Parameter selection for network communication links using reinforcement learning

1. A method of operating a communication network that includes a plurality of nodes, the method comprising:

receiving, by one or more processors, input data related to a state of the communication network including locations of the plurality of nodes and trajectories of the plurality of nodes for a first time interval;

determining, by the one or more processors, a first implementation policy for the communication network based on the input data, the first implementation policy being a set of features for forming one or more communication links in the communication network over the first time interval, the one or more communication links including a given communication link between a terrestrial terminal and a high-altitude platform terminal;

operating, by the one or more processors, the communication network to implement the first implementation policy in the first time interval;

determining, by the one or more processors, a utility value associated with the first implementation policy as a function of a performance metric of the communication network in the first time interval;

determining, by the one or more processors, a second implementation policy for the communication network for a second time interval based at least in part on the utility value associated with the first implementation policy; and

operating, by the one or more processors, the communication network to implement the second implementation policy in the second time interval.

2. The method of claim 1 , wherein the plurality of nodes includes one or more balloons.

3. The method of claim 1 , wherein the input data also includes data related to operation of the communication network in a geographic area that includes the terrestrial terminal.

4. The method of claim 1 , wherein the set of features includes characteristics for one or more communication beams for transmission or reception at each node of the communication network.

5. The method of claim 1 , wherein the one or more processors form a neural network.

6. The method of claim 5 , wherein further comprising training, by the one or more processors, the neural network by:

receiving training input data related to state information of the communication network;

determining a training policy based on the training input data;

simulating the training policy based on internal and external influences of the communication network; and

determining a mining-utility value of the training policy according to the simulation.

7. The method of claim 1 , wherein determining the first implementation policy includes:

identifying a trend in features of policies stored in a database; and

selecting features for the first implementation policy that increase performance metric of the communication network according to the trend.

8. A method of operating a communication network that includes a plurality of nodes, the method comprising:

receiving, by one or more processors, input data related to a state of the communication network including locations of the plurality of nodes and trajectories of the plurality of nodes for a first time interval;

determining, by the one or more processors, a training policy based on the input data, the training policy being a set of features for forming a first set of communication links in the commination network over the first time interval;

simulating by the one or more processors, the training policy over the first time interval based on internal and external influences of the communication network;

determining, by the one or more processors, a utility value of the training policy as a function of a performance metric of the communication network in the simulation;

determining, by the one or more processors, a first implementation policy based at least in part on the utility value associated with the training policy, the first implementation policy being a set of features for forming a second set of communication links in the communication network over a second time interval, the second set of communication links including a given communication link between a terrestrial terminal and a high-altitude platform terminal; and

operating, by the one or more processors the communication network to implement the first implementation policy in the second time interval.

9. The method of claim 8 , further comprising:

determining, by the one or more processors, a second utility value associated with the first implementation policy as a function of a second performance metric of the communication network in the second time interval;

determining, by the one or more processors, a second implementation policy for the communication network for a third time interval based at least in part on the second utility value associated with the first implementation policy; and

operating, by the one or more processors, the communication network to implement the second implementation policy in the third time interval.

10. The method of claim 8 , wherein the input data also includes data related to operation of the communication network in a geographic area that includes the terrestrial terminal.

11. The method of claim 8 , wherein the set of features includes characteristics for one or more communication beams for transmission or reception at each node of the communication network.

12. The method of claim 8 , wherein the one or more processors form a neural network.

13. The method of claim 8 , wherein determining the training policy includes:

identifying a trend in features of policies stored in a database; and

selecting features that increase the performance metric of the communication network according to the trend.

14. The method of claim 8 , wherein determining the first implementation policy includes:

identifying a trend in features of policies stored in a database, the policies stored in the database including the training policy; and

selecting features that increase a second performance metric of the communication network according to the trend.

15. The method of claim 8 , wherein the plurality of nodes includes one or more balloons.

16. A system comprising:

a memory storing policies for a communication network, each policy being a set of features for forming one or more communication links in the communication network over a given time interval and being associated with a corresponding utility value, each utility value being a function of a performance metric of the communication network for the given time interval;

one or more processors capable of accessing the memory, the one or more processors being configured to:

receive input data related to a state of the communication network;

determine a training policy based on the input data, the training policy being a set of features for forming one or more first communication links in the communication network over a first time interval;

simulate the training policy over the first time interval based on internal and external influences of the communication network;

determine a training utility value of the training policy as a function of a training performance metric of the communication network in the simulation;

determine a first implementation policy based at least in part on the training utility value associated with the training policy, the first implementation policy being a set of features for forming one or more second communication links in the communication network over a second time interval, the one or more second communication links including a given communication link between a terrestrial terminal and a high-altitude platform terminal; and

transmit instructions to one or more nodes of the communication network, the instructions being configured to came the one or more nodes of the communication network to implement the first implementation policy in the second time interval.

17. The system of claim 16 , wherein the training policy is determined based on a trend in features of the policies stored in the memory that maximizes the training performance metric of the communication network.

18. The system of claim 16 , wherein the one or more processors are further configured to store the training policy in the memory in association with the determined utility value.

19. The system of claim 18 , wherein the first implementation policy is determined based on a trend in features of the policies stored in the memory that maximizes the performance metric of the communication network.

20. The system of claim 16 , wherein the one or more processors form a neural network.