Optical path-designing method

Abstract

In the present invention, an optical path-designing device, by using a scheme that has been reinforcement-trained so as to reduce the amount of optical fiber resources used in an optical network having a plurality of nodes connected by optical fibers, selects an optical path demand to which the optical fiber resources are to be allocated, on the basis of frequency allocation state information indicating available frequencies in each of the optical fibers and optical path demand group information indicating a plurality of optical path demands indicated by combinations of start point nodes and end point nodes. The optical path-designing device refers to the frequency allocation state information, and determines a path and a frequency in the optical network to be allocated to the selected optical path demand such that a low or high frequency is given higher priority.

Description

Optical path design method

[0001] The present invention relates to an optical path design method.

[0002] In optical networks that multiplex multiple optical signals in the frequency domain of optical fibers, communication resources such as routes and frequencies to be allocated in response to demand for communication paths (lightpaths) connecting source and destination nodes are determined. To estimate the amount of optical fiber and frequency resources required to accommodate a large number of lightpath demands in an optical network, it is necessary to determine the route and frequency of each lightpath so that the required frequency for the entire optical network is minimized. There are two methods for solving this problem: mathematical programming and heuristic methods (see, for example, Non-Patent Document 1). While the former can obtain an exact solution based on the designed objective function, the time required to calculate the solution increases exponentially with the calculation scale. Therefore, it is difficult to obtain a solution for large optical networks in a realistic time frame. The latter allows for fast lightpath design based on the algorithm designer's policies. However, the performance of the algorithm depends on the network being designed, etc.

[0003] Meanwhile, the use of reinforcement learning has been proposed in an optical path design method that determines the route and frequency for a single optical path demand (see, for example, Non-Patent Document 2). In this reinforcement learning, an agent that designs optical paths repeatedly improves the policy to minimize the reward (here, the number of fibers required or the amount of frequency resources) obtained as a result of designing an optical network. This is expected to lead to the realization of an optical path design algorithm with higher frequency resource utilization efficiency.

[0004] H. Zang, JP Jue, and B. Mukherjee, “A Review of Routing and Wavelength Assignment Approaches for Wavelength- Routed Optical WDM Networks,” Optical Networks Magazine, p. 47-60, January 2000. Modulation and Spectrum Assignment in Elastic Optical Networks,” Journal of Lightwave Technology, vol. 37, no. 16, pp. 4155-4163, Aug. 2019.

[0005] By applying reinforcement learning as a design method for multiple optical path demands, it may be possible to construct an optical path design algorithm that outperforms conventional heuristic methods for large-scale optical networks where mathematical programming cannot be applied. However, no optical path design method using reinforcement learning has been demonstrated to date.

[0006] In view of the above circumstances, an object of the present invention is to provide an optical path design method that can design an optical path that achieves higher frequency resource utilization efficiency in response to multiple optical path demands.

[0007] An optical path design method according to one aspect of the present invention is an optical path design method comprising: a selection step of selecting an optical path demand to be allocated optical fiber resources based on frequency allocation state information indicating available frequencies in each of the optical fibers and optical path demand group information indicating multiple optical path demands indicated by combinations of the starting node and the ending node, using a reinforcement learned strategy to reduce the amount of optical fiber resources used in an optical network having a plurality of nodes connected by optical fibers; and an allocation step of determining, by referring to the frequency allocation state information, a route and frequency in the optical network to be allocated to the optical path demand selected in the selection step, so as to prioritize smaller or larger frequencies.

[0008] The present invention makes it possible to design optical paths that achieve higher frequency resource utilization efficiency in response to multiple optical path demands.

[0009] FIG. 1 is a configuration diagram of an optical path design device according to a first embodiment of the present invention. FIG. 2 is a diagram for explaining the operation of an optical path demand selection unit according to the first embodiment. FIG. 3 is a diagram for explaining post-processing of an action space generation unit according to the first embodiment. FIG. 4 is a diagram for explaining the operation of an optical path design unit according to the first embodiment. FIG. 5 is a flow diagram showing reinforcement learning processing of the optical path design device according to the first embodiment. FIG. 6 is a configuration diagram of an optical path design device according to a second embodiment. FIG. 7 is a diagram for explaining the operation of an action space generation unit according to the second embodiment. FIG. 8 is a diagram showing the hardware configuration of the optical path design device according to the first and second embodiments.

