OTN digital twin network generation method and system based on longitudinal federated learning

Abstract

The application discloses an OTN digital twin network generation method and system based on longitudinal federal learning, exchanges data after homomorphic encryption of data of each single-domain physical network through longitudinal federal learning, constructs a single-domain training set according to a fault root cause label and related alarm information corresponding to the fault root cause label, trains a cross-domain fault root cause identification model through the single-domain training set, judges convergence according to a public model parameter and a model parameter update amount in the training process, and thus generates an OTN digital twin network according to the trained cross-domain fault root cause identification model and topology information of each single-domain physical network. The embodiment of the application meets the privacy protection demand of related data such as alarm information, user service data and the like of each single domain in a multi-domain network, and in the case that a multi-domain orchestration system is an edge server, the method of the embodiment of the application can also utilize the computing capacity of the edge device for parallel training, and improve the model training efficiency.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology applications, and in particular to a method and system for generating OTN digital twin networks based on vertical federated learning. Background Technology

[0002] Digital twins enable monitoring, analysis, prediction, diagnosis, training, and simulation, and the simulation results can be fed back to the physical object to help optimize and make decisions. Technologies related to digital twin model construction, real-time updates of the digital model's state, and simulation analysis and control decision-making based on digital twins can be collectively referred to as DT (Digital Twin) technology.

[0003] In the telecommunications field, the application of DT (Digital Transmission Technology) can enable the analysis of the entire OTN (Over-the-Network) network. The DT network layer is a model abstraction of the entire cross-domain OTN physical network. Communication between DT network elements is not limited by the physical network space, and the visibility and operability of DT network elements are not limited by the spatial or control restrictions of the physical network domains. Therefore, as the OTN DT network layer of the cross-domain OTN physical network, the functional model of the OTN DT network layer needs to have the ability to perform global analysis of the entire cross-domain OTN physical network, and sample data is collected from each domain to train the functional model of the OTN DT network layer.

[0004] Generally speaking, each single domain in the entire cross-domain OTN physical network has a large amount of training data to collect and process. In addition, each single domain has privacy protection requirements for data information related to alarms, network topology, and user services. It is not appropriate to open all of them and report them centrally. Therefore, how to enable each single domain management system to collect complete training sample data for the cross-domain fault root cause identification model and complete the training, enhance the generalization ability of the model inference, and protect the data privacy of each single domain has become a bottleneck in the modeling of the digital twin functional model of the cross-domain OTN network, and urgently needs to be solved. Summary of the Invention

[0005] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0006] This invention provides a method and system for generating OTN digital twin networks based on vertical federated learning. It can construct training sets in an encrypted manner through vertical federated learning, protect data privacy, and train cross-domain fault root cause identification models, thereby solving the modeling bottleneck of OTN digital twin networks.

[0007] In a first aspect, embodiments of the present invention provide a method for generating OTN digital twin networks based on longitudinal federated learning, applicable to any single-domain management system within an OTN multi-domain physical network system. The OTN multi-domain physical network system further includes a multi-domain orchestration system, and the multi-domain orchestration system and the single-domain management system have a cross-domain fault root cause identification model with the same structure. The method includes:

[0008] Homomorphic encryption is performed on the local root cause markers to obtain encrypted root cause markers;

[0009] Receive all encrypted alarm sample sequences corresponding to the encrypted fault root cause marker, wherein the encrypted alarm sample sequences are obtained by homomorphically encrypting the relevant alarm information by the single-domain management system corresponding to the single domain;

[0010] A single-domain training set is generated based on the encrypted fault root cause markers and the encrypted alarm sample sequences;

[0011] Train the local cross-domain fault root cause identification model based on the single-domain training set to obtain the model parameter update amount of the cross-domain fault root cause identification model;

[0012] The model parameter update is reported to the multi-domain orchestration system, which then generates an OTN digital twin network based on the model parameter update and the topology information of each single-domain management system.

[0013] Secondly, embodiments of the present invention provide a method for generating OTN digital twin networks based on longitudinal federated learning, applied to a multi-domain orchestration system in an OTN multi-domain physical network system. The OTN multi-domain physical network system further includes a single-domain management system. The multi-domain orchestration system and the single-domain management system have a cross-domain fault root cause identification model with the same structure. The method includes:

[0014] Receive the model parameter update amount generated by the single-domain management and control system, and generate an OTN digital twin network based on the model parameter update amount and the topology information of each single-domain physical network;

[0015] The model parameter update quantity is obtained by the single-domain management system training the cross-domain fault root cause identification model corresponding to the single domain based on the single-domain training set. The single-domain training set is generated by the single-domain management system based on the encrypted fault root cause marker and the encrypted alarm sample sequence corresponding to the encrypted fault root cause marker. The encrypted fault root cause marker is obtained by the single-domain management system homomorphically encrypting the fault root cause marker of the single domain. The encrypted alarm sample sequence is obtained by the single-domain management system homomorphically encrypting the relevant alarm information of the single domain.

[0016] Thirdly, embodiments of the present invention provide a single-domain management system, including at least one processor and a memory for communicatively connecting to the at least one processor; the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the OTN digital twin network generation method as described in the first aspect.

