Building management system with augmented deep learning using combined regression and artificial neural network modeling

1. A method for controlling a plant, comprising:

using a neural network modeling technique to calculate a neural network prediction based on plant input data;

using a second modeling technique to calculate a second prediction based on the plant input data, wherein the second modeling technique is different than the neural network modeling technique;

determining, by comparing a location of the plant input data in a multi-dimensional modeling space to one or more thresholds, whether to use (1) the neural network prediction without the second prediction, (2) the second prediction without the neural network prediction, or (3) both the neural network prediction and the second prediction;

generating a combined prediction using one or both of the neural network prediction and the second prediction in accordance with a result of the determining; and

controlling the plant using the combined prediction.

2. The method of claim 1 , wherein the second modeling technique comprises at least one of a regression modeling technique, a physics-based modeling technique, a grey-box modeling technique, and a low-order empirical modeling technique.

3. The method of claim 1 , wherein comparing the location of the plant input data in the multi-dimensional modeling space to the one or more thresholds comprises determining a distance between the location of the plant input data and a different location in the multi-dimensional modeling space.

4. The method of claim 3 , wherein comparing the location of the plant input data in the multi-dimensional modeling space to the one or more thresholds comprises comparing the distance to a predetermined distance.

5. The method of claim 1 , wherein the one or more thresholds comprise a confidence limit associated with the neural network modeling technique.

6. The method of claim 1 , wherein comparing the location of the plant input data in the multi-dimensional modeling space to the one or more thresholds comprises determining whether the location of the plant input data is within a confidence limit associated with the neural network modeling technique; and

the combined prediction is generated using the neural network prediction without the second prediction in response to determining that the location of the plant input data is within the confidence limit.

7. The method of claim 1 , wherein comparing the location of the plant input data in the multi-dimensional modeling space to the one or more thresholds comprises determining whether the location of the plant input data is outside a confidence limit associated with the neural network modeling technique; and

the combined prediction is generated using the second prediction without the neural network prediction in response to determining that the location of the plant input data is outside the confidence limit.

8. The method of claim 1 , wherein comparing the location of the plant input data in the multi-dimensional modeling space to the one or more thresholds comprises determining whether the location of the plant input data is between a first confidence limit and a second confidence limit associated with the neural network modeling technique; and

the combined prediction is generated using both the neural network prediction and the second prediction in response to determining that the location of the plant input data is between the first confidence limit and the second confidence limit.

9. A method for controlling a plant, comprising:

using a neural network modeling technique to calculate a neural network prediction based on plant input data;

using a second modeling technique to calculate a second prediction based on the plant input data, wherein the second modeling technique is different than the neural network modeling technique;

determining a distance between a location of the plant input data in a multi-dimensional modeling space and a different location in the multi-dimensional modeling space;

calculating a combined prediction using the neural network prediction, the second prediction, and the distance; and

controlling the plant using the combined prediction.

10. The method of claim 9 , wherein the different location is a centroid of a cluster in the multi-dimensional modeling space associated with the neural network modeling technique.

11. The method of claim 9 , wherein the combined prediction is calculated as a continuous function of the neural network prediction, the second prediction, and the distance.

12. The method of claim 9 , wherein calculating the combined prediction comprises:

assigning a weight to the neural network prediction or the second prediction based on the distance; and

calculating the combined prediction as a weighted average of the neural network prediction and the second prediction using the weight.

13. The method of claim 9 , wherein the second modeling technique comprises at least one of a regression modeling technique, a physics-based modeling technique, a grey-box modeling technique, and a low-order empirical modeling technique.

14. One or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

using a neural network modeling technique to calculate a neural network prediction based on input data;

using a second modeling technique to calculate a second prediction based on the input data, wherein the second modeling technique is different than the neural network modeling technique;

determining, by comparing a location of the input data in a multi-dimensional modeling space to one or more thresholds, whether to use (1) the neural network prediction without the second prediction, (2) the second prediction without the neural network prediction, or (3) both the neural network prediction and the second prediction; and

generating a combined prediction using one or both of the neural network prediction and the second prediction in accordance with a result of the determining;

controlling a system or device using the combined prediction.

15. The non-transitory computer-readable media of claim 14 , wherein generating the combined prediction comprises:

determining a distance between the location of the input data in the multi-dimensional modeling space and a different location in the multi-dimensional modeling space; and

calculating the combined prediction using the neural network prediction, the second prediction, and the distance.

16. The non-transitory computer-readable media of claim 15 , wherein the different location is a centroid of a cluster in the multi-dimensional modeling space associated with the neural network modeling technique.

17. The non-transitory computer-readable media of claim 15 , wherein the combined prediction is calculated as a continuous function of the neural network prediction, the second prediction, and the distance.

18. The non-transitory computer-readable media of claim 14 , wherein the second modeling technique comprises at least one of a regression modeling technique, a physics-based modeling technique, a grey-box modeling technique, and a low-order empirical modeling technique.

19. The non-transitory computer-readable media of claim 14 , wherein comparing the location of the input data in the multi-dimensional modeling space to the one or more thresholds comprises:

determining a distance between the location of the input data and a different location in the multi-dimensional modeling space; and

comparing the distance to a predetermined distance.

20. The non-transitory computer-readable media of claim 14 , wherein the one or more thresholds comprise a confidence limit associated with the neural network modeling technique.