CN112560204B - Optical network route optimization method based on LSTM deep learning and related device thereof - Google Patents

Abstract

One or more embodiments of the present specification provide an LSTM deep learning-based optical network route optimization method and a related apparatus thereof, which apply an LSTM prediction model in deep learning, learn a mapping relationship between link utilization and service remaining time characteristics in different scenes and a routing policy, take link and service overall information in a network as input, and rapidly obtain a reconstruction threshold through calculation of several layers of neural networks to decide whether to perform a routing optimization policy. The method solves the defects that the algorithm in the prior art is generally limited in feature learning capacity, poor in expression capacity of complex functions, single in optimization target parameters, limited in switch resources and huge in network resource overhead, and meanwhile, the threshold value can be changed in real time and efficiently along with the continuous arrival of the service, so that the condition of service blocking caused by network flow fluctuation or excessive bearing service volume is relieved, and the adaptive routing optimization of the optical network service is realized.

Description

A method of optical network routing optimization based on LSTM deep learning and related devices

technical field

One or more embodiments of this specification relate to the technical field of routing optimization, and in particular to a method for optimizing routing of an optical network based on LSTM deep learning and related devices.

Background technique

Traditional optical network routing optimization methods include rerouting and spectrum shifting. Rerouting means that some services on the non-optimal path are re-routed to select a new optimal path for transmission to improve the overall performance of the network. When the network carries services for transmission, many services are not allocated on the shortest or optimal path because the resource requirements of the services are not met. When the service request on the shortest path is removed, the service on other non-optimal paths can be transferred to the shortest or optimal path through rerouting, thereby optimizing the use of network resources and improving the overall performance of the network. Spectrum relocation refers to finding whether there are idle low-order frequency slot resources under the original path of the service. If there is, the service will be directly moved to the free frequency slot. If not, the service will not be operated. Due to the frequent establishment and release of service connection requests, the fragmentation of spectrum resources is very high. Spectrum relocation for services can integrate spectrum resources, so that more services can be deployed to achieve the purpose of network optimization.

In recent years, some research results of optical network routing optimization methods combined with deep learning methods have begun to emerge, but the proposed optical network routing optimization algorithms based on deep learning have some defects: First, only traditional machine learning or shallow Layer artificial neural network algorithms, these traditional algorithms usually have problems such as limited feature learning ability and poor expression ability for complex functions. Second, most of the current solutions are to predict a certain parameter in the network, and then achieve the purpose of indirect optimization of service routing based on these network parameters. However, network service routing optimization is a complicated problem, and optimization for a single target parameter is usually It is difficult to achieve significant route optimization performance gains. Third, it is necessary to deploy a machine learning algorithm in each switch, and then the switch collects this information for routing calculation. However, in actual scenarios, the resources of the switch are very limited, and it is difficult to meet the computing resource requirements of the machine learning algorithm. Fourth, both need to use the traffic matrix as the input of the deep learning algorithm, but accurate traffic matrix measurement in real scenarios requires huge network resource overhead.

Contents of the invention

In view of this, the purpose of one or more embodiments of this specification is to propose an optical network routing optimization method based on LSTM deep learning and related devices to solve the problem of limited feature learning ability of algorithms in the prior art, and complex functions The problem of poor expression ability, single optimization target parameter, limited switch resources and huge network resource overhead.

Based on the above purpose, one or more embodiments of this specification provide an optical network routing optimization method and related devices based on LSTM deep learning, wherein the method includes:

receiving the request of the service to be deployed, analyzing the attribute of the service to be deployed, and calculating the shortest path according to the attribute of the service to be deployed;

judging whether the frequency spectrum on the shortest path is sufficient;

If not enough, obtain the remaining time of the deployed service on the congested path that satisfies the amount of spectrum required by the service to be deployed, calculate the time difference between the duration of the service to be deployed and the remaining time of the deployed service, and The to-be-deployed service finds a suboptimal path for deployment;

Collect the overall path spectrum occupancy data in the optical network, input the collected data into the trained LSTM deep learning model, and obtain the preset reconstruction threshold;

Judging whether the time difference reaches the preset reconstruction threshold, and assigning paths according to the judging result.

After the judging whether the frequency spectrum on the shortest path is sufficient, includes:

If it is sufficient, select and allocate to the wavelength spectrum of the shortest path of the service to be deployed, update the spectrum status information on the shortest path, and end the allocation.