[0017] Fourthly, embodiments of the present invention provide a multi-domain orchestration system, including at least one processor and a memory for communicatively connecting to the at least one processor; the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the OTN digital twin network generation method as described in the second aspect.

[0018] Fifthly, embodiments of the present invention provide an OTN multi-domain physical network system, including the single-domain management system described in the third aspect and the multi-domain orchestration system described in the fourth aspect, wherein the multi-domain orchestration system is connected to the single-domain management system.

[0019] The OTN digital twin network generation method provided in this invention has at least the following beneficial effects: Data from each single-domain physical network is homomorphically encrypted and exchanged through vertical federated learning; a single-domain training set is constructed based on fault root cause markers and corresponding alarm information; a cross-domain fault root cause identification model is trained using the single-domain training set; convergence is determined during training based on common model parameters and model parameter update amounts; and an OTN digital twin network is generated based on the trained cross-domain fault root cause identification model and the topology information of each single-domain physical network. This invention satisfies the privacy protection requirements for alarm information, user service data, and other related data in each single domain of a multi-domain network. Furthermore, when the multi-domain orchestration system acts as an edge server in the OTN multi-domain physical network, the method of this invention can also utilize the computing power of edge devices for parallel training, improving model training efficiency.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the technical solutions of the present invention and constitute a part of the specification. They are used together with the examples of the present invention to explain the technical solutions of the present invention and do not constitute a limitation on the technical solutions of the present invention.

[0022] Figure 1 This is an overall method flowchart of an OTN digital twin network generation method executed by a single-domain management and control system according to an embodiment of the present invention;

[0023] Figure 2 This is a diagram of an OTN cross-domain physical network architecture provided in one embodiment of the present invention;

[0024] Figure 3 This is an architecture diagram of a cross-domain fault root cause identification model provided in one embodiment of the present invention;

[0025] Figure 4 This is a flowchart of a method for constructing a single-domain training set according to an embodiment of the present invention;

[0026] Figure 5 This is a flowchart of a method for constructing a single-domain training set based on fault root cause markers and related alarm information, provided by an embodiment of the present invention.

[0027] Figure 6 This is a flowchart of an iterative training process provided in an embodiment of the present invention.

[0028] Figure 7 This is a flowchart illustrating the method of a single-domain control system and a multi-domain orchestration system during a single-iteration training process according to an embodiment of the present invention.

[0029] Figure 8 This is an overall method flowchart of an OTN digital twin network generation method executed by a multi-domain orchestration system according to an embodiment of the present invention;

[0030] Figure 9 This is a flowchart of an iterative training process provided in one embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0032] The global communications industry is transitioning from the interconnected and cloud era to the intelligent era, with new opportunities and challenges driving a comprehensive transformation and upgrade of networks. Against this backdrop, the TM Forum proposed the concept of Autonomous Networks (AN) in 2019. After more than two years, the AN concept has gained industry consensus, aiming to enable the digital transformation of operator networks through the integration of network and digital technologies. This provides innovative ICT services and user experiences with zero waiting time, zero contact, and zero failures for vertical industries and consumers, and builds self-configuring, self-healing, and self-optimizing network capabilities throughout the entire lifecycle of operator network operations. Simultaneously, emerging digital twin technology is also entering a period of rapid development, widely applied in manufacturing, supply chains, and other fields to support industrial digital transformation. The digital transformation of the telecommunications sector also has a strong demand for digital twin technology, which is considered by the industry to be the foundation of digital transformation and a crucial support and component for realizing the AN autonomous network architecture and technology. Network digital twins, through precise perception of their own and external environmental states, establish digital mirrors of the internal and external environments, integrating simulation and predictive capabilities, playing a key enabling technology role in scenarios such as "low-cost trial and error," "intelligent decision-making," and "predictive maintenance."

[0033] When applied to operator communication network scenarios, under the AN self-intelligent network architecture, it is necessary to construct an OTN DT network model and generate an OTN DT network layer in the complex multi-domain OTN network environment. This allows for the analysis capability of the entire OTN network through DT technology. The DT network layer is a model abstraction of the entire cross-domain OTN physical network. Communication between DT network elements is not limited by physical network space, and the visibility and operability of DT network elements are not limited by spatial or control restrictions such as physical network domain division. Therefore, as the OTN DT network layer of the cross-domain OTN physical network, the functional model of this OTN DT network layer needs to have the ability to perform global analysis of the entire cross-domain OTN physical network, and sample data is collected from each domain to train the functional model of this OTN DT network layer. Taking the construction of a cross-domain OTN fault root cause identification functional model belonging to the DT case perception algorithm model of the cross-domain OTN network layer as an example, supervised learning is usually used to construct the cross-domain OTN fault root cause identification algorithm model. A complete training sample consists of the input part (alarms from each domain), the output fault root cause identification value of the RNN, and the root cause label provided by the domain where the fault root cause is located. In addition, the complexity of collecting training samples for the cross-domain fault root cause identification model also lies in:

[0034] a. Uncertainty in the occurrence of the root cause of the failure: the root cause of the failure of different training samples comes from different domains: it is very likely that the true root cause of the failure of the previous training sample occurs in domain A, and the root cause label is provided by domain A; while the true root cause of the failure of the next training sample occurs in domain B, and the root cause label is provided by domain B.

