Optical path designing device

Abstract

This optical path designing device comprises: a behavior information generation unit that inputs, to a neural network trained on the basis of a reinforcement learning policy, a state space vector which includes demand information including identification information of a start point and an end point for each node of an optical network and which includes allocation information for each frequency band in each path connecting a start point to an end point, and that generates a path behavior vector and a frequency behavior vector independently; and a determination unit that determines a combination of a path and a frequency band by either leaving, in a frequency behavior vector, each frequency band which can be allocated to any one of the links constituting the path selected on the basis of the element value of the path behavior vector and which satisfies a condition based on the demand information, or by leaving, in a path behavior vector, a path which includes each link that allows allocation of a frequency band selected on the basis of the element value of the frequency behavior vector and the condition based on the demand information.

Description

Optical path design equipment

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

[0002] Optical signals may be multiplexed in the frequency band of links (optical fibers) between nodes (communication nodes) in an optical network. In addition, an optical path design device allocates frequency bands (communication resources) to routes connecting start and end points of each node in the optical network according to demand information for the optical paths.

[0003] Non-Patent Document 1 discloses a method for designing optical paths that uses a neural network trained using a reinforcement learning technique to search for optical path parameters. In reinforcement learning, an agent designs the parameters of optical paths in an optical network. In Non-Patent Document 1, the reward to the agent increases as the number of required optical fibers (path length) and the amount of frequency resources decrease. The agent updates the reinforcement learning strategy (weighting of each neuron in the neural network) so as to maximize this reward. This realizes an optical path design algorithm that further improves the utilization efficiency of frequency resources.

[0004] FIG. 10 is a diagram showing an example of a frequency allocation status table. The frequency allocation status table includes frequency allocation information 11 (frequency usage information). The frequency allocation information 11 is information that represents the allocation status (usage status) of frequency bands in an optical network. In the frequency allocation information 11, an allocated frequency slot (an in-use slot) is represented as "1". An unallocated frequency slot (an empty slot) is represented as "0". The frequency band may be expressed in units of frequency slots. In FIG. 10, at least one of the "S" frequency slot numbers from slot number "S0" to slot number "S-1" is selected as a candidate number for a frequency slot to be allocated to the path (optical path) of an optical signal.

[0005] Fig. 11 is a diagram showing an example of demand information. The demand information of an optical path includes start and end point information 12. The start and end point information 12 includes identification information of the start and end points (start and end points). In Fig. 11, for the demand information of an optical path between node number "N0" and node number "N2", only the node numbers of the start and end points are expressed as "1" (one-hot expression). Note that the method of expressing the start and end point nodes is not limited to a specific expression method.

[0006] 12 is a diagram showing an example of a state space vector. The optical path design device flattens the frequency allocation information 11 so that it becomes a one-dimensional vector. Similarly, the optical path design device flattens the start and end point information 12 so that it becomes a one-dimensional vector. The state space vector is composed of the flattened frequency allocation information 11 and the start and end point information 12.

[0007] 13 is a diagram showing an example of input and output of a neural network that generates an action space vector. The weighting of each neuron in the neural network is updated based on a reinforcement learning strategy (the probability that an element selected in the action space vector is the optimal solution). The light path design device inputs a state space vector to the neural network. The neural network generates an action space vector from the state space vector based on the weighting of each neuron in the neural network.

[0008] 14 is a diagram showing an example of the action space vector 13. Each element of the generated action space vector 13 is associated with both the number of "K" different routes connecting the start point and the end point and a frequency slot number. The route and frequency slot corresponding to the maximum element value of the action space vector 13 are determined as design parameters representing communication resources allocated in response to the demand information of the lightpath.

[0009] Takafumi Tanaka and Masayuki Shimoda, “Pre- and post-processing techniques for reinforcement-learning-based routing and spectrum assignment in elastic optical networks,” Journal of Optical Communications and Networking, vol. 15, no. 12, pp. 1019-1029, Dec. 2023.

[0010] However, in large-scale optical networks, there are many candidate routes as design parameters. Furthermore, the number of elements (size) of the action space vector increases with the increase in the frequency bands that can be allocated to the optical fiber. The number of elements of the action space vector is equal to the product "K x S" of the number of candidate routes "K" and the number of frequency slots (frequency bands) that can be allocated to the routes "S." The number of neurons and the number of neuron layers in the neural network required for proper reinforcement learning increase with the number of elements of the action space vector. Therefore, more time and memory are required to obtain the optimal solution.

