Environment controller and method for generating a predictive model of a neural network through distributed reinforcement learning

Patent metadata

Field	Value
Publication number	US-11460209-B2
Application number	US-201916697886-A
Country	US
Kind code	B2
Filing date	Nov 27, 2019
Priority date	Aug 26, 2019
Publication date	Oct 4, 2022
Grant date	Oct 4, 2022

Abstract

Interactions between a training server and a plurality of environment controllers are used for updating the weights of a predictive model used by a neural network executed by the plurality of environment controllers. Each environment controller executes the neural network using a current version of the predictive model to generate outputs based on inputs, modifies the outputs, and generates metrics representative of the effectiveness of the modified outputs for controlling the environment. The training server collects the inputs, the corresponding modified outputs, and the corresponding metrics from the plurality of environment controllers. The collected inputs, modified outputs and metrics are used by the training server for updating the weights of the current predictive model through reinforcement learning. A new predictive model comprising the updated weights is transmitted to the environment controllers to be used in place of the current predictive model.

First claim

What is claimed is:

1. An environment controller comprising:
at least one communication interface;
memory for storing a predictive model comprising weights of a neural network; and
a processing unit comprising one or more processor configured to:
(a) determine at least one environmental characteristic value in an area;
(b) receive at least one set point via one of the at least one communication interface or a user interface of the environment controller;
(c) execute a neural network inference engine using the predictive model for generating one or more output based on inputs, the one or more output comprising one or more command for controlling a controlled appliance, the inputs comprising the at least one environmental characteristic value in the area and the at least one set point;
(d) modify the one or more command;
(e) transmit the one or more modified command to the controlled appliance via the at least one communication interface;
(f) generate at least one metric representative of an execution of the one or more modified command by the controlled appliance;
(g) transmit the inputs, the one or more output and the at least one metric to a training server via the at least one communication interface; and
(h) receive an update of the predictive model comprising updated weights from the training server via the at least one communication interface.

2. The environment controller of claim 1 , wherein steps (a) to (g) are repeated a plurality of times before the occurrence of step (h).

3. The environment controller of claim 1 , wherein modifying the one or more command comprises applying a random modification to the one or more command.

4. The environment controller of claim 1 , wherein modifying the one or more command comprises applying a modification selected among a set of one or more pre-defined modification to the one or more command.

5. The environment controller of claim 1 , wherein the modification of the one or more command is performed based on configuration data included in a configuration message received via the at least one communication interface.

6. The environment controller of claim 1 , wherein the at least one metric comprises at least one updated environmental characteristic value in the area determined by the processing unit after the transmission of the one or more modified command to the controlled appliance.

7. The environment controller of claim 1 , wherein the at least one metric comprises at least one measurement by the processing unit of a time required for reaching at least one corresponding environmental state in the area after the transmission of the one or more modified command to the controlled appliance.

8. The environment controller of claim 1 , wherein the at least one environmental characteristic value in the area comprises at least one of the following: a current temperature in the area, a current humidity level in the area, a current carbon dioxide (CO2) level in the area, and a current occupancy of the area.

9. The environment controller of claim 1 , wherein the at least one environmental characteristic value in the area comprises at least one of the following: a plurality of consecutive temperature measurements in the area, a plurality of consecutive humidity level measurements in the area, a plurality of consecutive carbon dioxide (CO2) level measurements in the area, and a plurality of consecutive determinations of an occupancy of the area.

10. The environment controller of claim 1 , wherein determining the environmental characteristic value in the area consists in receiving the environmental characteristic value from a sensor via the at least one communication interface.

11. The environment controller of claim 1 , wherein the at least one set point comprises at least one of the following: a target temperature, a target humidity level, and a target CO2 level.

12. The environment controller of claim 1 , wherein the one or more processor is further configured to determine at least one characteristic of the area; and the inputs further include the at least one characteristic of the area.

13. The environment controller of claim 12 , wherein the at least one characteristic of the area comprises at least one of the following: an area type identifier selected among a plurality of area type identifiers, one or more geometric characteristics of the area, and a human activity in the area.

14. The environment controller of claim 1 , wherein the area is a room inside a building.

15. The environment controller of claim 1 , wherein the controlled appliance consists of a Variable Air Volume (VAV) appliance.

16. The environment controller of claim 1 , wherein the one or more command includes at least one of the following: a command for controlling a speed of a fan, a command for controlling a pressure generated by a compressor, and a command for controlling a rate of an airflow through a valve.

17. The environment controller of claim 1 , wherein the neural network inference engine implements the neural network corresponding to the predictive model, the neural network comprising an input layer for receiving the inputs, followed by one or more intermediate hidden layers, followed by an output layer for outputting the one or more output.

18. A method for improving a predictive model of a neural network used for performing environment control, the method comprising:
storing in a memory of a computing device a predictive model comprising weights of a neural network;
(a) determining by a processing unit of the computing device at least one environmental characteristic value in an area;
(b) receiving at least one set point via one of a communication interface of the computing device or a user interface of the computing device;
(c) executing by the processing unit of the computing device a neural network inference engine using the predictive model for generating one or more output based on inputs, the one or more output comprising one or more command for controlling a controlled appliance, the inputs comprising the at least one environmental characteristic value in the area and the at least one set point;
(d) modifying by the processing unit of the computing device the one or more command;
(e) transmitting the one or more modified command to the controlled appliance via the communication interface of the computing device;
(f) generating by the processing unit of the computing device at least one metric representative of an execution of the one or more modified command by the controlled appliance;
(g) transmitting the inputs, the one or more output and the at least one metric to a training server via the communication interface; and
(h) receiving an update of the predictive model comprising updated weights from the training server via the communication interface.

19. The method of claim 18 , wherein the computing device is an environment controller.

20. The method of claim 18 , wherein steps (a) to (g) are repeated a plurality of times before the occurrence of step (h).

21. The method of claim 18 , wherein modifying the one or more command comprises applying a random modification to the one or more command.

22. The method of claim 18 , wherein modifying the one or more command comprises applying a modification selected among a set of one or more pre-defined modification to the one or more command.

23. The method of claim 18 , wherein the modification of the one or more command is performed based on configuration data include

Assignees

• Distech Controls Inc

Inventors

• Lupien Steve
• Gervais Francois

Classifications

• G06N3/092

  Reinforcement learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/098

  Distributed learning, e.g. federated learning · CPC title

• G06N5/04

  Inference or reasoning models · CPC title

• F24F2110/20

  Humidity · CPC title