[0035] b. Alarms and root cause labels of the input part belonging to the same training sample come from different domains. A single domain must obtain relevant alarm information and root cause labels provided by other domains in order to obtain a complete training sample; otherwise, it is impossible to train the RNN model of this domain.

[0036] Therefore, for model training of cross-domain OTN fault root cause identification, a single OTN domain control system cannot collect complete training samples and complete training within its own domain. However, if a unified multi-domain orchestration system that coordinates the various single-domain control systems completes the unified model training, the following problems arise:

[0037] 1. The collection and processing of large amounts of training data in each single domain increases the computational load of the multi-domain orchestration system, which is inconsistent with the functional positioning of the multi-domain orchestration system;

[0038] 2. Individual domains have privacy protection requirements for data information related to alarms, network topology, and user services. It is not advisable to fully open all such data and report it centrally to the multi-domain orchestration system.

[0039] In summary, how to enable each single-domain management and control system to collect complete training sample data for the cross-domain fault root cause identification model, complete the training, enhance the generalization ability of the model's inference, and protect the data privacy of each domain has become a bottleneck in the modeling of the DTcase functional model for cross-domain OTN networks, and urgently needs to be solved.

[0040] Based on this, embodiments of the present invention provide an OTN digital twin network generation method and system, which uses vertical federated learning technology to construct training sample data for a cross-domain fault root cause identification model, thereby solving data privacy issues and improving the model's generalization ability.

[0041] Reference Figure 1 This invention provides a method for generating an OTN digital twin network. The OTN digital twin network is mapped to an OTN multi-domain physical network system, which includes multiple single-domain physical networks. The OTN multi-domain physical network system further includes a single-domain management system and a multi-domain orchestration system. The single-domain management system corresponds one-to-one with the single-domain physical network. The single-domain management system and the multi-domain orchestration system have the same cross-domain fault root cause identification model. The method executed by the single-domain management system and the method steps executed by the multi-domain orchestration system in the OTN multi-domain physical network system are described in detail below in two parts:

[0042] For any single-domain management system in an OTN multi-domain physical network system, the method includes, but is not limited to, the following steps: S110, S120, S130, S140, and S150:

[0043] Step S110: Homomorphically encrypt the local root cause marker to obtain an encrypted root cause marker;

[0044] Step S120: Receive all encrypted alarm sample sequences corresponding to the encrypted fault root cause marker. The encrypted alarm sample sequences are obtained by homomorphically encrypting the relevant alarm information by the single-domain control system of the corresponding single domain.

[0045] Step S130: Generate a single-domain training set based on the encrypted fault root cause markers and encrypted alarm sample sequences;

[0046] Step S140: Train a local cross-domain fault root cause identification model based on the single-domain training set to obtain the model parameter update amount of the cross-domain fault root cause identification model.

[0047] Step S150: The model parameter update is reported to the multi-domain orchestration system, which then generates an OTN digital twin network based on the model parameter update and the topology information of each single-domain management system.

[0048] The main architecture of vertical federated learning includes three entities: entity A, entity B, and coordinator C. In the application scenario of vertical federated learning, the samples of A and B come from a common data provider user, and the data provider sample IDs are the same, but the feature dimensions of different samples are different. The simplified process of vertical federated learning includes:

[0049] Step 1: Coordinator C creates a key pair and sends the public key to parties A and B.

[0050] Step 2: Parties A and B encrypt and exchange the intermediate results. The intermediate results are used to help calculate the gradient and loss value.

[0051] Step 3: Parties A and B calculate the encryption gradient and add additional masks respectively. Party B also calculates the encryption loss. Parties A and B then send the encrypted results to Party C.

[0052] Step 4: Coordinator C decrypts the gradient and loss information and sends the results back to parties A and B. Parties A and B remove the masking from the gradient information and update the model parameters based on this gradient information.

[0053] Based on vertical federated learning, the scheme for constructing a cross-domain DT network layer based on vertical federated learning in this embodiment of the invention is a DT modeling scheme for isomorphic cross-domain VFML. Taking the construction of a cross-domain fault root cause identification model for a cross-domain OTN network DT case as an example, the structure of its OTN multi-domain physical network is as follows: Figure 2As shown: The OTN multi-domain physical network comprises multiple single-domain physical networks. These single-domain physical networks can be built based on switching technologies from the same vendor or different vendors. Each single-domain physical network has a single-domain management and maintenance center (OMC). Each single-domain management and maintenance center is connected to the multi-domain orchestration system. The multi-domain orchestration system is responsible for outputting the trained cross-domain fault root cause identification model and performing DT modeling based on the topology information of each single-domain physical network. The ideas and features of this embodiment are as follows:

[0054] 1. Individual network element nodes within a single domain lack AI training capabilities, while individual domain management systems possess AI training and modeling capabilities. Cross-vendor multi-domain orchestration systems also possess AI training and modeling capabilities. Using a cross-domain fault diagnosis and analysis case as a training sample: the root cause of the fault is marked in domain k, while the model training infers that the relevant input alarms for this root cause are scattered across other domains. Furthermore, the root cause and related alarm information from each domain belong to this cross-domain fault analysis case. Therefore, it is necessary to collect all relevant alarms and root cause information from each domain to train the model, which aligns with and satisfies the application scenarios and training conditions required for vertical federated learning.