[0011] Thus, there is a problem in that it is not possible to improve the efficiency of designing optical paths according to demand information while suppressing an increase in the size of the neural network that generates the behavior vectors used in designing optical paths.

[0012] In view of the above circumstances, the present invention aims to provide an optical path design device that can improve the efficiency of designing optical paths according to demand information while suppressing an increase in the size of the neural network that generates the behavior vectors used in designing optical paths.

[0013] One aspect of the present invention is an optical path design device comprising: an action information generation unit that inputs demand information including identification information of the start point and end point of each node in an optical network, and a state space vector including allocation information of each frequency band in each route connecting the start point to the end point, into a neural network trained based on a reinforcement learning strategy, and independently generates a route action vector whose elements are route candidates and a frequency action vector whose elements are frequency band candidates using the neural network; and a determination unit that determines a combination of the route and the frequency band by leaving in the frequency action vector each frequency band that can be assigned to any of the links constituting the route selected based on the element values ​​of the route action vector and that satisfies a condition based on the demand information, or by leaving in the route action vector each route that includes each link to which a frequency band selected based on the element values ​​of the frequency action vector and a condition based on the demand information can be assigned.

[0014] The present invention makes it possible to improve the efficiency of designing optical paths according to demand information while suppressing an increase in the size of the neural network that generates the behavior vectors used in designing optical paths.

[0015] 1 is a diagram illustrating an example of the configuration of an optical path design system in the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of an optical network in the first embodiment. FIG. 3 is a diagram illustrating an example of the configuration of a behavior information generation unit in the first embodiment. FIG. 4 is a diagram illustrating an example of the operation of a route determination unit and a frequency determination unit in the first embodiment. FIG. 5 is a flowchart illustrating an example of the operation of an optical path design system in the first embodiment. FIG. 6 is a diagram illustrating an example of the operation of a route determination unit and a frequency determination unit in the second embodiment. FIG. 7 is a diagram illustrating an example of the configuration of a behavior information generation unit, a route determination unit, and a frequency determination unit in the third embodiment. FIG. 8 is a diagram illustrating an example of the operation of a route determination unit and a frequency determination unit in the fourth embodiment. FIG. 9 is a diagram illustrating an example of the hardware configuration of an optical path design device in each embodiment. FIG. 10 is a diagram illustrating an example of a frequency allocation state table. FIG. 11 is a diagram illustrating an example of demand information. FIG. 12 is a diagram illustrating an example of a state space vector. FIG. 13 is a diagram illustrating an example of input and output of a neural network that generates an action space vector. FIG. 14 is a diagram illustrating an example of an action space vector.

[0016] Embodiments of the present invention will be described in detail with reference to the drawings. (First Embodiment) Fig. 1 is a diagram showing an example of the configuration of an optical path design system 1 in a first embodiment. The optical path design system 1 is a system that designs optical paths in an optical network. The optical path design system 1 includes an optical network 2, a control device 3, a demand information generation device 4, and an optical path design device 5.

[0017] The optical path design device 5 includes a state information generation unit 51, a reward generation unit 52, a strategy update unit 53, an action information generation unit 54, and a determination unit 55. The determination unit 55 includes a route determination unit 551 and a frequency determination unit 552.

[0018] The optical network 2 includes a plurality of nodes and links between the nodes. Under the control of the control device 3, the optical network 2 allocates a frequency band of an optical signal to a path (optical path) connecting a start point to a destination point of each node of the optical network 2.

[0019] FIG. 2 is a diagram showing an example of the configuration of the optical network 2 in the first embodiment. In the following, the number of nodes 21 constituting the optical network 2 is, for example, six. The starting point of the optical path is, for example, node 21-0 "N0." The ending point of the optical path is, for example, node 21-2 "N2." In the first embodiment, the number of frequency slots required for allocation of an optical path is, for example, one. Note that the number of frequency slots required for allocation of an optical path is determined based on, for example, the path length of the optical path (transmission distance of the optical signal) and the bit rate in the demand information.

[0020] Returning to FIG. 1 , the explanation of the exemplary configuration of the optical path design system 1 will continue. The control device 3 allocates the frequency band of the optical signal to the route connecting the start point and the end point based on the design information (parameters) acquired from the determination unit 55. The control device 3 outputs the design result of the optical network 2 to the remuneration generation unit 52. The control device 3 outputs the frequency allocation information to the status information generation unit 51. The frequency allocation information is information that represents the allocation status (usage status) of the frequency band in the optical network 2. The demand information generation device 4 generates demand information (information on communication demand) of the optical path in the optical network 2. The demand information includes identification information of the start and end points of each node in the optical network 2.

