randomly;

[0093] Performing reverse training on the frozen model based on the target WSS parameter set, and adjusting the target WSS parameter set based on the reverse training result;

[0094] The model is then trained backward based on the adjusted WSS parameter set until the mean absolute error between the predicted optical signal-to-noise ratio (OSNR) output by the model and the expected OSNR is minimized.

[0095] The WSS parameter set corresponding to the minimum mean absolute error is taken as the optimal WSS parameter set.

[0096] The functional implementation of each module in the above-mentioned optical signal-to-noise ratio equalization model construction system corresponds to each step in the above-mentioned optical signal-to-noise ratio equalization model construction method embodiment, and their functions and implementation processes are not further described here.

[0097] In a third aspect, an embodiment of the present application provides an optical signal-to-noise ratio equalization model construction device, which may be a device with data processing capabilities, such as a personal computer (PC), a laptop computer, or a server.

[0098] Reference Figure 5 , Figure 5 Schematic diagram of the hardware structure of the optical signal-to-noise ratio equalization model construction device involved in the embodiment of the present application. In the embodiment of the present application, the optical signal-to-noise ratio equalization model construction device may include a processor, a memory, a communication interface and a communication bus.

[0099] The communication bus may be of any type and is used to interconnect the processor, memory, and communication interface.

[0100] Communication interfaces include input / output (I / O) interfaces, physical interfaces, and logical interfaces, which are used to interconnect components within the optical signal-to-noise ratio (OSNR) equalization model building device, as well as interfaces used to interconnect the OSNR equalization model building device with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet, fiber, or ATM interfaces; user equipment can be displays or keyboards.

[0101] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0102] The processor may be a general-purpose processor that can call an optical signal-to-noise ratio (OSNR) equalization model construction program stored in a memory and execute the OSNR equalization model construction method provided in the embodiments of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the OSNR equalization model construction program is called can be referenced to the various embodiments of the OSNR equalization model construction method of the present application and will not be further described here.

[0103] Those skilled in the art will understand that Figure 5 The hardware structure shown in the figure does not constitute a limitation to the present application and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0104] In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium.

[0105] The optical signal-to-noise ratio equalization model construction program is stored on the readable storage medium of the present application, wherein when the optical signal-to-noise ratio equalization model construction program is executed by the processor, the steps of the optical signal-to-noise ratio equalization model construction method as described above are implemented.

[0106] The method implemented when the optical signal-to-noise ratio equalization model construction program is executed can refer to the various embodiments of the optical signal-to-noise ratio equalization model construction method of the present application, and will not be repeated here.

[0107] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0108] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit the "first", "second" and "third" to different types.

[0109] In the description of the embodiments of this application, the words "exemplary," "for example," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary," "for example," or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary," "for example," or "for example" is intended to present the relevant concepts in a concrete manner.

[0110] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0111] In some processes described in the embodiments of the present application, multiple operations or steps are included that appear in a specific order. However, it should be understood that these operations or steps may not be performed in the order in which they appear in the embodiments of the present application or may be performed in parallel. The sequence numbers of the operations are only used to distinguish between different operations, and the sequence numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations or steps may be performed in sequence or in parallel, and these operations or steps may be combined.

[0112] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device to execute the methods described in each embodiment of the present application.

[0113] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for constructing an optical signal-to-noise ratio equalization model, characterized in that: The following steps are involved: Pre-training a target machine learning model based on an initial training set to obtain an initial model, wherein the initial training set includes an expected optical signal-to-noise ratio and a corresponding WSS parameter set; Freeze the weight parameters in the initial model and perform reverse training on the frozen model to determine the optimal WSS parameter set; Configure the optimal WSS parameter set to the actual network transmission link to determine the true optical signal-to-noise ratio; Calculating a target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio; When the target mean absolute error is less than or equal to a preset mean absolute error threshold, an optical signal-to-noise ratio equalization model is generated to output target WSS parameters that can achieve optical signal-to-noise ratio equalization through the optical signal-to-noise ratio equalization model.

2. The method for constructing an optical signal-to-noise ratio equalization model according to claim 1, wherein: After the step of calculating a target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio, the method further includes: When the target mean absolute error is greater than the mean absolute error threshold, adding the mapping relationship between the true optical signal-to-noise ratio and the optimal WSS parameter set as training data to the initial training set to form a target training set; The weight parameters in the frozen model are unfrozen, and the target training set is used as input for retraining to obtain a new model, and the step of freezing the weight parameters in the initial model is performed based on the new model.