[0055] 2. In this solution, the multi-domain orchestration system acts as an edge server. Each single-domain management system reports the parameter updates of its cross-domain fault root cause identification model trained by its own AI algorithm to the multi-domain orchestration system via homomorphic encryption. The multi-domain orchestration system then decrypts and aggregates the reported model parameter updates from all single-domain management systems, while simultaneously updating the parameters of the cross-domain fault root cause identification common model constructed by the multi-domain orchestration system.

[0056] 3. The multi-domain orchestration system will broadcast the updated cross-domain fault root cause identification public model parameters to each single-domain management system.

[0057] 4. Each single-domain control system refreshes the cross-domain fault root cause identification model parameters of its own domain using common model parameters, and iteratively initiates the next round of model training and interaction with the multi-domain orchestration system accordingly.

[0058] 5. If a communication failure occurs between a single-domain management system and a multi-domain orchestration system, or if the node itself malfunctions and cannot upload the model parameter gradients of its own domain management system, this will not affect the multi-domain orchestration system's updating of its own common model parameters or its interaction with other single-domain management systems.

[0059] 6. This approach can select relevant OTN domains to participate in the training of longitudinal federated learning based on the batch of fault root cause training samples. For example, if the fault root cause labels in this batch are related to alarms in K OTN domains, then the model training and update for this batch occurs between the multi-domain orchestrator and these K single-domain control systems; while the next batch...

[0060] If the root cause label is related to alarms in M ​​OTN domains, then the model training and update for this batch occurs between the multi-domain orchestrator and these M single-domain control systems. The single-domain control systems participating in the training of the two batches may overlap or not, but the multi-domain orchestration system always participates in the training of the cross-domain root cause identification model in all batches, thereby ensuring that the final cross-domain root cause identification common model trained has stronger generalization ability and robustness.

[0061] 7. The multi-domain orchestration system will generate the entire network cross-domain OTN DT network layer based on the cross-domain fault root cause identification model obtained from the final training, the encrypted network topology information reported by each domain, and other functional model information.

[0062] 8. Since each domain is networked by different equipment manufacturers, each domain management system needs to encrypt the cross-domain fault root cause identification model parameter updates before reporting the update amount of its own domain model parameters; at the same time, the network topology information of each domain reported to the multi-domain orchestration system should also be encrypted as needed.

[0063] 9. When using longitudinal federated learning to train the common cross-domain fault root cause identification model, it is necessary to ensure that the RNN+Softmax algorithm model of each single-domain control system (taking RNN+Softmax as an example here) and the RNN+Softmax algorithm model of the multi-domain orchestration system that acts as an edge server during training have the same structure: including the vector attributes of the input vector of the RNN+Softmax model, the number of vector parameters, the number of layers of the RNN+Softmax model, the number of neurons in each layer, the activation functions and connection relationships between layers, the vector attributes of the output vector, the number of output vector parameters, etc., to ensure the unified training and synchronous updating of the RNN+Softmax model parameters.

[0064] Specifically, refer to Figure 3 The modeling method of this invention is as follows:

[0065] Taking a cross-domain fault root cause identification model with an RNN+softmax model structure as an example, for a certain batch of training samples, each fault root cause training sample based on the supervised learning RNN+Softmax model consists of the following two parts: Input: a sequence of cross-domain alarm samples from time 0 to t; Output: fault root cause label. In the training of the RNN+Softmax model for cross-domain fault root cause identification, it is assumed that the training of this batch of models involves K OTN domains, each domain has a fault root cause label, and each domain has relevant fault alarm information corresponding to a single fault root cause label. Let the number of fault root cause labels in domain k be n. kL Then for i kL ∈[1,n kL ]have:

[0066]

[0067] Among them, n A is a positive integer. The model input alarm sample vector at time t corresponding to each fault root cause label is composed of associated alarms in each of all K domains. Therefore, n A represents the upper limit value of the number of associated alarms per domain for splicing each input alarm sample vector. Assuming that the encrypted alarm sample sequences are all l-dimensional column vectors, then if the number of associated alarms n A provided by splicing this alarm sample vector in a certain domain is < l, the remaining other element items are filled with zeros.

[0068] is a scalar value, representing the i kL th fault root cause label in domain k. Suppose there are K OTN domains participating in this training batch, and there are m types of fault root causes in each domain. Then there are K * m values for the OTN multi-domain fault root causes, which can represent the domain where the root cause fault is located and the type of the fault root cause, that is, is the value range of.

[0069] x ikL represents the alarm sample sequence corresponding to the fault root cause label in domain k. represents the alarm sampling of the alarm sample sequence x ikL at time t, represented in vector form. This vector has a total of K * n A dimensional alarm elements, and the vector is composed of alarm elements from each domain. For example represents the associated alarm element from domain 1. Due to the zero-padding process, x ikL is also an l-dimensional vector.