The judging whether the time difference reaches the preset reconstruction threshold, and assigning according to the judging result includes:

If the time difference is greater than the preset reconfiguration threshold, the service to be deployed will be reconfigured after passing the time difference, select and allocate to the wavelength spectrum of the shortest path of the service to be deployed, and update the Spectrum status information and end allocation, if the time difference is less than or equal to the preset reconstruction threshold, end allocation.

The training steps of the LSTM deep learning model include:

Obtain historical network spectrum occupancy information and service information in the network;

According to the historical network spectrum occupancy status information and service information, generate a spectrum matrix S and a service matrix R according to time series, including obtaining corresponding spectrum matrices S ₁ , S ₂ under time series t ₁ , t ₂ to t _n to S _n and service matrices R ₁ , R ₂ to R _n ;

The spectrum matrix S _i and the service matrix R _i collected according to the current t _i are used as input data to input the initial LSTM prediction model, and t _i represents any of the time series t ₁ , t ₂ to t _n At a moment, use the spectrum matrix S _i+1 and the service matrix R _i+1 at the next moment t _{i +1} of t i as label data to train the selection of the preset reconstruction threshold;

According to the spectrum matrix and the service matrix input each time t _i , the predicted reconstruction threshold is obtained as an output for simulation, and the predicted reconstruction threshold is calculated according to the network traffic congestion rate obtained by using the predicted reconstruction threshold. Scoring, and using the predicted reconstruction threshold obtained at the time t _i as input information, obtaining the predicted reconstruction threshold at the time t _i+1 for simulation, scoring, and inputting until the end;

Repeat the above four steps to repeatedly train the LSTM model. The LSTM model will select different predicted reconstruction thresholds for simulation and score according to the traffic blocking rate in the network, and select the predicted reconstruction threshold with the highest score. The reconstruction threshold is used as the preset reconstruction threshold.

The service attributes include source node, sink node, service start time, service duration and required frequency spectrum;

The algorithm for calculating the shortest path of a service is the Dijkstra algorithm.

The method used for selecting and allocating the wavelength spectrum of the shortest path of the service to be deployed is the First-Fit algorithm.

The form of the spectrum matrix S is:

表示网络中路由节点x_i与x_j之间链路的频谱占用状况，若节点x_i与x_j不直连，那么

为0，若节点x_i与x_j直连且频谱占用率为0，那么

为1。Wherein the spectrum matrix S in

Indicates the spectrum occupancy status of the link between routing nodes x _i and x _j in the network. If nodes x _i and x _j are not directly connected, then

is 0, if node x _i is directly connected to x _j and the spectrum occupancy rate is 0, then

is 1.

The form of the business matrix R is:

分别代表业务R_i的源节点和宿节点；

代表业务R_i所需的频谱资源；

代表业务R_i的剩余时间；P_i代表业务R_i的工作路径，所述业务矩阵随着业务的到来和离去随时更新。Wherein the business matrix R in

represent the source node and the sink node of the service R _i respectively;

Represents the spectrum resource required by service R _i ;

Represents the remaining time of service R _i ; P _i represents the working path of service R _i , and the service matrix is updated at any time with the arrival and departure of services.

Based on the same inventive concept, one or more embodiments of this specification also provide an electronic device, including a memory, a processor, and a computer program stored on the memory and operable on the processor, and the processor executes the program When implementing any of the methods described above.

Based on the same inventive concept, one or more embodiments of this specification also provide a non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to make the A computer executes any of the methods described above.

It can be seen from the above that one or more embodiments of the present invention provide an optical network routing optimization method based on LSTM deep learning and its related devices, by learning the link utilization rate and service remaining time characteristics in different scenarios The complex mapping relationship with the routing strategy takes the overall information of links and services in the network as input, and the reconstruction threshold can be quickly obtained through several layers of neural network calculations to decide whether to implement a routing optimization strategy. This method solves the problems in the prior art. Algorithms usually have the defects of limited feature learning ability, poor expression ability for complex functions, single optimization target parameter, limited switch resources and huge network resource overhead. In the case of service congestion due to network traffic fluctuations or excessive bearer traffic, it is possible to implement adaptive routing optimization for optical network services.

Description of drawings

In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or prior art. Obviously, in the following description The accompanying drawings are only one or more embodiments of the present invention, and those skilled in the art can also obtain other drawings according to these drawings without creative work.

