Controller, method and computer program product using a neural network for adaptively controlling an environmental condition in a building
Environmental condition controller and method for controlling an environmental condition in an area of a building. The controller stores a predictive model generated by a neural network training engine, and a previously calculated environmental condition adjustment value (y n−1 ). The controller receives an environmental condition target value (x ref ), and an environmental condition measured value (x). The controller recursively calculates an environmental condition adjustment value (y n ) by executing a neural network inference engine using the predictive model for inferring the environmental condition adjustment value (y n ) based on the previously calculated environmental condition adjustment value (y n−1 ), the environmental condition target value (x ref ), the environmental condition measured value (x), and an adaptive proportionality value (k). The controller also generates and transmits a command based on the environmental condition adjustment value (y n ). The controller further stores the calculated environmental condition adjustment value (y n ) as the previously calculated environmental condition adjustment value (y n−1 ).
1. An environmental condition controller for controlling an environmental condition in an area of a building, the environmental condition controller comprising:
memory for storing:
a predictive model generated by a neural network training engine, the predictive model comprising weights of a neural network calculated by the neural network training engine; and
a previously calculated environmental condition adjustment value (y n−1 );
a communication interface for:
receiving an environmental condition target value (x ref ); and
receiving an environmental condition measured value (x);
a processing unit for:
calculating an environmental condition adjustment value (y n ) by executing a neural network inference engine, the neural network inference engine implementing a neural network using the predictive model for inferring the environmental condition adjustment value (y n ) based on inputs, the inputs comprising the previously calculated environmental condition adjustment value (y n−1 ), the environmental condition target value (x ref ), the environmental condition measured value (x), and an adaptive proportionality value (k);
generating and transmitting a command based on the environmental condition adjustment value (y n ); and
storing in the memory the calculated environmental condition adjustment value (y n ) as a previously calculated environmental condition adjustment value.
2. The environmental condition controller of claim 1 , wherein the calculation of the environmental condition adjustment value (y n ) is based on more than one previously calculated environmental condition adjustment value used as inputs of the neural network inference engine.
3. The environmental condition controller of claim 1 , wherein the environmental condition is one of the following: temperature, pressure, humidity or lighting.
4. The environmental condition controller of claim 1 , wherein the command is one of the following: heating, ventilating, cooling, humidifying, dehumidifying or changing lighting.
5. The environmental condition controller of claim 1 , wherein the processing unit further iteratively calculates the adaptive proportionality value (k) with the following equation:
k=e CΔt c −1
where:
C is calculated using the equation C=S/(y max −y n−1 );
S is the slope between two previous environmental condition adjustment values over time, and is calculated using the equation s=dy/dt;
y max is the environmental condition maximum output; and
Δt c is a time interval between each recursive calculation of the environmental condition adjustment value (y n ).
6. The environmental condition controller of claim 5 , wherein the processing unit iteratively calculates the adaptive proportionality value (k) when at least one of the following conditions is met:
the environmental condition adjustment value (y n ) exceeds the environmental condition value (x); or
when a difference between the environmental condition value (x) and the environmental condition adjustment (y n ) is greater than a predefined tolerated variance.
7. A method for controlling an environmental condition in an area of a building, the method comprising:
storing in a memory of an environmental condition controller a predictive model generated by a neural network training engine, the predictive model comprising weights of a neural network calculated by the neural network training engine;
storing in the memory a previously calculated environmental condition adjustment value (y n−1 );
receiving an environmental condition target value (x ref ) via a communication interface of the environmental condition controller;
receiving an environmental condition measured value (x) via the communication interface;
calculating by a processing unit of the environmental condition controller an environmental condition adjustment value (y n ) by executing a neural network inference engine, the neural network inference engine implementing a neural network using the predictive model for inferring the environmental condition adjustment value (y n ) based on inputs, the inputs comprising the previously calculated environmental condition adjustment value (y n−1 ), the environmental condition target value (x ref ), the environmental condition measured value (x), and an adaptive proportionality value (k);
generating and transmitting by the processing unit a command based on the environmental condition adjustment value (y n ); and
storing in the memory the calculated environmental condition adjustment value (y n ) as a previously calculated environmental condition adjustment value.