[0070] Referring to Figure 3 the cross-domain fault root cause identification model shown, the input sample sequence is The fault root cause output by the softmax classification layer is y ω (x ikL ), y ω (x ikL ) represents the fault root cause inferred by the RNN based on the i kL th alarm sample sequence. The probability of various fault root causes is represented in vector form (K * m dimension). The fault root cause can be "fiber aging", "equipment undervoltage", "optical module failure", etc., and the corresponding fault alarm information can be "loss of frame LOF", "loss of signal LOS", "optical power degradation PD", etc. Among them, ω(ω1, ω2, ω3,…, ω K*m ) represents the model parameter vector of RNN + softmax, θ(θα ,θ β b) represents the parameter vector of the RNN model.

[0071] From an AI perspective, this solution treats the cross-domain fault root cause identification model in the cross-OTN domain DT case constructed using RNN technology as a logistic regression model for solving multi-classification problems. The RNN model output uses the Softmax form, resulting in:

[0072]

[0073] Among them, the RNN model is used to evaluate n kL A sequence of related alarm samples x ikL The cost function for OTN cross-domain fault root causes obtained through inference can be expressed by the following formula:

[0074]

[0075] The 1{·} operation means that the expression within the curly braces is 1 if it is true and 0 if it is false.

[0076] ω(ω1,ω2,ω2,…,ω K*m It can be represented by a K*m dimensional column vector, ω j This represents the RNN model parameters related to the root cause value j.

[0077] The objective function for training the cross-domain fault root cause identification model in the OTN DT case is:

[0078]

[0079] The gradient of the objective function J(ω) with respect to the RNN model parameters ω can be expressed as follows, and the RNN model parameters ω can then be trained using the gradient descent iterative algorithm:

[0080]

[0081]

[0082] This represents the objective function J(ω) with respect to the RNN model parameters ω. j The gradient.

[0083] Reference Figure 4 The training set for the cross-domain fault root cause identification model can be constructed through the following steps:

[0084] Step S210: Merge the various encrypted alarm sample sequences to obtain a complete encrypted alarm training sample;

[0085] Step S220: Construct a single-domain training set based on the correspondence between the encrypted fault root cause markers and the complete encrypted alarm training samples.

[0086] Based on the method of encrypting and exchanging intermediate data between A and B in vertical federated learning, the root cause label of a fault in a certain single-domain control system and the relevant alarm information of all other single-domain control systems are homomorphically encrypted and exchanged to obtain encrypted root cause labels and encrypted alarm sample sequences corresponding to the encrypted root cause labels. A single-domain training set is constructed using the two sets of encrypted information, and the single-domain training set belongs to the single domain where the root cause label is located.

[0087] Specifically, refer to Figure 5 Let domain k be the domain in which the single-domain control system with fault root cause labeling resides. For any single-domain control system other than domain k, construct a single-domain training set for its respective domain, including:

[0088] Step S221: Determine the first alarm sample vector related to the fault root cause label based on the fault root cause label of domain k;

[0089] Step S222: The first alarm sample vector is processed by homomorphic encryption to obtain the first encrypted alarm sample vector and the first encrypted alarm sample vector is sent to other single domains;

[0090] Step S223: Receive the second encrypted alarm sample vector after other single-domain homomorphic encryption processing. The second alarm sample vector corresponding to the second encrypted alarm sample vector is related to the fault root cause label.

[0091] Step S224: Merge the first encrypted alarm sample vector and the second encrypted alarm sample vector to obtain the current single-domain encrypted alarm sample sequence;

[0092] Step S225: Construct a single-domain training set based on the encrypted fault root cause markers and encrypted alarm sample sequences provided by domain k. The encrypted fault root cause markers are obtained by homomorphically encrypting the fault root cause markers by domain k.

[0093] Reference Figure 2 As shown, taking a complete encrypted training sample obtained from a single domain 1 of the OTN from vendor 1 as an example, assume that the root cause label of the failure of this sample is provided by domain k:

[0094] First, mark the root cause of the failure. The corresponding RNN model performs homomorphic encryption on the portion of the input alarm sample vector in domain 1 at time t to obtain the first encrypted alarm sample vector (in this scheme, the encryption symbol is represented by en() and the decryption symbol is represented by dec()), and sends the encrypted alarm vector to other domains.

[0095] The expression for the first encrypted alarm sample vector is as follows:

[0096]

[0097] Indicating the fault root cause marker in domain 1 The corresponding RNN model input alarm sample vector at time t, the part of the vector that does not belong to the alarm information of domain 1 is padded with zeros.

[0098] Then, obtain the root cause markers from other domains. The corresponding RNN model's input alarm sample vector at time t is homomorphically encrypted, and the portion within each domain is then merged according to the homomorphic encryption merging formula (the merging formula is f(En(m1), En(m2), ..., En(mk)) = En(f(m1, m2, ..., mk))):

[0099]

[0100] The above two steps are the sampled sample sequence from time 0 to t. Adjust the sampling time and repeat the above two steps (assuming that the encrypted input alarm sample vector corresponding to other times is...). The encrypted input alarm sample vectors at all times are statistically analyzed, and finally a complete encrypted alarm sample sequence is obtained. and the encrypted root cause markers provided by domain k

[0101] Furthermore, other domains are processed using similar steps, each obtaining a complete encrypted cross-domain fault root cause identification training sample, i.e., the second encrypted alarm sample vector; following a similar method, each domain obtains all encrypted cross-domain fault root cause identification training samples. Following the algorithmic approach of federated learning, those not participating in the group n... kL Other OTN fields that provide input alarms and root cause labels for a given sample (i.e., fields other than the K fields) will not participate in the encrypted acquisition and exchange of related training samples, nor will they participate in the current group n. kL Training and parameter updating of federated learning models for cross-domain fault root cause identification of individual samples.