[0010] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0011] (First Embodiment) The offline optical path design problem is a problem of allocating a large number of optical path demands to an optical network so as to minimize the required amount of optical fiber resources. Conventionally, a method of applying reinforcement learning to the offline optical path design problem has not been established. In the first embodiment, optimization of the optical path design order for a large number of optical path demands using reinforcement learning is combined with route and frequency selection using a heuristic method based on the optimization. This enables optical path design that achieves higher frequency resource utilization efficiency than conventional heuristic methods for multiple optical path demands in a network of realistic scale.

[0012] 1 is a configuration diagram of an optical path design apparatus 10 according to a first embodiment of the present invention. The optical path design apparatus 10 is connected to an optical network system 20.

[0013] The optical network system 20 includes an optical network and a control device. The optical network is composed of multiple nodes connected by optical fibers. The nodes are optical communication devices that send, receive, and forward optical signals. The optical fibers transmit optical signals between the nodes. The control device sets optical paths and frequency bands to be assigned to the optical paths in the optical network. The optical paths are communication paths composed of optical fiber links (hereinafter simply referred to as "links") that connect two nodes, a start node and a finish node.

[0014] The optical path design device 10 performs optical path design in an optical network. Optical path design (hereinafter simply referred to as "design") involves allocating optical paths and frequency bands available in the optical network to optical path demands. The optical path design device 10 includes an optical path demand group storage unit 11, an optical path demand selection unit 12, an optical path design unit 17, and a simulation unit 18.

[0015] The optical path demand group storage unit 11 stores given optical path demand group information. The optical path demand group information indicates one or more optical path demands for the optical network. The optical path demand is an optical path that is in demand, i.e., an optical path for which communication resources have not been allocated. The optical path demand is represented by the start and end nodes and the bandwidth used. The start and end nodes are a combination of the start and end nodes of the optical path. The communication resources include optical fiber resources such as optical fiber links and frequencies.

[0016] The optical path demand selection unit 12 selects a single optical path demand for the next optical path design from the multiple optical path demands indicated by the optical path demand group information. The optical path demand selection unit 12 has a state space generation unit 13, an action space generation unit 14, a reward calculation unit 15, and a policy update unit 16.

[0017] The state space generation unit 13 generates a state space vector based on the optical path demand group information acquired from the optical path demand group storage unit 11 and the frequency allocation state table acquired from the simulation unit 18 or the optical network system 20. The frequency allocation state table indicates the usage status of each frequency in each optical fiber link. The state space vector is a vector combining the optical path demand group information and the frequency allocation state table.

[0018] The action space generation unit 14 generates an action space vector from the state space vector. This uses a function for determining a strategy in reinforcement learning. A neural network is generally used for this function. The action space vector represents a weight for each combination of start and end nodes and bandwidth. Based on the weight represented by the action space vector, the action space generation unit 14 selects the next optical path demand to be designed, which is specified by the combination of start and end nodes and bandwidth.

[0019] The reward calculation unit 15 calculates a reward based on the result of the lightpath design performed for the optical network. The policy update unit 16 executes a reinforcement learning algorithm using the state space vector used in the lightpath design, the action space vector generated using the state space vector, the reward obtained as a result of the lightpath design for the lightpath demand selected based on the action space vector, etc., and updates the weights of the neural network used as a policy by the action space generation unit 14.

