Control sequence generation system and methods
A model receives a target demand curve as an input and outputs an optimized control sequence that allows equipment within a physical space to be run optimally. A thermodynamic model is created that represents equipment within the physical space, with the equipment being laid out as nodes within the model according to the equipment flow in the physical space. The equipment activation functions comprise equations that mimic equipment operation. Values flow between the nodes similarly to how states flow between the actual equipment. The model is run such that a control sequence is used as input into the neural network; the neural network outputs a demand curve which is then checked against the target demand curve. Machine learning methods are then used to determine a new control sequence. The model is run until a goal state is reached.
1. A device for control sequence generation, comprising:
a processor in communication with a memory, the processor being configured to:
retrieve from a memory associated with the processor, a neural network which models a plurality of controlled devices;
retrieve from the memory associated with the processor, a desired demand curve;
receive a simulated control sequence that comprises instructions to control the plurality of controlled devices over time for a controlled building zone;
run the neural network using the simulated control sequence or a next simulated control sequence as input and receiving a simulated demand curve as output, wherein the simulated demand curve is a time series of zone energy inputs;
compute a cost function using the simulated demand curve and the desired demand curve, wherein computing the cost function comprises determining a difference between the desired demand curve and the simulated demand curve and wherein computing the cost function further comprises performing automatic differentiation backward through the neural network producing the next simulated control sequence;
iteratively execute the running computing, and using steps until a goal state is reached;
determine that the next simulated control sequence is a generated control sequence upon the goal state being reached; and
use at least a portion of the generated control sequence to run at least one of the plurality of controlled devices to adjust an environmental state of the controlled building zone.
2. The device of claim 1 , wherein the cost function comprises a next time series control sequence of operation error, a next time series output path error, an energy cost, a state cost, a short cycling frequency, a frequency of sequence changes, a equipment life cost, or a comfort value.
3. The device of claim 1 , wherein the neural network is a heterogenous physics network.
4. The device of claim 3 , wherein the heterogenous physics network comprises nodes that comprise equations that model thermodynamic behavior of equipment.
5. The device of claim 4 , wherein the nodes are laid out with reference to the plurality of controlled devices.
6. The device of claim 5 , wherein at least one node has multiple variables associated with it.
7. The device of claim 6 , wherein at least two nodes have different activation functions.
8. The device of claim 7 , wherein at least one activation function comprises multiple equations.
9. The device of claim 1 , wherein a node within the neural network accepts a time series curve representing weather as input.
10. The device of claim 1 , wherein the plurality of controllable devices comprise HVAC devices.
11. A method executed by a processor of control sequence generation, comprising:
retrieving from a memory associated with the processor, a neural network of which models a plurality of controllable devices;
retrieving from the memory associated with the processor, a desired demand curve;
receiving a simulated control sequence that comprises instructions to control the plurality of controllable devices over time for a building zone;
running the neural network using the simulated control sequence or a next simulated control sequence as input and receiving a simulated demand curve as output, wherein the simulated demand curve is a time series of zone energy inputs;
computing a cost function using the simulated demand curve and the desired demand curve, wherein computing the cost function comprises determining a difference between the desired demand curve and the simulated demand curve and wherein computing the cost function further comprises performing automatic differentiation backward through the neural network producing the next simulated control sequence;
iteratively executing the running, computing, and using steps until a goal state is reached;
determining that the next simulated control sequence is a generated control sequence upon the goal state being reached; and
using at least a portion of the generated control sequence to run at least one of the plurality of controllable devices to adjust an environmental state of the controlled building zone.
12. The method of claim 11 , wherein the goal state comprises the cost function being minimized, the neural network running for a specific time, or the neural network running a specific number of iterations.
13. The method of claim 11 , wherein the neural network comprises multiple nodes, at least two of the multiple nodes having different activation functions, and wherein at least one activation function has multiple variables that are passed to a different node.
14. The method of claim 11 , wherein the plurality of controllable devices comprise HVAC devices.
15. The method of claim 11 , wherein the neural network comprises nodes that comprise activation functions that model thermodynamic behavior of equipment.
16. The method of claim 14 , wherein the nodes are laid out with reference to connections between the plurality of controllable devices.
17. A non-transitory computer-readable storage medium configured with instructions executable by processor for creation of a desired control sequence from a desired demand curve, the non-transitory computer-readable storage medium comprising:
instructions for retrieving from a memory associated with the processor, a neural network which models a plurality of controllable devices;
instructions for retrieving from the memory associated with the processor, a desired demand curve;
instructions for receiving a simulated control sequence that comprises instructions to control the plurality of controllable devices over time for a controllable building zone;
instructions for running the neural network using the simulated control sequence or a next simulated control sequence as input and receiving a simulated demand curve as output, wherein the simulated demand curve is a time series of zone energy inputs;
instructions for computing a cost function using the simulated demand curve and the desired demand curve, wherein computing the cost function comprises determining a difference between the desired demand curve and the simulated demand curve and wherein computing the cost function further comprises performing automatic differentiation backward through the neural network producing the next simulated control sequence;
instructions for iteratively executing the running, computing, and using steps until a goal state is reached;
instructions for determining that the next simulated control sequence is a desired time series control sequence upon the goal state being reached; and
instructions for using at least a portion of the desired time series control sequence to run at least one of the plurality of controllable devices to adjust an environmental state of the controlled building zone.
18. The non-transitory computer-readable storage medium of claim 17 , wherein the neural network comprises multiple nodes, at least two of the multiple nodes having different activation functions, and wherein at least one activation function has multiple variables that are passed to a different node.
19. The non-transitory computer-readable storage medium of claim 18 , wherein each of at least two nodes in the multiple nodes location in the neural network correspond to each of at least two of the controllable devices.
20. The non-transitory computer-readable storage medium of claim 19 , further comprising the each of at least two nodes are arranged in the neural network equivalently to an arrangement of connections of the each of the controllable devices.