[0102] Based on the aforementioned cross-domain fault root cause identification model and the constructed single-domain training set, the training process for the cross-domain fault root cause identification model in cross-domain OTNcase is proposed below.

[0103] The overall training process for the longitudinal federated learning of the RNN+Softmax model for cross-domain fault root cause identification in multi-domain orchestration systems and the corresponding RNN models of each single-domain control system is as follows:

[0104] Initialize the common model parameters ω0 of the cross-domain fault root cause identification model of the multi-domain orchestration system, and set the iteration count k = 0; collect a single-domain training set for cross-domain fault root cause identification with fault root cause labels in single domain k, which has n elements. kL n samples; using n kL The cross-domain fault root cause identification model of the multi-domain orchestration system and each single-domain control system is trained with samples. Each training iteration increments the iteration count by 1 and determines whether k exceeds the limit K. If it does not exceed the limit K, the iteration continues; otherwise, the training ends.

[0105] Reference Figure 6 The training process in step S150 above may include the following steps:

[0106] Step S310: Report the model parameter update to the multi-domain orchestration system;

[0107] Step S320: Receive the common model parameters issued by the multi-domain orchestration system. The common model parameters are obtained based on the model parameter update amount and the initial common model parameters. The initial common model parameters are issued by the multi-domain orchestration system to the single-domain control system before iterative training.

[0108] Step S330: Update the model parameters of the local cross-domain fault root cause identification model according to the public model parameters;

[0109] Step S340: Iteratively train the cross-domain fault root cause identification model based on the model parameter update amount and common model parameters until the cross-domain fault root cause identification model meets the termination condition, so that the multi-domain orchestration system can generate an OTN digital twin network based on the trained cross-domain fault root cause identification model and the topology information of each single-domain control system.

[0110] Since the single-domain control system and the multi-domain orchestration system have the same cross-domain fault root cause identification model, they can iteratively train by passing model parameters during the training process. After each training iteration, the single-domain control system calculates the gradient and sends the updated model parameters to the multi-domain orchestration system. The multi-domain orchestration system determines convergence based on the updated model parameters. If convergence has not occurred, it calculates new common model parameters based on the updated model parameters and sends the new common model parameters back to the single-domain control system. This process is repeated iteratively until the parameters converge.

[0111] Reference Figure 7 Based on the above iterative training process, the single iteration process of this embodiment of the invention is as follows:

[0112] Step S410: The single-domain management system receives the updated public model parameters;

[0113] Step S420: The single-domain control system calculates the gradient of the updated common model parameters and performs homomorphic encryption processing based on the cross-domain fault root cause identification model of its own single domain.

[0114] Step S430: Determine the model parameter update amount of the cross-domain fault root cause identification model based on the gradient calculation results and report the model parameter update amount to the multi-domain orchestration system;

[0115] In step S440, the multi-domain orchestration system updates the common model parameters according to the model parameter update amount and distributes the common model parameters to the single-domain management system.

[0116] For a single iteration: Assume the current iteration number is p, and the common model parameter of the current multi-domain orchestration system is ω. p The multi-domain orchestration system uses ω as the common model parameter. p The data is distributed to each individual domain management system, and each individual domain management system uses the common model parameter ω. p Gradient calculations are performed to obtain the updated model parameters, and these updated parameters are reported to the multi-domain orchestration system. The multi-domain orchestration system then determines the convergence condition.

[0117]

[0118] If convergence is achieved, the iteration ends; if convergence fails, the common model parameters are updated according to the following formula:

[0119]

[0120] This increments the iteration count by 1, i.e., p+1, where ω p+1 These are the common model parameters used in the (P+1)th iteration. ω p+1 The data will continue to be distributed to each single-domain management system for the P+1th iteration.

[0121] For the (P+1)th iteration, assume ω p The data has already been distributed to various single-domain management systems. Based on this, the vertical federated learning training interaction process between the multi-domain orchestration system and the single-domain management system (taking single-domain 1 as an example) in round p+1 is as follows:

[0122] First, the cost function of the single-domain 1 cross-domain fault root cause identification model is applied to ω. p Calculate the gradient and perform homomorphic encryption:

[0123]

[0124] Then, the model parameter update amount is calculated, and this model parameter update amount is homomorphically encrypted and reported to the multi-domain orchestration system by single domain 1. The model parameter update amount is calculated according to the following formula:

[0125]

[0126] Finally, the multi-domain orchestration system obtains the updated parameters of the encryption model for all single domains, including single domain 1. Then, using the formula described above for updating the common model parameters, the common model parameters ω in the (p+1)th round are updated. p+1 And distribute it to each individual domain.