Fig. 1 is a flowchart of an optical network routing optimization method according to one or more embodiments of the present specification;

Fig. 2 is a flowchart of specific steps of an optical network routing optimization method according to one or more embodiments of the present specification;

FIG. 3 is an optical network topology diagram of one or more embodiments of this specification;

FIG. 4 is a network virtual topology diagram of one or more embodiments of this specification;

Fig. 5 is a schematic structural diagram of an electronic device according to one or more embodiments of the present specification.

detailed description

In order to make the purpose, technical solutions and advantages of the present disclosure clearer, the present disclosure will be further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present specification shall have ordinary meanings understood by those skilled in the art to which the present disclosure belongs.

As mentioned in the background technology section, the routing optimization methods in the prior art still have the problems of limited feature learning ability, poor expression ability for complex functions, single optimization target parameters, limited switch resources and huge network resource overhead, and it is difficult to meet the requirements of the network. It is required for route optimization when traffic fluctuates or bears too much traffic. During the process of implementing the present disclosure, the applicant found that rerouting services can alleviate network congestion, but the probability and frequency of rerouting must be fully considered, that is, whether routing reconfiguration is necessary. After some services are processed and resources are released, idle spectrum resources are left on the path, and services on non-shortest or optimal paths in the network need a reconfiguration operation probability to determine whether reconfiguration is required. The selection of the reconfiguration frequency is one of the key considerations, which needs to be dynamically set according to the specific performance of the network. If the frequency is too low, too few reconfiguration times will have little effect, and the network performance will not be improved, and the expected effect will not be achieved; if the frequency is too high, the network will frequently perform service reconfiguration, which will increase the complexity of organization and aggravate network traffic. The computing load will also cause the new incoming business to fail to choose the shortest path, which may cause more low-latency demand business to be blocked.

In view of this, one or more embodiments of this specification provide an optical network routing optimization scheme based on LSTM deep learning, applying the LSTM prediction model in deep learning, by learning the link usage rate and service remaining time in different scenarios The complex mapping relationship between features and routing strategies takes the overall information of links and services in the network as input, and quickly obtains the reconstruction threshold through several layers of neural network calculations to decide whether to implement routing optimization strategies. This solution solves the defects of limited feature learning ability, poor expression ability for complex functions, single optimization target parameters, limited switch resources and huge network resource overhead in the existing algorithms. It can be changed in real time and efficiently, thereby alleviating the situation of service congestion caused by network traffic fluctuations or carrying too much business, and realizing the self-adaptive routing optimization of optical network services.

Hereinafter, the technical solution of one or more embodiments in this specification will be described in detail through specific embodiments.

First, one or more embodiments of this specification provide a method for optimizing optical network routing based on LSTM deep learning. Referring to FIG. 1, the method includes the following steps:

101: Receive a request of a service to be deployed, analyze the attributes of the service to be deployed, and calculate the shortest path according to the attributes of the service to be deployed.

102: Determine whether the frequency spectrum on the shortest path is sufficient.

103: If not enough, obtain the remaining time of the deployed service on the congested path that satisfies the amount of spectrum required by the service to be deployed, and calculate the time difference between the duration of the service to be deployed and the remaining time of the deployed service, And find a suboptimal path for the service to be deployed for deployment.

104: Collect the spectrum occupancy data of the overall path in the optical network, input the collected data into the trained LSTM deep learning model, and obtain a preset reconstruction threshold.

105: Determine whether the time difference reaches the preset reconstruction threshold, and allocate paths according to the determination result.

It can be seen that through the above method, the reconfiguration threshold can dynamically optimize the optical network in real time and efficiently with the continuous arrival of services, thereby alleviating service congestion caused by network traffic fluctuations or excessive traffic loads.

