Neural network methods for defining system topology

Patent metadata

Field	Value
Publication number	US-12572807-B2
Application number	US-202017009713-A
Country	US
Kind code	B2
Filing date	Sep 1, 2020
Priority date	Jun 5, 2020
Publication date	Mar 10, 2026
Grant date	Mar 10, 2026

Abstract

A neural network in one embodiment is built by decomposing a structure into different building materials creating neurons that represent building materials and open spaces in a structure. Subsystems in the building have their neurons concatenated together to create same length neuron strings. In some embodiments, neurons in a short neuron string are split to make longer neuron strings. In some embodiments, neurons are added to some neuron strings to represent inside features, air features, and outside features.

First claim

We claim:

1 . A neural network creation system, the system comprising: a memory and a vector processor configured to:
determine structure elements for at least a first subsystem and a second subsystem in a digital representation of a structure; build a neuron for each structure element in the first subsystem and the second subsystem, each neuron comprising: an input, an output, and a parameter; build extra neurons for each subsystem that has fewer structure elements than a subsystem with a greatest number of neurons, such that each has an equal number of neurons; connect the neurons associated with the first subsystem making a first neuron string; and connect the neurons associated with the second subsystem making a second neuron string, the input of each next one of the neurons in a neuron string being connected to the output of a preceding one of the neurons to create a neural network; assign a value to the parameter in at least one neuron, the value representing a thermodynamic aspect of an associated structure element the neuron represents; such that as the neural network is run, the parameter value changes, reflecting thermodynamic aspects of the structure element; use changes of the parameter value to determine a control setting on a piece of equipment within the structure; and such that the first neuron string and the second neuron string can be run simultaneously on the vector processor.

2 . The system of claim 1 , further comprising building an inner neuron which comprises representation of state within the structure, and connecting the first neuron string to the inner neuron.

3 . The system of claim 2 , wherein the structure comprises a first room and a second room, a set of neuron strings is associated with the first room and a set of neuron strings is associated with the second room, a wall is shared between the first room and the second room, the shared wall is represented by one wall string in the set of neuron strings associated with the first room, the shared wall is represented by no strings in the set of neuron strings associated with the second room, and wherein n inner neuron associated with the second neuron string is connected to the wall.

4 . The system of claim 1 , wherein the parameter comprises resistance or capacitance.

5 . The system of claim 2 , further comprising building an outside neuron which comprises representation of air state outside the structure, and wherein the outside neuron is connected to the first neuron string.

6 . The system of claim 2 , wherein representation of state within the structure comprises a calculation comprising air volume, furniture volume, heat value related to number of people in the structure, or heat value related to lighting in the structure.

7 . The system of claim 1 , further comprising determining the value of at least one neuron parameter for at least one neuron; and
reporting the value using an output device.

8 . The system of claim 1 , further comprising the memory and processor further configured to send an instruction to the piece of equipment.

9 . A method performed by a vector processor for creation of neural networks, the method comprising:
determining structure elements for at least a first subsystem and a second subsystem in a digital representation of a structure; building a neuron for each structure element in the first subsystem and the second subsystem, each neuron comprising: an input, an output, and a parameter; building extra neurons for each subsystem that has fewer structure elements than a subsystem with a greatest number of structure elements, such that each subsystem has an equal number of neurons; connecting neurons associated with the first subsystem making a first neuron string; and connecting the neurons associated with the second subsystem making a second neuron string, an input of each next one of the neurons in a neuron string being connected to output of a preceding one of the neurons to create a neural network; assigning a value to the parameter in at least one neuron, the value representing a thermodynamic aspect of an associated structure element the neuron represents; such that as the neural network is run, the parameter value changes, reflecting thermodynamic aspects of the structure element; using the change of the parameter value to determine a control setting on a piece of equipment within the structure; and such that the first neuron string and the second neuron string can be run simultaneously on the vector processor.

10 . The method of claim 9 , further comprising building an inner neuron which comprises representation of state within the structure, and connecting the first neuron string to the inner neuron.

11 . The method of claim 10 , further comprising building a ground neuron which comprises representation of ground state outside the structure, and connecting the ground neuron to a neuron string whose structure it is representing connects to ground.

12 . The method of claim 11 , further comprising building an outside neuron which comprises representation of outside state outside the structure, and connecting the outside neuron to a neuron string whose structure it is representing connects to an outside surface.

13 . The method of claim 12 , comprising building an indoor air neuron that comprises a resistance value and a capacitance value, and connecting the indoor air neuron between the outside neuron and the first neuron string.

14 . The method of claim 9 , further comprising sending an instruction to the piece of equipment.

15 . A non-transitory computer-readable storage medium configured with executable instructions to perform a method for creation and discretization of neural networks, the non-transitory computer-readable storage medium comprising:
instructions for determining structure elements for at least a first subsystem and a second subsystem in a digital representation of a structure, the first subsystem being physically next to the second subsystem; instructions for building a neuron for each structure element in the first subsystem and the second subsystem each neuron comprising: an input, an output, and a parameter; instructions for building an extra neuron for each subsystem that has fewer structure elements than a subsystem with a greatest number of structure elements, such that each subsystem has an equal number of structure elements; instructions for connecting neurons associated with the first subsystem making a first neuron string; and connecting the neurons associated with the second subsystem making a second neuron string, an input of each next one of the neurons in a neuron string being connected to output of a preceding one of the neurons to create a neural network; such that as the neural network is run, a value of the parameter changes, reflecting thermodynamic aspects of an associated structure element; instructions for using the change of the parameter value to determine a control setting on a piece of equipment within the structure; and such that the first neuron string and the second neuron string can be run simultaneously on a vector processor.

16 . The non-transitory computer-readable storage medium of claim 15 , comprising instructions for connecting an inside air neuron to the first neuron string, making an air neuron string.

17 . The non-transitory computer-readable storage medium of claim 16 , further comprising instructions for connecting an inner neuron, which comprises representation of state within the structure, to the air neuron string.

18 . The non

Assignees

• Harvey Troy Aaron
• Fillingim Jeremy David
• Passivelogic Inc

Inventors

• Harvey Troy Aaron
• Fillingim Jeremy David

Classifications

• B60H1/00285

  for vehicle seats · CPC title

• G06N3/048

  Activation functions · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G05B13/04

  involving the use of models or simulators · CPC title

• F24F2140/50

  Load · CPC title