[0127] Where en(g1) represents the update amount of the encryption model parameters obtained from gradient calculation. This represents the update amount of the encrypted model parameters of the single domain reported by the single-domain 1 control system to the multi-domain orchestration system in the p+1th iteration, where a is the learning rate.

[0128] Reference Figure 8 For a multi-domain orchestration system in an OTN multi-domain physical network system, the method includes, but is not limited to, the following step S510:

[0129] Step S510: Receive the model parameter update amount generated by the single-domain management system, and generate an OTN digital twin network based on the model parameter update amount and the topology information of each single-domain physical network.

[0130] Specifically, the model parameter update quantity is obtained by the single-domain control system training the cross-domain fault root cause identification model of the corresponding single domain based on the single-domain training set. The single-domain training set is generated by the single-domain control system based on the encrypted fault root cause marker and the encrypted alarm sample sequence corresponding to the encrypted fault root cause marker. The encrypted fault root cause marker is obtained by the single-domain control system homomorphically encrypting the fault root cause marker of the single domain. The encrypted alarm sample sequence is obtained by the single-domain control system homomorphically encrypting the relevant alarm information of the single domain.

[0131] Similarly, referring to steps S110 to S150 above, the multi-domain orchestration system performs iterative training based on the model parameter update amount uploaded by the single-domain control system, and generates an OTN digital twin network based on the results of the iterative training and the topology information of each single-domain physical network.

[0132] Reference Figure 9 Conversely, on the multi-domain orchestration system side, performing step S510 iterative training to generate the OTN digital twin network may specifically include the following steps:

[0133] Step S511: Generate common model parameters based on the model parameter update amount and the initial common model parameters, and distribute the common model parameters. The initial common model parameters are distributed by the multi-domain orchestration system to the single-domain control system before iterative training.

[0134] Step S512: Iteratively train the cross-domain fault root cause identification model based on the model parameter update amount and common model parameters until the cross-domain fault root cause identification model meets the termination condition.

[0135] Step S513: Generate an OTN digital twin network based on the trained cross-domain fault root cause identification model and the topology information of each single-domain control system.

[0136] The termination conditions for iterative training of the multi-domain orchestration system are the same as those for iterative training of the single-domain control system, and will not be repeated here.

[0137] The above scheme uses vertical federated learning technology to homomorphically encrypt data from each single-domain physical network. A single-domain training set is constructed based on fault root cause markers and corresponding alarm information. This training set is then used to train a cross-domain fault root cause identification model. During training, convergence is determined by the common model parameters and the amount of model parameter updates. Finally, an OTN digital twin network is generated based on the trained cross-domain fault root cause identification model and the topology information of each single-domain physical network. This invention satisfies the privacy protection requirements for alarm information, user service data, and other related data in each single domain of a multi-domain network. Furthermore, when the multi-domain orchestration system acts as an edge server in the OTN multi-domain physical network, the method of this invention can leverage the computing power of edge devices for parallel training, improving model training efficiency.

[0138] It is worth noting that the solutions in this embodiment of the invention can be applied not only to OTN networks, but also to other homogeneous networks, such as PTN, POTN, and IP networks. Although the above cross-domain fault identification models are all illustrated using an RNN+softmax structure as an example, it is clear that in the field of neural network algorithms, there are many variations or alternative algorithms for RNNs. Besides softmax, the classification layer can have other types of classification algorithms, which will not be listed here. Those skilled in the art can select appropriate algorithms to construct cross-domain fault identification models according to actual conditions.

[0139] This invention also provides a single-domain management system, including at least one processor and a memory for communicatively connecting to the at least one processor; the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the aforementioned OTN digital twin network generation method on one side of the single-domain management system.

[0140] This invention also provides a multi-domain orchestration system, including at least one processor and a memory for communicatively connecting to the at least one processor; the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the aforementioned OTN digital twin network generation method on one side of the multi-domain orchestration system.

[0141] This invention also provides an OTN multi-domain physical network system, including a single-domain management system and a multi-domain orchestration system that execute the aforementioned OTN digital twin network generation method. The single-domain management system and the multi-domain orchestration system are connected to achieve data interaction.

[0142] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0143] The above is a detailed description of the preferred embodiments of this application. However, this application is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for generating OTN digital twin networks based on longitudinal federated learning, applied to any single-domain management system in an OTN multi-domain physical network system, wherein the OTN multi-domain physical network system further includes a multi-domain orchestration system, and the multi-domain orchestration system and the single-domain management system have a cross-domain fault root cause identification model with the same structure; the method includes: Homomorphic encryption is performed on the local root cause markers to obtain encrypted root cause markers; Receive all encrypted alarm sample sequences corresponding to the encrypted fault root cause marker, wherein the encrypted alarm sample sequences are obtained by homomorphically encrypting the relevant alarm information by the single-domain management system corresponding to the single domain; A single-domain training set is generated based on the encrypted fault root cause markers and the encrypted alarm sample sequences; Train the local cross-domain fault root cause identification model based on the single-domain training set to obtain the model parameter update amount of the cross-domain fault root cause identification model; The model parameter update is reported to the multi-domain orchestration system, which then generates an OTN digital twin network based on the model parameter update and the topology information of each single-domain management system.