[0021] The optical path design device 5 (agent) determines the optimal route and frequency band in the optical network 2 based on demand information of optical paths in the optical network 2. The status information generator 51 acquires demand information (communication demand information) of optical paths in the optical network 2 from the demand information generator 4. The status information generator 51 acquires frequency allocation information of the optical network 2 from the control device 3.

[0022] The state information generator 51 generates a state space vector based on the demand information and frequency allocation information. Here, the state information generator 51 flattens the frequency allocation information so that it becomes a one-dimensional vector. Similarly, the state information generator 51 flattens the start and end point information 12 so that it becomes a one-dimensional vector. The state space vector is composed of the flattened frequency allocation information and start and end point information.

[0023] The reward generator 52 acquires from the control device 3 the design results of the optical path based on the design information (parameters) determined by the determiner 55 (for example, information representing the results of the number of optical fibers required and the amount of frequency resources). The reward generator 52 generates reward information for reinforcement learning based on the design results of the optical path. The reinforcement learning is, for example, multitask reinforcement learning. The reward generator 52 outputs the reward information to the policy updater 53.

[0024] The policy update unit 53 updates the policy in reinforcement learning based on the reward information, the action information generation unit 54 input to the state space vector, and the action space vector generated by the action information generation unit 54. The policy update unit 53 executes reinforcement learning of the neural network that generates the action space vector based on the policy. That is, the policy update unit 53 updates the weighting of each neuron in the neural network that generates the action space vector based on the policy.

[0025] The policy update unit 53 updates the reinforcement learning policy for each piece of demand information so that the reward is maximized as the number of required optical fibers (path length) and the amount of frequency resources become smaller. This realizes a lightpath design algorithm (scalable lightpath design) that improves the utilization efficiency of frequency resources.

[0026] The policy update unit 53 may use a computer to perform a simulation in advance of a combination (design information) of a route and a frequency band of the optical network 2, based on demand information for any start point and end point of each node of the optical network 2. The policy update unit 53 may update the weighting of each neuron of a neural network that generates an action space vector, based on the simulation result.

[0027] The behavior information generation unit 54 includes a neural network that generates a behavior space vector. The behavior information generation unit 54 performs reinforcement learning on the neural network that generates the behavior space vector based on the policy determined by the policy update unit 53. The behavior information generation unit 54 inputs the state space vector to this neural network. The behavior information generation unit 54 acquires the behavior space vector from this neural network.

[0028] The behavior space vector is composed of a path behavior vector and a frequency behavior vector. The path behavior vector is a vector whose elements are candidate paths connecting the start point and the end point of each node in the optical network 2. The frequency behavior vector is a vector whose elements are candidate frequency bands to be assigned to links between nodes in the optical network 2. The behavior information generator 54 generates the path behavior vector and the frequency behavior vector independently using a neural network that generates the behavior space vector.

[0029] The determination unit 55 independently acquires a path behavior vector and a frequency behavior vector from the neural network. The path determination unit 551 determines a path based on the element values ​​of the path behavior vector (the probability that the selection is the optimal solution for each path). The frequency determination unit 552 leaves each frequency band that can be assigned to any of the links that make up the selected path in the frequency behavior vector. In other words, the frequency determination unit 552 excludes each frequency band that cannot be assigned to any of the links that make up the selected path from the frequency behavior vector. The determination unit 55 transmits design information (parameters) that represent the combination of the determined path and the frequency band with the largest element value in the frequency behavior vector to the control device 3.

[0030] 3 is a diagram showing an example of the configuration of the behavior information generation unit 54 in the first embodiment. The behavior information generation unit 54 includes a neural network 541. The neural network 541 generates an action space vector from a state space vector based on a reinforcement learning strategy. The action space vector includes a path action vector whose elements are path candidates and a frequency action vector whose elements are frequency band candidates. The neural network 541 outputs the path action vector and the frequency action vector independently to the determination unit 55.

[0031] The number of elements (size) of the action space vector is the sum "S+K" of the number of elements of the action space vector corresponding to the number of frequency slot candidates "S" and the number of elements of the action space vector corresponding to the number of route candidates "K".

