Artificial intelligence watchdog for distributed system synchronization

A method of using artificial intelligence for maintaining synchronization between centralized and distributed train control models, the method comprising:

• providing a centralized or cloud-based computer processing system in one or more of a back-office server or a plurality of servers remote from a train;
• providing one or more distributed, edge-based computer processing systems on-board one or more locomotives of the train, wherein each of the distributed computer processing systems is communicatively connected to the centralized computer processing system;
• receiving, at a data acquisition hub communicatively connected to one or more of databases and a plurality of sensors associated with one or more locomotives or other components of a train, real-time and historical configuration, structural, and operational data in association with inputs derived from real time and historical contextual data relating to a plurality of trains operating under a variety of different conditions for use as training data;
• creating, using a centralized virtual system modeling engine included in the centralized computer processing system, one or more centralized models of one or more actual train control systems in operation on-board the one of more locomotives of the train based at least in part on data received from the data acquisition hub, wherein a first one of the centralized models is utilized in a process of generating a first set of output control commands for a first train control scenario to be implemented by an energy management system associated with the one or more locomotives;
• creating, using one or more distributed virtual system modeling engines included in the one or more distributed computer processing systems, one or more edge-based models of one or more actual train control systems in operation on-board the one or more locomotives of the train based at least in part on data received from the data acquisition hub, wherein a first one of the edge-based models is utilized in a process of generating a second set of output control commands for a second train control scenario to be implemented by an energy management system associated with the one or more locomotives;
• receiving the training data from the data acquisition hub by a machine learning engine included in at least one of the centralized and distributed computer processing systems;
• receiving the first centralized model from the centralized virtual system modeling engine by the machine learning engine;
• receiving the first edge-based model from one of the distributed virtual system modeling engines by the machine learning engine;
• comparing, using the machine learning engine, the first set of output control commands generated by the first centralized model for the first train control scenario and the second set of output control commands generated by the first edge-based model for the second train control scenario;
• training, using the machine learning engine, a learning system using the training data to enable the machine learning engine to safely mitigate a divergence discovered between the first and second sets of output control commands using a learning function including at least one learning parameter, wherein training the learning system includes:
• providing the training data as an input to the learning function, the learning function being configured to use the at least one learning parameter to generate an output based on the input;
• causing the learning function to generate the output based on the input;
• comparing the output to one or more of the first and second sets of output control commands to determine a difference between the output and the one or more of the first and second sets of output control commands;
• and
• modifying the at least one learning parameter and the output of the learning function to decrease the difference responsive to the difference being greater than a threshold difference and based at least in part on actual real time and historical information on in-train forces and train operational characteristics acquired from a plurality of trains operating under a variety of different conditions;
• and
• adjusting one or more of throttle requests, dynamic braking requests, and pneumatic braking requests for the one or more locomotives of the train using an energy management system associated with the one or more locomotives of the train, wherein the adjusting is based at least in part on the modified output of the learning function used by the learning system which has been trained by the machine learning engine.

A locomotive control system, comprising:

• a centralized or cloud-based computer processing system located in one or more of a back-office server or a plurality of servers remote from a train;
• one or more distributed, edge-based computer processing systems located on-board one or more locomotives of the train, wherein each of the distributed computer processing systems is communicatively connected to the centralized computer processing system;
• a data acquisition hub communicatively connected to one or more of databases and a plurality of sensors associated with the one or more locomotives or other components of the train and configured to acquire real-time and historical configuration, structural, and operational data in association with inputs derived from real time and historical contextual data relating to a plurality of trains operating under a variety of different conditions for use as training data;
• a centralized virtual system modeling engine included in the centralized computer processing system and configured to create one or more centralized models of one or more actual train control systems in operation on-board the one of more locomotives of the train based at least in part on data received from the data acquisition hub, wherein a first one of the centralized models is utilized in a process of generating a first set of output control commands for a first train control scenario to be implemented by an energy management system associated with one or more of the locomotives;
• one or more distributed virtual system modeling engines included in one or more of the distributed computer processing systems, each of the one or more distributed virtual system modeling engines being configured to create one or more edge-based models of one or more actual train control systems in operation on-board the one or more locomotives of the train based at least in part on data received from the data acquisition hub, wherein a first one of the edge-based models is utilized in a process of generating a second set of output control commands for a second train control scenario to be implemented by an energy management system associated with the one or more of the locomotives;
• a machine learning engine included in at least one of the centralized and distributed computer processing systems, the machine learning engine being configured to:
• receive the training data from the data acquisition hub;
• receive the first set of output control commands from the centralized virtual system modeling engine;
• receive the second set of output control commands from one of the distributed virtual system modeling engines;
• compare the first set of output control commands and the second set of output control commands;
• and
• train a learning system using the training data to enable the machine learning engine to safely mitigate any difference between the first and second sets of output control commands using a learning function including at least one learning parameter, wherein training the learning system includes:
• providing the training data as an input to the learning function, the learning function being configured to use the at least one learning parameter to generate an output based on the input;
• causing the learning function to generate the output based on the input;
• comparing the output to one or more of the first and second sets of output control commands to determine a difference between the output and the one or more of the first and second sets of output control commands;
• and
• modifying the at least one learning parameter and the output of the learning function to decrease the difference responsive to the difference being greater than a threshold difference;
• and
• an energy management system associated with the one or more locomotives of the train and configured to adjust one or more of throttle requests, dynamic braking requests, and pneumatic braking requests for the one or more locomotives of the train based at least in part on the modified output of the learning function.