[0020] The lightpath design unit 17 performs lightpath design for the lightpath demand selected by the action space generation unit 14 of the lightpath demand selection unit 12, and allocates communication resources. Specifically, the lightpath design unit 17 determines the lightpath to be allocated to the lightpath demand and the design parameters of the lightpath. The design parameters include the center frequency of the lightpath. Furthermore, when multiple modulation methods and multiple symbol rates are available, the design parameters also include parameters specifying the modulation method and symbol rate selected from among them. The lightpath design unit 17 outputs lightpath design information in which the lightpath and the determined design parameters are set to the simulation unit 18 or the optical network system 20.

[0021] The simulation unit 18 simulates the optical network. When reinforcement learning is being performed in the optical path demand selection unit 12, the simulation unit 18 simulates the optical network, thereby enabling the allocation and deletion of optical paths to any point in a simulated manner. The simulation unit 18 sets optical paths, frequencies, and the like in the simulated optical network based on the optical path design information received from the optical path design unit 17. The simulation unit 18 outputs to the optical path demand selection unit 12 the design results of optical paths in the simulated optical network and a frequency allocation state table indicating the usage status of frequencies in each optical fiber link.

[0022] After the reinforcement learning is completed, the optical path design device 10 may output optical path design information to the optical network system 20. The control device of the optical network system 20 sets optical paths, frequencies, and the like in the optical network based on the optical path design information received from the optical path design device 10. In this case, the optical path design device 10 acquires a frequency allocation state table and optical path design results from the control device of the optical network system 20.

[0023] 2 is a diagram for explaining specific operations in the optical path demand selection unit 12. The frequency allocation state table that the optical path demand selection unit 12 acquires from the simulation unit 18 or the optical network system 20 includes the frequency usage status of each optical fiber link of the optical network. For example, let the number of nodes in the optical network be N (N is an integer of 2 or more), and let the node with node number i (i is an integer of 0 or more and (N-1) or less) be referred to as node #i. The optical network shown in FIG. 2 is an example where N=6. Let the optical fiber link between node #i and adjacent node #h (i<h, h is an integer of 1 or more and (N-1) or less) be referred to as link (i-h).

[0024] Here, the unit frequency band is defined as a frequency slot. There are L frequency slots available for use in one link (L is an integer equal to or greater than 1). When allocating an optical path to a demand for an optical path, a frequency slot number of a frequency to be allocated to the optical path is selected from frequency slot numbers #1 to #L. In the frequency allocation status table shown in FIG. 2, 1 is set if the frequency band of each frequency slot number is in use in each optical fiber link, and 0 is set if it is unused.

[0025] The optical path demand group information includes information on the number of optical path demands given to each start and end node. The information on each optical path demand includes the combination of the node number of the start node and the node number of the end node of the optical path, and the bit rate required by the optical path. A start and end node whose start point is node #i and whose end point is node #j, and a start and end node whose start point is node #j and whose end point is node #i, are referred to as start and end nodes (i-j) (i<j, j is an integer between 1 and (N-1) inclusive). The bit rate represents the size of the frequency band used. There may be multiple optical path demands between the same start and end nodes, and each of these optical path demands may have a different bit rate.

[0026] The state space generation unit 13 generates a state space vector using the frequency allocation state table and the optical path demand group information. The state space vector is data input to the neural network 140 of the action space generation unit 14. In FIG. 2, the state space vector is flattened as a one-dimensional vector. That is, the state space generation unit 13 uses a one-dimensional vector obtained by combining a vector obtained by rewriting the frequency allocation state table into a one-dimensional vector and a vector obtained by rewriting the optical path demand group information into a one-dimensional vector as the state space vector. The state space generation unit 13 may process the frequency allocation state table and the optical path demand group information into any format, such as by adding additional information to generate the state space vector.

[0027] The action space generation unit 14 includes a neural network 140. This neural network 140 receives a state space vector as input and outputs an action space vector. The size of the action space vector is equal to the product of the number of types of start and end nodes in the optical path demand group information and the number of bit rate types. Each element of the action space vector corresponds to a combination of start and end nodes and bit rates in the optical path demand group information. The state space vector shown in FIG. 3 is a one-dimensional vector in which elements corresponding to four types of bit rates of 100, 200, 300, and 400 [Gbps] of each start and end node, i.e., start and end node (0-1), start and end node (0-2), ..., start and end node (4-5), are arranged.