2. The OTN digital twin network generation method according to claim 1, characterized in that, Each encrypted alarm sample sequence is an l-dimensional column vector, and the remaining elements in the encrypted alarm sample sequence, except for the elements related to the alarm information, are padded with zeros.

3. The OTN digital twin network generation method according to claim 2, characterized in that, The step of generating a single-domain training set based on the encrypted fault root cause markers and the encrypted alarm sample sequences includes: Merging the various encrypted alarm sample sequences yields a complete encrypted alarm training sample; A single-domain training set is constructed based on the correspondence between the encrypted fault root cause markers and the complete encrypted alarm training samples.

4. The OTN digital twin network generation method according to claim 1, characterized in that, The step of reporting the model parameter update to the multi-domain orchestration system, so that the multi-domain orchestration system can generate an OTN digital twin network based on the model parameter update and the topology information of each of the single-domain management systems, includes: The updated model parameters are reported to the multi-domain orchestration system. The system receives common model parameters issued by the multi-domain orchestration system. These common model parameters are obtained based on the model parameter update amount and the initial common model parameters. The initial common model parameters are issued by the multi-domain orchestration system to the single-domain management system before iterative training. Update the model parameters of the local cross-domain fault root cause identification model based on the public model parameters; The cross-domain fault root cause identification model is iteratively trained based on the model parameter update amount and the common model parameters until the cross-domain fault root cause identification model meets the termination condition, so that the multi-domain orchestration system generates an OTN digital twin network based on the trained cross-domain fault root cause identification model and the topology information of each single-domain control system.

5. The OTN digital twin network generation method according to claim 4, characterized in that, The termination condition for iterative training of the cross-domain fault root cause identification model is: Where K represents a homomorphic encryption operator with K single fields, en() represents the homomorphic encryption operator, and dec() represents the public-key decryption operator for homomorphic encryption. This represents the amount of model parameter updates for the k-th single domain in the p-th iteration.

6. The OTN digital twin network generation method according to claim 5, characterized in that, If the cross-domain fault root cause identification model fails to meet the termination condition during iterative training, the common model parameters are updated as follows: in These are the common model parameters used in the (p+1)th iteration.

7. The OTN digital twin network generation method according to claim 1, characterized in that, Before reporting the aforementioned topology information, the following is also included: The topology information is encrypted.

8. A method for generating OTN digital twin networks based on longitudinal federated learning, applied to a multi-domain orchestration system in an OTN multi-domain physical network system, wherein the OTN multi-domain physical network system further includes a single-domain management system, and the multi-domain orchestration system and the single-domain management system have a cross-domain fault root cause identification model with the same structure; the method includes: Receive the model parameter update amount generated by the single-domain management system, and generate an OTN digital twin network based on the model parameter update amount and the topology information of each single-domain management system; The model parameter update quantity is obtained by the single-domain management system training the cross-domain fault root cause identification model corresponding to the single domain based on the single-domain training set. The single-domain training set is generated by the single-domain management system based on the encrypted fault root cause marker and the encrypted alarm sample sequence corresponding to the encrypted fault root cause marker. The encrypted fault root cause marker is obtained by the single-domain management system homomorphically encrypting the fault root cause marker of the single domain. The encrypted alarm sample sequence is obtained by the single-domain management system homomorphically encrypting the relevant alarm information of the single domain. The step of generating an OTN digital twin network based on the model parameter update amount and the topology information of each of the single-domain management and control systems includes: Based on the model parameter update amount and the initial common model parameters, common model parameters are generated and distributed. The initial common model parameters are distributed by the multi-domain orchestration system to the single-domain management system before iterative training. The cross-domain fault root cause identification model is iteratively trained based on the model parameter update amount and the common model parameters until the cross-domain fault root cause identification model meets the termination condition. An OTN digital twin network is generated based on the trained cross-domain fault root cause identification model and the topology information of each of the single-domain management and control systems.

9. The OTN digital twin network generation method according to claim 8, characterized in that, The termination condition for iterative training of the cross-domain fault root cause identification model is: Where K represents a homomorphic encryption operator with K single fields, en() represents the homomorphic encryption operator, and dec() represents the public-key decryption operator for homomorphic encryption. This represents the amount of model parameter updates for the k-th single domain in the p-th iteration.

10. The OTN digital twin network generation method according to claim 9, characterized in that, If the cross-domain fault root cause identification model fails to meet the termination condition during iterative training, the common model parameters are updated as follows: in These are the common model parameters used in the (p+1)th iteration.

11. The OTN digital twin network generation method according to claim 8, characterized in that, The cross-domain fault root cause identification model consists of multiple RNN units and a softmax classification layer.

12. A single-domain control system, characterized in that, It includes at least one processor and a memory for communicatively connecting to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the OTN digital twin network generation method as described in any one of claims 1 to 7.

13. A multi-domain orchestration system, characterized in that, It includes at least one processor and a memory for communicatively connecting with said at least one processor; The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the OTN digital twin network generation method as described in any one of claims 8 to 11.

14. An OTN multi-domain physical network system, characterized in that, It includes the single-domain management system as described in claim 12 and the multi-domain orchestration system as described in claim 13, wherein the multi-domain orchestration system is connected to the single-domain management system.

Images