In one or more embodiments of this specification, an LSTM deep learning model for predicting a preset reconstruction threshold is applied. The LSTM deep learning model needs to be trained to realize the function of predicting the preset reconstruction threshold. Correspondingly, the training method steps of the LSTM prediction model are as follows:

Step 1: Collect historical spectrum occupancy information and service information in the network;

Step 2: Generate the spectrum matrix S and service matrix R according to the time series from the collected historical network spectrum occupancy data and business information, including the time series t ₁ , t ₂ to t _n , and obtain the corresponding spectrum matrices S ₁ , S ₂ to S _n and service matrices R ₁ , R ₂ to R _n ;

Among them, the network routing nodes are x ₁ , x ₂ to x _n in turn; the link spectrum occupancy matrix S at the current moment:

表示网络中路由节点x_i与x_j之间链路的频谱占用状况，若节点x_i与x_j不直连，那么

为0，若节点x_i与x_j直连且频谱占用率为0，那么

为1；已部署业务矩阵R：any of them

Indicates the spectrum occupancy status of the link between routing nodes x _i and x _j in the network. If nodes x _i and x _j are not directly connected, then

is 0, if node x _i is directly connected to x _j and the spectrum occupancy rate is 0, then

is 1; deployed business matrix R:

分别代表业务R_i的源节点和宿节点；

代表业务R_i所需的频谱资源；

代表业务R_i的剩余时间；P_i代表业务R_i的工作路径。业务矩阵随着业务的到来和离去随时更新。Among them, in the business matrix R

represent the source node and the sink node of the service R _i respectively;

Represents the spectrum resource required by service R _i ;

Represents the remaining time of service R _i ; P _i represents the working path of service R _i . The business matrix is updated at any time with the arrival and departure of business.

Step 3: Use the historical spectrum matrix S _i and business matrix R _i collected at current t _i as the input data into the LSTM prediction model, t _i represents any moment in the time series t ₁ , t ₂ to t _n , Use the spectrum matrix S _i+1 and service matrix R _i+1 at the next time t _{i +1} of t i as label data to train the selection of the reconstruction time threshold.

Step 4: According to the spectrum matrix and service matrix input each time t _i , simulate the output predicted reconstruction threshold, and score the predicted reconstruction threshold according to the network traffic congestion rate obtained by using the predicted reconstruction threshold. The predicted reconstruction threshold obtained at time t _i is used as input information, and the predicted reconstruction threshold obtained at time t _i+1 is simulated, scored, and input until the end.

Step 5: Repeat steps 1 to 4 to continuously train the LSTM model repeatedly. Although the historical data input each time is the same, the LSTM model will select different prediction reconstruction thresholds for simulation and score according to the traffic congestion rate in the network , repeated tests to select the prediction reconstruction threshold with the highest score, so that the selected reconstruction threshold makes the blocking rate of the current network as small as possible, and at the same time makes the gap between the input data and the label data at the next moment predicted by the prediction network as small as possible. In this way, the LSTM model can obtain the optimal preset prediction result at the next moment according to the input spectrum matrix and business matrix information at the current moment.

As shown in Figure 2, it is a flow chart of specific steps of an optical network routing optimization method according to an embodiment of the present invention, including the following steps:

201: The service request arrives, and the optimal path is calculated. Dynamic arrival of optical network services, analysis of service attributes, such as source node, sink node, service start time, service duration, number of spectrum required, etc.; at the same time, use Dijkstra algorithm to calculate the optimal service path according to the source and sink nodes;

202: Determine whether the link frequency spectrum is sufficient. Judging whether the spectrum resource on each link is sufficient along the optimal path can be divided into two situations according to the judgment result:

If the frequency spectrum on the optimal path link is sufficient, skip the following steps and go to step 208;

If the frequency spectrum on a certain link of the optimal path is insufficient, perform step 203;

203: Obtain a time difference. Obtain the service set of the congested road section on the optimal path, filter out the deployed services that meet the number of spectrum required by the service to be deployed, then select the service with the smallest remaining time, and obtain its remaining time T _end ; calculate the duration T _hold of the service to be deployed time difference T _r with T _end ;

204: Deploy the service to be reconfigured. Construct a virtual topology, disconnect the congested link with insufficient spectrum resources in the virtual topology, and then use the Dijkstra algorithm to calculate the path for the service to be reconfigured, and deploy the service on the suboptimal path;

205: Use LSTM deep learning model. Firstly, the overall link spectrum data in the current network is collected, and the link spectrum occupancy matrix S and the deployed service matrix R in the network at the current moment are used as input data. Then input the matrix S and R into the pre-trained LSTM deep learning model to obtain the preset reconstruction time threshold T _TH at the next moment;

206: Determine whether to perform service reconstruction. Judging the time difference T _r and the reconstruction time threshold T _TH at the next moment, there are two cases according to the judgment result:

If the time difference T _r does not meet the reconstruction time threshold T _TH , the process ends;

If the time difference T _r satisfies the reconstruction time threshold T _TH , go to step 207;

207: Business reconstruction. After the service to be reconfigured passes through the remaining time T _end of the deployed service screened in step 203 on the suboptimal path, the screened deployed service leaves, the link spectrum resource status information is updated, and the service is reconfigured, and step 208 is performed; first After the screened deployed services leave, the status information of link spectrum resources is updated, and then the First-Fit algorithm is used to select and allocate spectrum resources on the optimal path to realize the deployment of services to be reconfigured;

208: Service deployment. Use the First-Fit algorithm to reasonably select and allocate link spectrum resources for services, update the spectrum status information on the path, and the process ends.

Based on the above method, one or more embodiments of this specification also provide a 6-node, 8-link topology network optimization method, including:

As shown in Figure 3, it is an optical network topology diagram of an embodiment of the present invention, and the specific steps of its optimization method are:

First, the deep learning model is trained under the topology network of 6 nodes and 8 links in Figure 3:

Step 1: Collect historical spectrum occupancy status information and service information in the network. The parameters include the link spectrum occupancy rate at each moment; the source node, sink node, required spectrum resource, remaining time and working path of the deployed service.

Step 2: Obtain the historical spectrum matrix and service matrix, and generate a 6×6 link spectrum occupancy matrix S _i at time t _i , t _i+1 , ..., t _i+n :

And the deployed business matrix R _i :

Step 3: At this time, the original data used for training the model is obtained, that is, the link spectrum occupancy matrix S _i , S _i+1 , ..., under the time series t _i , t _i+1 , ..., t _i+n S _i+n and deployed service matrices R _i , R _i+1 , . . . , R _i+n . Feed raw data into LSTM.

分别进行模拟，根据业务阻塞率进行打分。Step 4: Assuming the kth training, the reconstruction threshold obtained at time t _i , t _i+1 , ..., t _i+n is

Simulate separately and score according to the service blocking rate.

Step 5: Repeat steps 1-4 to continuously train the LSTM model repeatedly, so that the selected reconstruction threshold makes the blocking rate of the current network at the next moment as small as possible. Make the LSTM model obtain the optimal preset prediction result at the next moment according to the input spectrum matrix and business matrix information at the current moment, and the LSTM model training is completed at this time.

After the model pre-training is completed, follow the steps below to complete the request after the business arrives:

Step 1: The service request r arrives, analyze the request attributes, assuming that the service is from node 1 to node 6, the service duration is T _hold , the number of spectrums required is N _s , and the optimal path is calculated by Dijkstra algorithm as 1-2- 4-6;

Step 2: Determine whether the number of spectrums on path 1-2-4-6 satisfies the number N _s of spectrums required by service r. Assume that link 2-4 has insufficient spectrum resources to deploy service r because of its large traffic volume. Determine whether service r needs to be reconfigured; if the number of spectrums on 1-2-4-6 meets the number N _s of spectrums required by service r, directly perform step 8;

和

假设

因此计算得出时间差

Step 3: Since the link 2-4 is congested, at this time, filter out the service that satisfies the spectrum number N _s required by the service r on the link 2-4, assuming that there are r _a and r _b on the link 2-4 The spectrum number of a business is greater than N _s , and the remaining time is

and

suppose

Therefore, the calculated time difference

Step 4: As shown in Figure 4, it is the network virtual topology diagram of this embodiment, construct a virtual topology, disconnect the congested link with insufficient spectrum resources in the virtual topology, and use the Dijkstra algorithm to calculate the path for the service r, then the most The optimal path is 1-2-3-5-6, and this path can successfully deploy service r after judgment;

Step 5: First collect the overall link spectrum data in the current network, obtain the link spectrum occupancy matrix and the deployed service matrix in the network at the current moment as input data, and then input the matrix S and R into the pre-trained LSTM depth Learning the model to obtain a preset reconstruction time threshold T _TH at the next moment;

Step 6: Determine the size of the time difference T _r and the preset reconstruction time threshold T _TH , if T _r >T _TH , execute step 7, if T _r <T _TH , the process ends;

后，链路2-4上的业务r_a离去，释放频谱资源，业务r进行业务重构，在次优路径1-2-3-5-6上搬移到最优路径1-2-4-6，执行步骤8；Step 7: Business r passes time on path 1-2-3-5-6

Afterwards, service r _a on link 2-4 leaves, releases spectrum resources, service r performs service reconstruction, and moves to the optimal path 1-2-4 on the suboptimal path 1-2-3-5-6 -6, execute step 8;

Step 8: Use the First-Fit algorithm to reasonably select and allocate link spectrum resources for service r, update the spectrum status information on the path, and the process ends.