[0028] The action space generation unit 14 selects the start and end nodes and bit rate corresponding to the element with the maximum value set in the output action space vector as the optical path demand to be designed next.

[0029] It is also possible to select an appropriate optical path demand by adding post-processing after the generation of the action space vector by the action space generation unit 14. This will be explained using FIG.

[0030] FIG. 3 is a diagram illustrating post-processing performed by the action space generation unit 14 after generating an action space vector. For ease of understanding, the action space vector is shown in a table format with the vertical axis representing the start and end nodes and the horizontal axis representing the bit rate. The action space generation unit 14 calculates the product of the action space vector output by the neural network 140 and the optical path demand mask. This optical path demand mask indicates whether or not a combination of start and end nodes and bit rates for which there is optical path demand exists in the optical path demand group information. Elements of the optical path demand mask are set to 1 if optical path demand exists and 0 if there is no optical path demand. By calculating the logical product of corresponding elements in the optical path demand mask and the action space vector, combinations of start and end nodes and bit rates for which there is no optical path demand can be excluded from the action space vector. The optical path design unit 17 selects the optical path demand corresponding to the element with the maximum value set in the action space vector excluding the non-existent optical path demand as the next design target. This ensures that the selected optical path demand exists.

[0031] 4 is a diagram for explaining the operation of the lightpath design unit 17. The lightpath design unit 17 determines the route and frequency of the optical network to be assigned to the lightpath demand based on the start and end nodes and bit rate of the lightpath demand selected by the action space generation unit 14. In the example of FIG. 4, a case will be explained where the design target selected for a six-node optical network is an optical path demand with start and end nodes (0-2) and a bit rate of 100 Gbps (gigabits per second).

[0032] First, the optical path design unit 17 searches for a route connecting node #0 and node #2. For example, the optical path design unit 17 searches for K routes (K is an integer equal to or greater than 1) using a commonly known K-shortest paths algorithm. Note that other route searches can also be used. In this example, it is assumed that the optical path design unit 17 obtains three candidates: route candidate #1 "0-1-2", route candidate #2 "0-5-1-2", and route candidate #3 "0-5-4-2". Route candidate #1 "0-1-2" is a route that uses link (0-1) and link (1-2). Route candidate #2 "0-5-1-2" is a route that uses link (0-5), link (5-1), and link (1-2). Route candidate #3 "0-5-4-2" is a route that uses link (0-5), link (5-4), and link (4-2).

[0033] Next, the optical path design unit 17 determines the frequency to be assigned to each route candidate. Therefore, the optical path design unit 17 refers to the frequency allocation state table and selects an assignable frequency slot number for each route candidate. The lower diagram in Fig. 4 shows the frequency allocation state table for the optical fiber links included in the route candidates for the start and end nodes (0-2). Here, the smaller the frequency slot number, the lower the frequency.

[0034] An optical path is restricted to using a uniform frequency for all passing links, and furthermore, it is necessary to select consecutive frequency slot numbers according to the size of the frequency band. For example, in route candidate #1, the frequency slot numbers that can be used in common on all links, link (0-1) and link (0-2), are 3 and 4. If the required number of frequency slots for a demand of 100 Gbps is 1, it is necessary to use either frequency slot 3 or frequency slot 4, and if the required number of frequency slots is 2, it is necessary to use both frequency slot 3 and frequency slot 4. On the other hand, if the required number of frequency slots is 3, there is no available frequency band, so route #1 is determined to be unallocatable. It is assumed that the frequency slot granularity required for the bit rate is predetermined.