3. The method for constructing an optical signal-to-noise ratio equalization model according to claim 1, wherein: The expected optical signal-to-noise ratio includes multiple steps. After the step of generating an optical signal-to-noise ratio equalization model when the target mean absolute error is less than or equal to a preset error threshold, the method further includes: The step of freezing the weight parameters in the initial model is performed based on the optical signal-to-noise ratio equalization model corresponding to the next expected optical signal-to-noise ratio and the previous expected optical signal-to-noise ratio until the training of all expected optical signal-to-noise ratios is completed to generate a final optical signal-to-noise ratio equalization model.

4. The method for constructing an optical signal-to-noise ratio equalization model according to claim 1, wherein: The reverse training of the frozen model to determine the optimal WSS parameter set includes: Re-initialize a target WSS parameter set corresponding to the expected optical signal-to-noise ratio randomly; Performing reverse training on the frozen model based on the target WSS parameter set, and adjusting the target WSS parameter set based on the reverse training result; The model is then trained backward based on the adjusted WSS parameter set until the mean absolute error between the predicted optical signal-to-noise ratio (OSNR) output by the model and the expected OSNR is minimized. The WSS parameter set corresponding to the minimum mean absolute error is taken as the optimal WSS parameter set.

5. An optical signal-to-noise ratio equalization model construction system, characterized in that: include: a pre-training module for pre-training a target machine learning model based on an initial training set to obtain an initial model, wherein the initial training set includes an expected optical signal-to-noise ratio and its corresponding WSS parameter set; The retraining module is used to freeze the weight parameters in the initial model and perform reverse training on the frozen model to determine the optimal WSS parameter set; A determination module configured to configure the optimal WSS parameter set to the actual network transmission link to determine the true optical signal-to-noise ratio; a calculation module, configured to calculate a target mean absolute error based on the actual optical signal-to-noise ratio and the expected optical signal-to-noise ratio; A construction module is used to generate an optical signal-to-noise ratio equalization model when the target mean absolute error is less than or equal to a preset mean absolute error threshold, so as to output a target WSS parameter that can achieve optical signal-to-noise ratio equalization through the optical signal-to-noise ratio equalization model.

6. The optical signal-to-noise ratio equalization model construction system according to claim 5, wherein: The construction module is further configured to, when the target mean absolute error is greater than the mean absolute error threshold, add the mapping relationship between the true optical signal-to-noise ratio and the optimal WSS parameter set as training data to the initial training set to form a target training set; The retraining module is also used to unfreeze the weight parameters in the frozen model, and retrain with the target training set as input to obtain a new model, and perform the step of freezing the weight parameters in the initial model based on the new model.

7. The optical signal-to-noise ratio equalization model construction system according to claim 5, wherein: The expected optical signal-to-noise ratio includes a plurality of expected optical signal-to-noise ratios, and the retraining module is further used for: The step of freezing the weight parameters in the initial model is performed based on the optical signal-to-noise ratio equalization model corresponding to the next expected optical signal-to-noise ratio and the previous expected optical signal-to-noise ratio until the training of all expected optical signal-to-noise ratios is completed to generate a final optical signal-to-noise ratio equalization model.

8. The optical signal-to-noise ratio equalization model construction system according to claim 5, wherein: The retraining module is specifically used for: Re-initialize a target WSS parameter set corresponding to the expected optical signal-to-noise ratio randomly; Performing reverse training on the frozen model based on the target WSS parameter set, and adjusting the target WSS parameter set based on the reverse training result; The model is then trained backward based on the adjusted WSS parameter set until the mean absolute error between the predicted optical signal-to-noise ratio (OSNR) output by the model and the expected OSNR is minimized. The WSS parameter set corresponding to the minimum mean absolute error is taken as the optimal WSS parameter set.

9. An optical signal-to-noise ratio equalization model construction device, characterized in that: The optical signal-to-noise ratio equalization model construction device includes a processor, a memory, and an optical signal-to-noise ratio equalization model construction program stored in the memory and executable by the processor, wherein when the optical signal-to-noise ratio equalization model construction program is executed by the processor, the steps of the optical signal-to-noise ratio equalization model construction method according to any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores an optical signal-to-noise ratio equalization model construction program, wherein when the optical signal-to-noise ratio equalization model construction program is executed by a processor, the steps of the optical signal-to-noise ratio equalization model construction method according to any one of claims 1 to 4 are implemented.

Images