Based on the same inventive concept, corresponding to any method in any of the above embodiments, one or more embodiments of this specification also provide an electronic device, including a memory, a processor, and a computer stored on the memory and capable of running on the processor A program, when the processor executes the program, implements the optical network routing optimization method based on LSTM deep learning described in any one of the above embodiments.

FIG. 5 shows a schematic diagram of a more specific hardware structure of an electronic device provided by this embodiment. The device may include: a processor 1010 , a memory 1020 , an input/output interface 1030 , a communication interface 1040 and a bus 1050 . The processor 1010 , the memory 1020 , the input/output interface 1030 and the communication interface 1040 are connected to each other within the device through the bus 1050 .

The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit, central processing unit), a microprocessor, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute related programs to realize the technical solutions provided by the embodiments of this specification.

The memory 1020 may be implemented in the form of ROM (Read Only Memory, read only memory), RAM (Random Access Memory, random access memory), static storage device, dynamic storage device, and the like. The memory 1020 can store operating systems and other application programs. When implementing the technical solutions provided by the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 1020 and invoked by the processor 1010 for execution.

The input/output interface 1030 is used to connect the input/output module to realize information input and output. The input/output/module can be configured in the device as a component (not shown in the figure), or can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, and the like.

The communication interface 1040 is used to connect a communication module (not shown in the figure), so as to realize the communication interaction between the device and other devices. The communication module can realize communication through wired means (such as USB, network cable, etc.), and can also realize communication through wireless means (such as mobile network, WIFI, Bluetooth, etc.).

Bus 1050 includes a path that carries information between the various components of the device (eg, processor 1010, memory 1020, input/output interface 1030, and communication interface 1040).

It should be noted that although the above device only shows the processor 1010, the memory 1020, the input/output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components. In addition, those skilled in the art can understand that the above-mentioned device may only include components necessary to implement the solutions of the embodiments of this specification, and does not necessarily include all the components shown in the figure.

The electronic device in the above embodiments is used to implement the corresponding LSTM deep learning-based optical network routing optimization method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

Based on the same inventive concept, corresponding to the method in any of the above embodiments, one or more embodiments of this specification also provide a non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium stores computer instructions The computer instructions are used to make the computer execute the optical network routing optimization method based on LSTM deep learning as described in any one of the above embodiments.

The computer-readable medium in this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology. Information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Flash memory or other memory technology, Compact Disc Read-Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridge, tape magnetic disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

The computer instructions stored in the storage medium of the above embodiments are used to make the computer execute the optical network routing optimization method based on LSTM deep learning as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which are not described here. Let me repeat.

It can be seen from the above that the present invention applies the LSTM prediction model in deep learning to update the predicted optimal reconstruction threshold in real time based on the spectrum status information and service information in the optical network link, instead of optimizing for a single target parameter. At the same time, there is no need to use the traffic matrix as the input of the deep learning algorithm, which is more in line with the real scene situation. At the same time, the reconstruction threshold is updated and changed in real time according to the arrival of dynamic services, which can keep the network in the current optimal situation all the time, fully consider the probability and frequency of reconstruction, and can well change the network performance.

The foregoing describes specific embodiments of this specification. Other implementations are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. Multitasking and parallel processing are also possible or may be advantageous in certain embodiments.

Those of ordinary skill in the art should understand that: the discussion of any of the above embodiments is exemplary only, and is not intended to imply that the scope of the present disclosure (including claims) is limited to these examples; under the idea of the present disclosure, the above embodiments or Combinations can also be made between technical features in different embodiments, steps can be implemented in any order, and there are many other variations of the different aspects of one or more embodiments of this specification as described above, which are not included in the details for the sake of brevity. supply.

The description of one or more embodiments is intended to embrace all such alterations, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omission, modification, equivalent replacement, improvement, etc. made within the spirit and principles of one or more embodiments of this specification shall fall within the protection scope of the present disclosure.