[0032] In the action space vector generated by the neural network 541, as long as the number of elements of the frequency action vector is "S" and the number of elements of the path action vector is "K", the configuration of the neural network 541 may be any configuration. For example, a layer for generating only the path action vector may not be provided after the fully connected layer in the neural network 541. For example, a layer for generating only the frequency action vector may not be provided after the fully connected layer in the neural network 541.

[0033] Next, an operation example of the optical path design system 1 will be described. Fig. 4 is a diagram showing an operation example of the route determination unit 551 and the frequency determination unit 552 in the first embodiment. In the following, the number of elements "K" of the route space vector is, for example, three. The number of elements "S" of the frequency space vector is, for example, six.

[0034] The route determination unit 551 includes a predetermined route candidate table. In the route candidate table, start and end points are associated with route numbers and link configurations. The "K" different link configurations (routes) for each start and end point are obtained by, for example, executing a k-shortest path algorithm for combinations of start and end points in the optical network 2.

[0035] The route determination unit 551 determines the route candidate number corresponding to the maximum element value of the route space vector. The element value of the route space vector represents the probability that selecting the route corresponding to that element value is the optimal solution for maximizing the reward. In FIG. 4, the route determination unit 551 selects the route candidate number "K1" corresponding to the maximum element value "0.3". The route determination unit 551 transmits the route number "K1" to the control device 3 as route design information (parameter).

[0036] The route determination unit 551 obtains from the route candidate table the identification information "N0-N2" of the start and end points obtained from the demand information, and information on "N0-N5-N4-N2" which is composed of three links "N0-N5", "N5-N4", and "N4-N2" corresponding to the route number "K1".

[0037] The route determination unit 551 refers to the frequency allocation information acquired from the control device 3. The route determination unit 551 extracts, from the frequency allocation information, frequency slots that can be assigned to route "K1" for each link corresponding to route number "K1". In Fig. 4, the route determination unit 551 extracts, from the frequency allocation information, frequency slot numbers "S2", "S3", and "S4" that can be assigned in common to three links "N0-N5", "N5-N4", and "N4-N2" that correspond to route "K1".

[0038] The route determination unit 551 generates frequency assignment feasibility information based on the extraction result from the frequency assignment information. In the frequency assignment feasibility information, a frequency slot that can be assigned to a link (a free slot) is represented as "1." In the frequency assignment feasibility information, a frequency slot that cannot be assigned to a link (a slot in use) is represented as "0."

[0039] The numbers of the frequency slots that can be commonly assigned to the three links "N0-N5", "N5-N4", and "N4-N2" are number "S2", number "S3", and number "S4". The path determination unit 551 generates a frequency mask vector. In the frequency mask vector illustrated in FIG. 4, the element values ​​corresponding to the frequency slot numbers "S2", "S3", and "S4" are "1", and the other element values ​​are "0".

[0040] The frequency determination unit 552 multiplies the frequency action vector by the frequency mask vector element by element. This results in a frequency action vector from which frequency bands that cannot be assigned to the path "K1" have been removed. The frequency determination unit 552 selects the frequency slot number "S2" corresponding to the maximum element value "0.3" of the obtained frequency action vector from the frequency action vector. The frequency determination unit 552 transmits the selected frequency slot number "S2" to the control device 3 as frequency design information (parameter).

[0041] The frequency determination unit 552 may transmit the frequency slot number "S4" corresponding to the second largest element value "0.2" in the obtained frequency action vector as frequency design information to the control device 3. The frequency determination unit 552 may transmit the frequency slot number "S3" corresponding to the third largest element value "0.1" in the obtained frequency action vector as frequency design information to the control device 3.

[0042] 5 is a flowchart showing an example of the operation of the optical path design system 1 according to the first embodiment. The determination unit 55 selects each candidate route and frequency slot based on demand information (step S101). The determination unit 55 determines whether each candidate route and frequency slot has been determined as the final route and frequency slot (step S102).

[0043] If it is determined that each candidate route and frequency slot has been determined as the final route and frequency slot (step S102: YES), the determination unit 55 determines that the optical path design is possible. The determination unit 55 transmits design information (parameters) representing the combination of the determined route and the frequency band with the maximum element value in the frequency behavior vector to the control device 3 (step S103).

