# US 2019/0378005 A1 · Continuous learning of simulation trained deep neural network model

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0378005 A1
- Document ID: US20190378005A1
- Status: Published application
- Kind: A1
- Application number: 16/003,443
- Filed: 2018-06-08
- Published: 2019-12-12
- Earliest claimed: 2018-06-08
- Assignees: Technip France
- Applicants: Technip France
- Inventors: James Francis O'SULLIVAN; Djoni Eka SIDARTA; Ho Joon LIM
- Web page: https://patentgrep.com/patent/US20190378005A1

## CPC

- G06N 3/08: Learning methods
- B63B 21/50: Anchoring arrangements {or methods} for special vessels, e.g. for floating drilling platforms or dredgers
- B63B 71/10: using computer simulation, e.g. finite element method [FEM] or computational fluid dynamics [CFD]
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/0499: Feedforward networks
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G06F 30/20: Design optimisation, verification or simulation (optimisation, verification or simulation of circuit designs G06F30/30)
- G06N 3/02: Neural networks
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- B63B: SHIPS OR OTHER WATERBORNE VESSELS; EQUIPMENT FOR SHIPPING (arrangements of vessel ventilation, heating, cooling, or air-conditioning B63J2/00; floating substructures as supports of dredgers or soil-shifting machines E02F9/06)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

The present invention provides a system and method of side-stepping the need to retrain neural network model after initially trained using a simulator by comparing real-world data to data predicted by the simulator for the same inputs, and developing a mapping correlation that adjusts real world data toward the simulation data. Thus, the decision logic developed in the simulation-trained model is preserved and continues to operate in an altered reality. A threshold metric of similarity can be initially provided into the mapping algorithm, which automatically adjusts real world data to adjusted data corresponding to the simulation data for operating the neural network model when the metric of similarity between the real world data and the simulation data exceeds the threshold metric. Updated learning can continue as desired, working in the background as conditions are monitored.

## Claims

### Claim 1

A method of operating a neural network model, comprising:
providing simulation data to the neural network model, the simulation data including at least one simulated value of a variable and at least one simulation result corresponding to the at least one simulated value of the variable;
providing decision logic to the neural network model to process the simulation data for a neural network model output;
obtaining real world data including at least one real world value of the variable and at least one real world result corresponding to the at least one real world value of the variable;
comparing the real world data with the simulation data to determine a metric of similarity between the real world data and the simulation data;
providing the real world data as operational data for the neural network model if the metric of similarity is simulation data greater than a threshold metric to produce an output corresponding to an output with the simulation data from the neural network model;
adjusting the real world data to correspond with the simulation data and providing the adjusted data as the operational data for the neural network model if the metric is not greater than the threshold metric; and
operating the neural network model with the adjusted data that corresponds at least within the range of the simulation data to produce an output with the adjusted data corresponding to an output with the simulation data from the neural network model.

### Claim 2 (depends on 1)

The method of claim 1, wherein operating the neural network model comprises operating the neural network for monitoring a floating system.

### Claim 3 (depends on 1)

The method of claim 1, wherein the simulation data and real world data can apply to mooring lines on a floating system.

### Claim 4 (depends on 1)

The method of claim 1, wherein the simulation data include multiple variables for the simulation result.

### Claim 5 (depends on 1)

The method of claim 1, wherein providing simulation data to the neural network model comprises providing data for a floating system comprising mean floating system offset in the surge direction, mean floating system offset in the sway direction, and total mass of the floating system as variables and the floating system long drift period in the surge direction and the floating system long drift period in the sway direction as results.

### Claim 6 (depends on 5)

The method of claim 5, wherein obtaining real world data comprises obtaining the floating system long drift period in the surge direction in the real world and the floating system long drift period in the sway direction in the real world.

### Claim 7

A system comprising:
a central processing unit;
a non-transitory computer readable memory including processor executable program instructions that, when executed cause the central processing unit to perform operations comprising:
storing simulation data in memory for a neural network model, the simulation data including at least one simulated value of a variable and at least one simulation result corresponding to the at least one simulated value of the variable;
storing decision logic simulation data in the memory for the neural network model to process the simulation data;
storing real world data in the memory including at least one real world value of the variable and at least one real world result corresponding to the at least one real world value of the variable;
comparing the real world data with the simulation data to determine a metric of similarity between the real world data and the simulation data;
providing the real world data as operational data for the neural network model if the metric of similarity is greater than a threshold metric simulation data to produce an output corresponding to an output with the simulation data from the neural network model;
adjusting the real world data to correspond with the simulation data and providing the adjusted data as the operational data for the neural network model if the metric is not greater than the threshold metric; and
operating the neural network model with the adjusted data that corresponds at least within the range of the simulation data to produce an output with the adjusted data corresponding to an output with the simulation data from the neural network model.