[0035] Furthermore, the number of required frequency slots may differ depending on the route candidate. For example, route candidate #1 has a hop count of two, while route candidate #2 and route candidate #3 have a hop count of three. In this case, the frequency bandwidth required to achieve transmission that satisfies the bit rate of the optical path demand may be larger than that of route candidate #1. This is because, generally, when optical fiber communication is performed at a fixed bit rate, noise is added to the signal in proportion to the distance, so wideband signals using a modulation method with a low multilevel (such as QPSK (Quadrature Phase Shift Keying)) are used for long-distance transmission paths, while narrowband signals using a modulation method with a high multilevel (such as 16QAM (Quadrature Amplitude Modulation)) that emphasizes frequency utilization efficiency are used for short-distance transmission paths. For example, route candidate #1 uses a 16QAM signal with one slot, while route candidate #2 and route candidate #3 require a QPSK signal with two slots in order for the signal to reach the destination, resulting in a larger required bandwidth. The optical path design unit 17 determines the number of required frequency slots taking into consideration the size of the frequency band (bit rate) of the optical path demand and the modulation method and symbol rate according to the transmission distance (number of hops), and calculates the allocatable frequency slots.

[0036] The optical path design unit 17 refers to the frequency allocation state table, calculates the frequency slots that can be assigned to each of the route candidates #1, #2, and #3, and selects the route candidate and frequency slot number among them with the smallest frequency slot number. If the required frequency slot for all three route candidates #1, #2, and #3 is 1, the candidate frequency slot number for route candidate #1 is 3, the candidate frequency slot number for route candidate #2 is 1, and the candidate frequency slot number for route candidate #3 is 3. Therefore, the optical path design unit 17 obtains route candidate #2 with frequency slot number 1 as the design result.

[0037] If the number of required frequency slots differs for each route candidate, the optical path design unit 17 may select, as a design result, a route candidate having a smaller number of frequency slots with higher priority than a route candidate having a smaller frequency slot number. Alternatively, the optical path design unit 17 may select a route candidate having a larger frequency slot number.

[0038] 5 is a flow diagram showing the reinforcement learning process of the optical path design device 10. First, the state space generation unit 13 of the optical path design device 10 acquires the frequency allocation state table of the optical network being simulated from the simulation unit 18 and reads the optical path demand group information from the optical path demand group storage unit 11. For example, as shown in FIG. 2, the state space generation unit 13 combines the frequency allocation state table and the optical path demand group information to generate a state space vector (step S1). The action space generation unit 14 inputs the state space vector generated in step S1 to the neural network 140 and obtains an output of the action space vector (step S2).

[0039] The action space generation unit 14 selects an optical path demand to be the target of optical path design based on the weight of each combination of start and end nodes and bit rates indicated by the action space vector calculated in step S2 (step S3). Specifically, the action space generation unit 14 selects a combination of start and end nodes and bit rates corresponding to the element having the maximum value set in the action space vector generated by the processing shown in Fig. 2 or the action space vector obtained by performing the post-processing shown in Fig. 3 after the processing shown in Fig. 2.

[0040] The lightpath design unit 17 allocates lightpaths and frequencies of the optical network to the lightpath demands selected in step S3 (step S4). Specifically, as shown in FIG. 4, the lightpath design unit 17 refers to the frequency allocation state table and allocates lightpath routes and frequency slots to the lightpath demands using a heuristic method. The action space generation unit 14 outputs lightpath design information, in which the allocated lightpaths and design parameters indicating the center frequencies based on the allocated frequency slots are set, to the simulation unit 18. The lightpath design information may further include design parameters indicating a modulation method, a symbol rate, etc.

[0041] The simulation unit 18 simulates the results of setting optical paths and frequencies in the optical network in response to the optical path demand selected in step S3 in accordance with the optical path design information. The simulation unit 18 outputs an optical path design result indicating the results of setting optical paths and frequencies in the optical network to the optical path demand selection unit 12. The reward calculation unit 15 calculates a reward for the optical path and frequency assignment result using the optical path design result (step S5). Any function can be used as the reward function for calculating the reward. For example, a reinforcement learning model that efficiently utilizes the frequency resources of the entire optical network can be configured by setting the reward function so that the amount of change in the maximum value of the operating frequency in each optical fiber in the optical network is minimized before and after the optical path assignment. In this case, for example, the reward calculation unit 15 obtains the maximum value of the operating frequency in each optical fiber from the optical path design result, and calculates a reward using the reward function by using the obtained maximum value of the operating frequency and the maximum value of the operating frequency previously obtained.