[0044] If it is determined that the candidates for the route and frequency slot have not been determined as the final route and frequency slot (for example, if it is determined that the selected route "K1" does not have an allocable frequency band) (step S102: NO), the determination unit 55 determines whether or not there are remaining route candidates (route numbers other than the selected route number) for the start and end points (step S104).If it is determined that there are no remaining route candidates (step S104: NO), the determination unit 55 determines that it is possible to design an optical path and terminates the path design process (step S105).

[0045] If it is determined that there are remaining route candidates (step S104: YES), the determination unit 55 selects a route to which a frequency slot can be allocated from the remaining route candidates. This reduces the possibility that an optical path cannot be designed (step S106). In the route action vector, the element value corresponding to the route candidate to which allocation was not possible is reset to "0". In the route action vector, the determination unit 55 allocates a frequency slot to the route selected from the remaining route candidates (step S107). The determination unit 55 returns the process to step S102.

[0046] The reward generator 52 may reduce the reward each time a process of selecting another route candidate from the remaining route candidates in the route action vector is executed, thereby reducing the possibility that a non-optimal element has the maximum value in the route action vector output from the neural network 541.

[0047] As described above, the behavior information generator 54 inputs demand information including identification information of the start point and end point of each node in the optical network 2 and a state space vector including allocation information of each frequency band on each route connecting the start point and the end point to the neural network 541 trained based on a reinforcement learning strategy. The behavior information generator 54 uses the neural network 541 to independently generate a route behavior vector whose elements are candidate routes and a frequency behavior vector whose elements are candidate frequency bands.

[0048] The determination unit 55 determines the combination of the route "K1" and the frequency band "S2" by leaving in the frequency action vector each frequency band that can be assigned to any of the links that make up the route "K1" selected based on the maximum element value "0.3" of the route action vector and that satisfies the condition based on the demand information (the condition that it can be assigned to a route connecting the start point "N0" and the end point "N2").

[0049] When there is no frequency band that can be assigned to the route selected as a candidate, the determination unit 55 may exclude the selected candidate and select another route from the route action vectors. For example, when there is no frequency band that can be assigned to the route "K0" selected as a candidate, the determination unit 55 may exclude the selected candidate "K0" and select the route "K1". The determination unit 55 may execute a process of determining a combination of the route "K1" selected after excluding the candidate "K0" and a frequency band.

[0050] This makes it possible to improve the efficiency of designing optical paths in accordance with the demand for optical paths (optical path demand information) while suppressing an increase in the size of the neural network that generates the behavior vectors used in designing optical paths.

[0051] Second Embodiment The second embodiment is mainly different from the first embodiment in that the demand information includes not only identification information of the start and end points but also bit rate information. The second embodiment will be described focusing on the differences from the first embodiment.

[0052] 6 is a diagram illustrating an example of the operation of the route determination unit 551 and the frequency determination unit 552 in the second embodiment. When optical paths having a plurality of different bit rates can be mixed in the optical network 2, the demand information may include not only identification information of the start and end points but also bit rate information for each hop number (transmission distance) of the path along which the optical signal is transmitted. The bit rate information may be expressed using at least one of a modulation method and a symbol rate.

[0053] 6, the identification information of the start and end points is, for example, identification information "N0-N2." The route with the largest element value in the route space vector is, for example, route "K0." The route determination unit 551 detects the link configuration "N0-N1-N2" by referring to the route candidate table based on the identification information "N0-N2" and the route "K0."

[0054] The route determination unit 551 extracts from the frequency allocation information the numbers "S2", "S3", "S4", and "S5" of frequency slots that can be commonly assigned to the link "N0-N1" and the link "N1-N2" that make up the route "K0". The route determination unit 551 generates frequency allocation feasibility information based on the extraction results from the frequency allocation information.

[0055] The required slot number table indicates the number of frequency slots required for each combination of the number of hops in a path for transmitting an optical signal and a bit rate. The number of hops in a path for transmitting an optical signal may be expressed by the transmission distance of the optical signal or by the path length. The required slot number table is determined in advance.

[0056] The path determination unit 551 generates a frequency number mask vector based on the required slot number table and the numbers of frequency slots that can be commonly assigned to all links in the frequency assignment feasibility information. Here, the path determination unit 551 selects one of the commonly assignable frequency slot numbers "S2," "S3," "S4," and "S5" as the first slot number. For example, when the number of required frequency slots is three, the path determination unit 551 determines the first slot number so that the three consecutive frequency slot numbers from the first slot number among the commonly assignable frequency slot numbers remain in the frequency action vector. In FIG. 6, the first slot number is at least one of the numbers "S2" and "S3." In FIG. 6, the element values ​​corresponding to the frequency slot numbers "S2" and "S3" in the frequency number mask vector are "1," and the element values ​​corresponding to the frequency slot numbers "S0," "S1," "S4," and "S5" are "0."