Claims (8)

1. An optical network route optimization method based on LSTM deep learning is characterized by comprising the following steps:

receiving a request of a service to be deployed, analyzing the attribute of the service to be deployed, and calculating the shortest path of the service to be deployed according to the attribute of the service to be deployed;

judging whether the spectrum on the shortest path is sufficient;

if not, acquiring the residual time of the deployed service on the blocked path which meets the number of frequency spectrums needed by the service to be deployed, calculating the time difference between the duration time of the service to be deployed and the residual time of the deployed service, and searching a suboptimal path for the service to be deployed for deployment;

collecting the data of the whole path spectrum occupation condition in the optical network, and inputting the collected data into a trained LSTM deep learning model to obtain a preset reconstruction threshold;

the training step of the LSTM deep learning model comprises the following steps:

the first step is as follows: acquiring historical network spectrum occupation condition information and service information in a network;

the second step is that: network spectrum occupation condition information according to the historyService information, generating a spectrum matrix S and a service matrix R according to the time sequence, including in the time sequence t ₁ 、t ₂ Obtaining corresponding frequency spectrum matrixes S1, S2 to Sn and service matrixes R1, R2 to Rn when the frequency spectrum matrixes are tn;

wherein, the form of the service matrix R is as follows:

wherein in the traffic matrix R

Respectively representing services R _i A source node and a sink node;

representative service R _i The required spectrum resources;

representative service R _i The remaining time of (c); p _i Representative service R _i The service matrix is updated at any time along with the arrival and the departure of the service;

the third step: according to the current t _i The collected spectrum matrix S _i And said traffic matrix R _i Input the initial LSTM prediction model as input data, t _i Is represented in the time series t ₁ 、t ₂ To t _n At any one time, use t _i Next time t of _i+1 Of the spectrum matrix S _i+1 And said traffic matrix R _i+1 As label data, training the selection of the preset reconstruction threshold;

the fourth step: according to each time t _i Inputting the frequency spectrum matrix and the service matrix, simulating a predicted reconstruction threshold obtained by output, scoring the predicted reconstruction threshold according to a network service blocking rate obtained by using the predicted reconstruction threshold, and scoring the t _i The prediction reconstruction threshold value obtained at the moment is taken as input information to obtain the t _i+1 Simulating, scoring and inputting the prediction reconstruction threshold value of the moment until the moment is ended;

repeating the four steps, and repeatedly training the LSTM model, wherein the LSTM model can select different prediction reconstruction thresholds for simulation and carries out scoring according to the service blocking rate in the network, and the prediction reconstruction threshold with the highest score is selected as a preset reconstruction threshold;

and judging whether the time difference reaches the preset reconstruction threshold value or not, and distributing paths according to the judgment result.

2. The LSTM deep learning-based optical network route optimization method of claim 1, wherein after determining whether the spectrum on the shortest path is sufficient, the method comprises:

and if the shortest path is sufficient, selecting and allocating the wavelength spectrum of the shortest path of the service to be deployed, updating the spectrum state information on the shortest path and finishing allocation.

3. The method of claim 1, wherein the determining whether the time difference reaches the preset reconstruction threshold and performing the distribution according to the determination result comprises:

if the time difference is greater than the preset reconstruction threshold, the service to be deployed is reconstructed after the time of the time difference, the wavelength spectrum of the shortest path of the service to be deployed is selected and allocated, the spectrum state information on the path is updated, the allocation is finished, and if the time difference is less than or equal to the preset reconstruction threshold, the allocation is finished.

4. The LSTM deep learning-based optical network route optimization method of claim 1, wherein the service attributes include source node, sink node, service start time, service duration and required spectrum number;

the algorithm for calculating the shortest service path is Dijkstra algorithm.

5. The LSTM deep learning-based optical network route optimization method of claim 3, wherein the method adopted for selecting and allocating the wavelength spectrum with shortest path to be deployed is a First-First algorithm.

6. The LSTM deep learning-based optical network route optimization method of claim 1, wherein the spectrum matrix S is of the form:

wherein in the spectral matrix S

Representing a routing node x in a network _i And x _j The spectrum occupation of the link between them, if node x _i And x _j Not directly connected, then

Is 0, if node x _i And x _j Direct coupling and spectrum occupancy of 0, then

Is 1.

7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 6 when executing the program.

8. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1 to 6.

Images