[0042] The policy update unit 16 uses the reward obtained in step S5 and a reinforcement learning algorithm to update the weights of the neural network 140 used by the action space generation unit 14 in step S2 (step S6). Any reinforcement learning algorithm can be applied in this embodiment. Information required for learning, in addition to the reward, differs depending on the reinforcement learning algorithm. Such information may be stored in a storage unit (not shown) provided in the optical path demand selector 12 as needed. The timing for updating the weights of the neural network 140 can also be set arbitrarily. For example, if a large amount of training data is required to use mini-batch learning, the process of step S6 is performed to update the weights of the neural network once the required amount of data is obtained. Otherwise, the weight update process of step S6 is skipped.

[0043] The state space generation unit 13 deletes the optical path demand selected in step S3 from the optical path demand group information. The state space generation unit 13 determines whether or not at least one optical path demand exists in the optical path demand group information (step S7). If the state space generation unit 13 determines that at least one optical path demand exists, it repeats the processing from step S1. In step S1, the state space generation unit 13 acquires a frequency allocation state table from the simulation unit 18, and this frequency allocation state table shows a state in which optical paths are allocated to the optical network for the optical path demand selected in step S3.

[0044] If the state space generation unit 13 determines that there is no optical path demand in the optical path demand group information, it determines whether the learning in step S6 or the design process from step S1 to step S4 has been repeated a specified number of times (step S8). If the learning or design process has not been repeated a specified number of times, the state space generation unit 13 returns the optical path demand group and the frequency allocation state of the optical network to their initial states (step S9), and repeats the process from step S1 again. On the other hand, if the state space generation unit 13 determines that the learning or design process has been repeated a specified number of times, it ends the process of FIG. 5.

[0045] When designing an optical network for the optical network system 20, the optical path design device 10 performs the processes of steps S1 to S4 and step S7. In step S1, the optical path demand selector 12 obtains a frequency assignment table from the optical network system 20. In step S3, the neural network 140 trained by the process shown in FIG. 5 is used. In step S4, the optical path design unit 17 outputs optical path design information to the optical network system 20. The optical path design device 10 may further perform the processes of steps S5 and S6. In this case, the optical path demand selector 12 of the optical path design device 10 obtains the optical path design result from the optical network system 20.

[0046] If the bit rate of all lightpath demands is the same, the lightpath demand group information indicates the demand number of each start and end node. In this case, the action space vector has the same number of elements as the number of types of start and end nodes, and each element corresponds to a different output start node.

[0047] Second Embodiment In the first embodiment, the optical path design unit 17 determines the route of the optical path to be designed. In this embodiment, the action space generation unit of the optical path demand selection unit determines the route of the optical path to be designed. The second embodiment will be described focusing on the differences from the first embodiment.

[0048] Fig. 6 is a block diagram showing the configuration of an optical path design device 10a according to a second embodiment. In Fig. 6, the same components as those in the optical path design device 10 according to the first embodiment shown in Fig. 1 are denoted by the same reference numerals, and their description will be omitted. The optical path design device 10a shown in Fig. 6 differs from the optical path design device 10 shown in Fig. 1 in that it includes an optical path demand selection unit 12a and an optical path design unit 17a instead of the optical path demand selection unit 12 and the optical path design unit 17. The optical path demand selection unit 12a shown in Fig. 6 differs from the optical path demand selection unit 12 shown in Fig. 1 in that it includes an action space generation unit 14a instead of the action space generation unit 14.