[0057] The frequency determination unit 552 multiplies the frequency action vector by the frequency mask vector element by element, thereby excluding frequency bands that cannot be assigned to the route "K1" and obtaining a frequency action vector corresponding to the bit rate information for each hop number.

[0058] Even if the number of frequency slots is fixed according to the bit rate for each hop count, the optical path design method exemplified in the second embodiment is applicable.

[0059] As described above, the determination unit 55 determines the number of frequency bands based on the bit rate based on the demand information and the number of hops of the route. For example, the determination unit 55 determines the number of frequency bands to be "3" based on the bit rate "200 Gbps" based on the demand information and the number of hops "2" of the route "K0".

[0060] The determination unit 55 determines combinations of the determined number of frequency bands and routes. For example, the determination unit 55 may determine a combination of three frequency bands (frequency slot numbers "S2", "S3", and "S4") and route "K0" as design information (parameters). For example, the determination unit 55 may determine a combination of three frequency bands (frequency slot numbers "S3", "S4", and "S5") and route "K0" as design information (parameters). Note that these three consecutive frequency bands in the frequency action vector may be represented using the first slot number (number "S2" or number "S3").

[0061] This makes it possible to improve the efficiency of designing optical paths according to demand information while suppressing an increase in the size of the neural network that generates the behavior vectors used in designing optical paths.

[0062] (Third Embodiment) In the third embodiment, the main difference from the second embodiment is that information on the route candidates determined (selected) by the route determination unit 551 is input to a partial neural network that selects a frequency slot number. In the third embodiment, the differences from the second embodiment will be mainly described.

[0063] 7 is a diagram showing an example of the configuration of the behavior information generation unit 54, the route determination unit 551, and the frequency determination unit 552 in the third embodiment. The route determination unit 551 determines route candidates from route motion vectors. The frequency determination unit 552 includes a partial neural network that selects a frequency slot number. The frequency determination unit 552 inputs information about the route candidates determined (selected) by the route determination unit 551 to this partial neural network.

[0064] The format of the route candidate information may be a route action vector format (number of elements "K") or a route candidate number format (number of elements "1"). The route action vector format may be a one-hot representation format of the route action vector (number of elements "K").

[0065] As described above, the determination unit 55 may input the determined path to the partial neural network that determines the frequency band. The determination unit 55 may determine a combination of the path and the frequency band by acquiring the determined frequency band from the partial neural network.

[0066] This makes it possible to improve the efficiency of designing optical paths according to demand information while suppressing an increase in the size of the neural network that generates the behavior vectors used in designing optical paths. Also, the frequency determination unit 552 can generate frequency behavior vectors with higher accuracy.

[0067] (Fourth Embodiment) In the fourth embodiment, the main difference from the first to third embodiments is that the frequency slot number is determined (selected) before the path is determined (selected). In the fourth embodiment, the difference from the first embodiment will be mainly described.

[0068] 8 is a diagram showing an example of the operation of the route determination unit 551 and the frequency determination unit 552 in the fourth embodiment. The frequency determination unit 552 determines the frequency slot number corresponding to the maximum element value of the frequency action vector as the frequency slot number "S1". The route determination unit 551 extracts the numbers (number "K0", number "K1", and number "K2") of "K" routes connecting the start point and the end point for the identification information of the start and end points in the demand information from the route candidate table.

[0069] The route determination unit 551 generates a route mask vector. The route mask vector indicates, for each route, whether or not a frequency band having the extracted frequency slot number as the leading slot number can be assigned. In the route mask vector illustrated in FIG. 8 , the element values ​​corresponding to route numbers “K1” and “K2” are “1”, and the other element values ​​are “0”.

[0070] The route determination unit 551 multiplies the route action vector by the route mask vector element by element. This results in a route action vector excluding the route "K0" that cannot be assigned to the frequency band of frequency slot number "S1". The route determination unit 551 selects the route number "K1" corresponding to the maximum element value "0.3" of the obtained route action vector from the route action vector. The route determination unit 551 transmits the selected route number "K1" to the control device 3 as route design information (parameter).

