Granted Patent B2

US 12,680,715 · App. 17/991,438 · Granted Jul 14, 2026

Inventors: Robert D. Turney (Watertown, WI); Henry O. Marcy, V (Milwaukee, WI)

Assignee: Tyco Fire & Security GmbH

F24F11/63

Abstract

Claims (33)

1 . A method for controlling equipment to affect a variable state or condition of a space, comprising:

training weights of a neural network by performing a training process comprising outputting simulated control dispatches from the neural network based on simulated inputs to the neural network, simulating costs predicted to result from operating the equipment in accordance with the simulated control dispatches, and using the simulated costs as input to adjust the weights to reduce the simulated costs;

generating a control dispatch for the equipment by applying a measurement relating to the space as an input to the neural network; and

controlling the equipment in accordance with the control dispatch.

2 . The method of claim 1 , wherein the training of the weights is performed by a first computing system and the generating of the control dispatch is performed by a second computing system.

3 . The method of claim 2 , wherein the first computing system is a cloud computing system and wherein the second computing system is a building edge controller.

4 . The method of claim 2 , further comprising communicating the weights from the first computing system to the second computing system.

5 . The method of claim 1 , wherein the simulating of the costs comprises modeling equipment performance predicted to result from the simulated control dispatches using a model distinct from the neural network.

6 . The method of claim 1 , wherein the simulated inputs comprise simulated values of the variable state or condition of the space and simulated utility rate information.

7 . The method of claim 1 , wherein the simulating of the costs comprises:

identifying a state-space thermal model for the space;

defining a cost function using the state-space thermal model; and

calculating, using the cost function, the costs predicted to result from operating the equipment over a simulated time period in accordance with the simulated control dispatches from the neural network.

8 . The method of claim 1 , wherein the control dispatch comprises one or more of a temperature setpoint, temperature schedule, humidity setpoint, airflow setpoint, power level, on/off setting, damper position, fan speed, compressor frequency, or resource consumption allocation.

9 . A system for controlling building equipment, comprising:

a first computing system programmed to train weights of a neural network by performing a training process comprising outputting simulated control dispatches for the building equipment from the neural network, generating simulated costs predicted to result from operating the building equipment in accordance with the simulated control dispatches, and using the simulated costs as input to adjust the weights to reduce the simulated costs; and

a second computing system programmed to receive the weights from the first computing system, generate an online control dispatch for the building equipment using the neural network, and control the building equipment in accordance with the online control dispatch.

10 . The system of claim 9 , wherein the first computing system is remote from the second computing system and the second computing system is co-located with the building equipment.

11 . The system of claim 9 , wherein the first computing system is a cloud computing system and the second computing system is a building edge controller.

12 . The system of claim 9 , wherein the second computing system is configured to receive a stream of measurements relating to the building equipment, preprocess the stream of measurements to generate inputs for the neural network, and generate the online control dispatch for the equipment by applying the inputs to the neural network.

13 . The system of claim 9 , further comprising a sensor communicable with the second computing system, wherein the second computing system is configured use a measurement from the sensor as an input to the neural network, and wherein the first computing system is configured to train the weights of the neural network without using the measurement from the sensor.

14 . The system of claim 9 , wherein the first computing system is configured to generate the simulated costs predicted to result from operating the building equipment in accordance with the simulated control dispatches by applying the simulated control dispatches as inputs to a model other than the neural network.

15 . A building system, comprising:

building equipment comprising a local controller configured to control the building equipment to affect a variable state or condition of a building by generating control dispatches for the building equipment using a neural network; and

a computing system separate from the local controller and configured to train the neural network by:

providing simulated inputs to the neural network to obtain simulated control dispatches;

modeling simulated costs predicted to result from the simulated control dispatches; and

using the simulated costs as input to adjust the neural network in a manner that reduces the simulated costs.

16 . The building system of claim 15 , wherein the computing system is configured to reformat the neural network for execution on the local controller and provide the reformatted neural network to the local controller.

17 . The building system of claim 15 , wherein the computing system separate from the local controller is remote from the building equipment and the local controller.

18 . The building system of claim 15 , wherein the computing system separate from the local controller has more processing power and memory than the local controller.

19 . The building system of claim 15 , wherein the building equipment is heating, ventilation, or air conditioning equipment.

20 . The building system of claim 15 , wherein the modeling of the simulated costs comprises using a model other than the neural network to predict the simulated costs based on the simulated control dispatches.

Assignments (2)

Continuity (4)

References Cited (103)