[0049] FIG. 7 is a diagram for explaining the operation of the action space generation unit 14a. The input of the neural network 140a used by the action space generation unit 14a is a state space vector similar to the input of the neural network 140 used by the action space generation unit 14 of the first embodiment. The output of the neural network 140a includes route candidate information in addition to the start and end node and bit rate information similar to the action space vector output by the neural network 140 of the first embodiment. That is, the action space vector output by the neural network 140a has elements corresponding to combinations of route candidates, start and end nodes, and bit rates. The action space generation unit 14a outputs to the lightpath design unit 17a the combination of route candidates, start and end nodes, and bit rates corresponding to the element with the maximum value set among the elements of the action space vector. The number K of route candidates and the K route candidates #1 to #K that can be taken for each start and end node are calculated in advance using the K shortest paths algorithm. The optical path design unit 17a determines an allocatable frequency for a candidate route corresponding to a route candidate number k (k is an integer between 1 and K).

[0050] For example, in FIG. 2, the elements with the maximum value set in the action space vector correspond to route candidate #1, start / end node (0-2), and bit rate 100 Gbps. The optical path design unit 17a selects the start / end node (0-2) and the optical path demand with a bit rate of 100 Gbps as the target for optical path design. Furthermore, from route candidate #1 "0-1-2", route candidate #2 "0-5-1-2", and route candidate #3 "0-5-4-2" that are predetermined for the start / end node (0-2), the optical path design unit 17a selects route candidate #1 "0-1-2". Based on the number of hops for route candidate #1 "0-1-2" and the bit rate of 100 Gbps, the optical path design unit 17a determines that the required number of frequency slots is 1. The optical path design unit 17a refers to the frequency allocation table and selects frequency slot 3, which has the smallest number, from available frequency slots 3 and 4 for both link (0-1) and link (1-2). As a result, the optical path design unit 17a allocates the route "0-1-2" and frequency slot 3 to the optical path demand for the start and end node (0-2) and the bit rate of 100 Gbps.

[0051] In this embodiment, by optimizing the routes that the start and end nodes can take using reinforcement learning, it is possible to achieve higher frequency resource utilization efficiency.

[0052] According to the embodiment described above, it is possible to realize an optical path design that achieves higher frequency resource utilization efficiency than conventional heuristic methods for a plurality of optical path demands, for a realistic network scale.

[0053] 8 is a diagram showing an example of the hardware configuration of the optical path design apparatus 10, 10a. The optical path design apparatus 10, 10a includes a processor 71, a storage unit 72, a communication interface 73, and a user interface 74.

[0054] The processor 71 is a central processing unit that performs calculations and control. The processor 71 is, for example, a CPU (central processing unit) or a GPU (graphics processing unit). The processor 71 reads and executes programs from the storage unit 72. The storage unit 72 further has a work area and the like when the processor 71 executes various programs. The communication interface 73 is connected to other devices so as to be able to communicate with them. The user interface 74 is an input device such as a keyboard, a pointing device (a mouse, a tablet, etc.), a button, a touch panel, etc., and a display device such as a display. Human operations are input via the user interface 74.

[0055] At least some of the functions of the optical path demand selection unit 12, the optical path design unit 17, and the simulation unit 18 of the optical path design device 10, and the optical path demand selection unit 12, the optical path design unit 17a, and the simulation unit 18 of the optical path design device 10a are realized by the processor 71 reading and executing a program from the storage unit 72. The program of the optical path design device 10, 10a may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The program of the optical path design device 10 may be transmitted via a telecommunications line. At least some of the functions of the optical path design device 10 may be realized using hardware such as an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a field-programmable gate array (FPGA). The optical path demand group storage unit 11 is realized using the storage unit 72.

[0056] The optical path design apparatus 10, 10a may be realized by a plurality of computer devices connected to a network. In this case, it is possible to arbitrarily select which of the plurality of computer devices to realize each functional unit of the optical path design apparatus 10, 10a. Furthermore, the same functional unit may be realized by a plurality of computer devices.