[0071] As described above, the determination unit 55 determines the combination of the route "K1" and the frequency band "S1" by leaving in the route action vector the routes "K1" and "K2" including each link to which the frequency band "S1" selected based on the maximum element value "0.4" of the frequency action vector and the condition based on the demand information (the condition that the frequency band "S1" can be assigned to a route connecting the start point "N0" and the end point "N2") In other words, the determination unit 55 of the first embodiment may determine (select) the frequency slot number before the route number, as in the fourth embodiment.

[0072] This makes it possible to improve the efficiency of designing optical paths in accordance with the demand for optical paths (optical path demand information) while suppressing an increase in the size of the neural network that generates the behavior vectors used in designing optical paths.

[0073] The determination unit 55 of the second embodiment may also determine (select) the frequency slot number before the route number. For example, the determination unit 55 selects, as the first slot number, the frequency band "S2" selected based on the maximum element value "0.3" of the frequency behavior vector and conditions based on demand information (conditions that the start point "N0", the end point "N2", the bit rate "200 Gbps", and the number of hops of the route "2" are satisfied). The determination unit 55 may determine the combination of the route "K0" and the frequency band "S2" by leaving, in the route behavior vector, the route "K0" including each link to which three consecutive frequency bands from this first slot number can be assigned.

[0074] The determination unit 55 of the third embodiment may also determine (select) the frequency slot number before the path number. For example, the determination unit 55 may input the determined frequency band number (frequency slot number) to a partial neural network that determines the path number. The determination unit 55 may determine a combination of a path and a frequency band by acquiring the determined path number from the partial neural network.

[0075] Each device of the present invention can be realized by a computer and a program, and the program can be recorded on a recording medium or provided via a communication network.

[0076] (Hardware Configuration) FIG. 9 is a diagram illustrating an example of the hardware configuration of the optical path design device 5 in each embodiment. The optical path design device 5 is realized as software by a processor 101, such as a CPU (Central Processing Unit), executing a program stored in a storage device 103 having a non-volatile recording medium (non-transitory recording medium) and a memory 102. The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as a flexible disk, a magneto-optical disk, a ROM (Read Only Memory), and a CD-ROM (Compact Disc Read Only Memory), and non-transitory recording media such as a storage device built into a computer system, such as a hard disk or a solid state drive (SSD). The communication unit 104 executes predetermined communication processing.

[0077] The optical path design device 5 may be realized using hardware including an electronic circuit (electronic circuit or circuitry) using, for example, an LSI (Large Scale Integrated circuit), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array).

[0078] Although an embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention.

[0079] The present invention is applicable to an optical path design system for an optical communication system.

[0080] 1...optical path design system, 2...optical network, 3...control device, 4...demand information generation device, 5...optical path design device, 11...frequency allocation information, 12...start and destination point information, 13...action space vector, 21...node, 51...state information generation unit, 52...reward generation unit, 53...strategy update unit, 54...action information generation unit, 55...determination unit, 101...processor, 102...memory, 103...storage device, 104...communication unit, 541...neural network, 551...route determination unit, 552...frequency determination unit

Claims

1. An optical path design device comprising: an action information generation unit that inputs demand information including identification information of the start point and end point of each node in an optical network and a state space vector including allocation information of each frequency band in each route connecting the start point to the end point into a neural network trained based on a reinforcement learning strategy, and uses the neural network to independently generate a route action vector whose elements are route candidates and a frequency action vector whose elements are frequency band candidates; and a determination unit that determines a combination of the route and the frequency band by leaving in the frequency action vector each frequency band that can be assigned to any of the links constituting the route selected based on the element values ​​of the route action vector and that satisfies a condition based on the demand information, or by leaving in the route action vector each route that includes each link to which a frequency band selected based on the element values ​​of the frequency action vector and a condition based on the demand information can be assigned.

2. The optical path design device of claim 1, wherein the determination unit determines the number of frequency bands based on the bit rate based on the demand information and the number of hops of the route, and determines the combination of the determined number of frequency bands with the route.

3. The optical path design device of claim 1, wherein the determination unit determines the combination of the route and the frequency band by inputting the determined route into a partial neural network that determines the frequency band and obtaining the determined frequency band from the partial neural network.

4. The optical path design device of claim 1, wherein the determination unit excludes the selected candidate and selects another route from the route behavior vectors if there is no frequency band that can be assigned to the route selected as a candidate.

Images