8. The method of claim 7 , wherein the following steps are performed in a recursive loop:
receiving the environmental condition measured value (x) via the communication interface;
calculating by the processing unit the environmental condition adjustment value (y n ) by executing the neural network inference engine;
generating and transmitting by the processing unit the command based on the environmental condition adjustment value (y n ), and
storing in the memory the calculated environmental condition adjustment value (y n ) as a previously calculated environmental condition adjustment value.
9. The method of claim 8 , wherein the recursive calculation of the environmental condition adjustment value (y n ) is based on more than one previously calculated environmental condition adjustment value used as inputs of the neural network inference engine.
10. The method of claim 8 , wherein the processing unit further iteratively calculates the adaptive proportionality value (k) using the following equation:
k=e CΔt c −1
where:
C is calculated using the equation C=S/(y max −y n−1 );
S is the slope between two previous environmental condition adjustment values over time, and is calculated using the equation s=dy/dt;
y max is the environmental condition maximum output; and
Δt c is a time interval between each recursive calculation of the environmental condition adjustment value (y n ).
11. The method of claim 10 , wherein the adaptive proportionality value (k) is iteratively calculated by the processing unit when at least one of the following conditions is met:
the environmental condition adjustment value (y n ) exceeds the environmental condition value (x); or
when a difference between the environmental condition value (x) and the environmental condition adjustment (y n ) is greater than a predefined tolerated variance.
12. The method of claim 7 , wherein the environmental condition is one of the following: temperature, pressure, humidity and lighting.
13. The method of claim 7 , wherein the command is one of the following: heating, ventilating, cooling, humidifying, dehumidifying and changing lighting.
14. A non-transitory computer program product comprising instructions deliverable via an electronically-readable media, such as storage media and communication links, the instructions when executed by a processing unit of an environmental condition controller providing for controlling an environmental condition in an area of a building by:
storing in a memory of the environmental condition controller a predictive model generated by a neural network training engine, the predictive model comprising weights of a neural network calculated by the neural network training engine;
storing in the memory a previously calculated environmental condition adjustment value (y n−1 );
receiving an environmental condition target value (x ref ) via a communication interface of the environmental condition controller;
receiving an environmental condition measured value (x) via the communication interface;
calculating by the processing unit an environmental condition adjustment value (y n ) by executing a neural network inference engine, the neural network inference engine implementing a neural network using the predictive model for inferring the environmental condition adjustment value (y n ) based on inputs, the inputs comprising the previously calculated environmental condition adjustment value (y n−1 ), the environmental condition target value (x ref ), the environmental condition measured value (x), and an adaptive proportionality value (k);
generating and transmitting by the processing unit a command based on the environmental condition adjustment value (y n ); and
storing in the memory the calculated environmental condition adjustment value (y n ) as a previously calculated environmental condition adjustment value.
15. The computer program product of claim 14 , wherein the calculation of the environmental condition adjustment value (y n ) is based on more than one previously calculated environmental condition adjustment value used as inputs of the neural network inference engine.
16. The computer program product of claim 14 , wherein the environmental condition is one of the following: temperature, pressure, humidity and lighting.
17. The computer program product of claim 14 , wherein the command is one of the following: heating, ventilating, cooling, humidifying, dehumidifying and changing lighting.
18. The computer program product of claim 14 , wherein the processing unit further iteratively calculates the adaptive proportionality value (k) using the following equation:
k=e CΔt c −1
where:
C is calculated using the equation C=S/(y max −y n−1 );
S is the slope between two previous environmental condition adjustment values over time, and is calculated using the equation s=dy/dt;
y max is the environmental condition maximum output; and
Δt c is a time interval between each recursive calculation of the environmental condition adjustment value (y n ).
19. The computer program product of claim 18 , wherein the adaptive proportionality value (k) is iteratively calculated by the processing unit when at least one of the following conditions is met:
the environmental condition adjustment value (y n ) exceeds the environmental condition value (x); or
when a difference between the environmental condition value (x) and the environmental condition adjustment (y n ) is greater than a predefined tolerated variance.
20. The computer program product of claim 14 , wherein the following steps are performed in a recursive loop:
receiving the environmental condition measured value (x) via the communication interface;
calculating by the processing unit the environmental condition adjustment value (y n ) by executing the neural network inference engine;
generating and transmitting by the processing unit the command based on the environmental condition adjustment value (y n ), and
storing in the memory the calculated environmental condition adjustment value (y n ) as a previously calculated environmental condition adjustment value.