[0057] In the optical path design method according to this embodiment, optical fiber resources are allocated to multiple optical path demands so as to reduce the amount of optical fiber resources required in an optical network having multiple nodes connected by optical fibers. Therefore, the optical path design device selects the start and end nodes of the optical path to which optical fiber resources are next allocated from the multiple optical path demands using a reinforcement learning strategy. That is, the optical path design device selects an optical path demand to which optical fiber resources are allocated based on frequency allocation status information indicating available frequencies in each optical fiber and optical path demand group information indicating multiple optical path demands represented by combinations of start and end nodes. After the selection, the optical path design device references the frequency allocation status information for the start and end nodes (optical path demands) of the selected optical path and determines the route and frequency to be allocated using a heuristic method. This heuristic method selects frequencies by preferentially allocating smaller or larger frequencies within the allocable frequency band.

[0058] Alternatively, the optical path design device may use a reinforcement learned strategy to select, from multiple optical path demands, the start and end nodes (optical path demands) of the optical path to which optical fiber resources are next to be allocated and the route connecting those start and end nodes, and then refer to frequency allocation status information and allocate frequencies to the start and end nodes and route of the selected optical path, giving priority to smaller or larger frequencies within the allocable frequency band.

[0059] The optical path demand may include bit rate information. In this case, the optical path design device may determine the bit rate of the optical path as well as the start and end nodes to which optical fiber resources are next allocated, using the policy. Furthermore, the optical path design device may select the start and end nodes of the optical path by excluding candidates that do not have start and end nodes of the optical path demand from the output of the policy. For example, the optical path design device selects an optical path demand to be allocated optical fiber resources based on the weight of the optical path demand included in the optical path demand group information, from among the weights of multiple optical path demands calculated using the policy based on the frequency allocation state information and the optical path demand group information.

[0060] The optical path design device may perform a process of reinforcement learning a policy to reduce the amount of optical fiber resources used in the optical network based on the state of the optical network in which the determined routes and frequencies have been assigned. Furthermore, the reward function used in the reinforcement learning may include a function that minimizes the amount of change in the maximum frequency used in the optical fiber in the optical network before and after the optical path assignment. Furthermore, the input of the neural network used as the policy may include the number of remaining optical paths for each start and end node.

[0061] Although the embodiments of the present invention have been described in detail above with reference to the drawings, the specific configurations are not limited to these embodiments, and include designs within the scope of the present invention that do not deviate from the gist of the present invention.

[0062] It can be used for designing optical paths in optical networks.

[0063] 10, 10a Optical path design device 11 Optical path demand group storage unit 12, 12a Optical path demand selection unit 13 State space generation unit 14, 14a Action space generation unit 15 Reward calculation unit 16 Policy update unit 17, 17a Optical path design unit 18 Simulation unit 20 Optical network system 140, 140a Neural network

Claims

1. An optical path design method comprising: a selection step of selecting an optical path demand to be allocated optical fiber resources based on frequency allocation status information indicating available frequencies in each of the optical fibers and optical path demand group information indicating multiple optical path demands indicated by combinations of the starting node and the ending node, using a strategy that has been reinforced learned to reduce the amount of optical fiber resources used in an optical network having multiple nodes connected by optical fibers; and an allocation step of referring to the frequency allocation status information to determine the route and frequency in the optical network to be allocated to the optical path demand selected in the selection step, giving priority to smaller or larger frequencies.

2. The optical path design method of claim 1, wherein in the selection step, the reinforced learned strategy is used to select an optical path demand to be allocated optical fiber resources and a route to be allocated to the optical path demand to be allocated optical fiber resources based on the frequency allocation state information and the optical path demand group information, and in the allocation step, the frequency allocation state information is referenced to determine the frequency to be allocated to the optical path demand and the route selected in the selection step so as to give priority to smaller or larger frequencies.

3. The optical path design method according to claim 1 or 2, wherein the optical path demand further includes bit rate information.

4. The optical path design method of claim 1, wherein in the selection step, the method uses the strategy to calculate a weight for each of a plurality of optical path demands based on the frequency allocation status information and the optical path demand group information, and selects an optical path demand to be subject to optical fiber resource allocation based on the weight for the optical path demand included in the optical path demand group information from among the calculated weights.

Images