Neural network initialization
A neural network representing a controlled space can be initialized by collecting state time series data that affects the controlled space such as weather, and also collecting sensor data from the controlled space at the same time. The time series data is used as input to a neural network that models the controlled space until an area in the neural network equivalent to the sensor is at or near the sensor state at a given time.
1 . A computer-implemented method for initializing and operating a thermodynamic neural network representing a building, the method comprising:
receiving, from at least one physical sensor located within the building, sensor values representing a state of a portion of the building;
receiving a sequence of time state values;
storing the sequence of time state values in a memory;
receiving a sensor value of the portion of the building;
storing the sensor value in the memory;
executing, by a processor, a neural network stored in the memory, the neural network comprising a first neuron and a second neuron, the first neuron representing a first structural component of the building and the second neuron representing a second structural component thermally coupled to the first structural component, each neuron comprising an activation function, the activation function comprising a permanent variable representing a physical property of a building structure, and a transient variable representing a thermodynamic state of the building;
propagating the sequence of time state values through the neural network over the sequence of time state values;
computing, by the activation function of the first neuron, a value of the transient variable of the first neuron using at least one of the sequence of time state values;
computing, by the activation function of the second neuron, a value of the transient variable of the second neuron using output from the first neuron;
iteratively updating the transient variable of the second neuron for successive time values until the transient variable corresponds to the sensor value;
after the transient variable corresponds to the sensor value, generating a thermodynamic state model of the portion of the building; and
determining an operating state of a heating or cooling system of the building based on the thermodynamic state model.
2 . The method of claim 1 , further comprising causing operation of a heating or cooling system of the building based on the thermodynamic state.
3 . The method of claim 1 , wherein the permanent variable represents a physical property of a building structure, and a transient variable represents a thermodynamic state.
4 . The method of claim 1 , wherein the permanent variable represents a physical property comprising heat capacity, thermal resistance, or heat transfer rate.
5 . The method of claim 4 , wherein the first structural component comprises a wall and the second structural component comprises a room.
6 . The method of claim 1 , wherein the transient variable represents temperature within the portion of the building.
7 . The method of claim 1 , further comprising determining an operating state for a heating system using the transient variable of the second neuron.
8 . The method of claim 7 , further comprising determining an operating state for a heating system using the heating requirement.
9 . The method of claim 1 , wherein activation functions in different neurons represent different sets of equations.
10 . The method of claim 9 , wherein the activation functions in different neurons model different materials.
11 . A system for modeling thermodynamic behavior of a building, comprising:
a processor;
a memory in operational communication with the processor; and
a neural network stored in the memory, the neural network comprising a plurality of neurons including a first neuron and a second neuron, the first neuron representing a first structural component of the building and the second neuron representing a second structural component thermally coupled to the first structural component, each neuron comprising an activation function, the activation function comprising a permanent variable representing a physical property of a building structure, and a transient variable representing a thermodynamic state;
wherein the processor is in communication with the memory configured to:
receive a value comprising a sensed state of the building;
receive values representing state;
propagate the values representing state through the neural network, comprising computing transient variable values of the neurons using the activation functions;
after the transient variable corresponds to the sensed state, operating the neural network to determine a thermodynamic state of a portion of the building;
determine a thermodynamic state of the portion of the building using the transient variable of the second neuron; and
determine an operating state of a heating or cooling system of the building based on the thermodynamic state.
12 . The system of claim 11 , wherein the thermodynamic state comprises temperature of the portion of the building.
13 . The system of claim 11 , wherein the control signal controls operation of a heating system or cooling system.
14 . The system of claim 11 , wherein the first neuron and the second neuron are arranged in the neural network to correspond spatially to structural components of the building.
15 . The system of claim 11 , wherein the neural network models thermodynamic heat transfer between rooms and walls of the building.
16 . A non-transitory computer-readable storage medium storing instructions which, when executed by a processor, cause the processor to:
execute, by the processor, a neural network stored in memory, the neural network comprising a first neuron and a second neuron, the first neuron representing a first structural component of the building and the second neuron representing a second structural component thermally coupled to the first structural component, each neuron comprising an activation function, the activation function comprising a permanent variable representing a physical property of a building structure, and a transient variable representing a thermodynamic state of a building;
receive a value comprising a sensed state of a building;
receive values representing state;
propagate the values representing state through a neural network, comprising computing transient variable values of the neurons using the activation functions;
after the transient variable corresponds to the sensed state, operating the neural network to determine a thermodynamic state of a portion of the building;
determine a thermodynamic state of the portion of the building using the transient variable of the second neuron; and
determine an operating state of a heating or cooling system of the building based on the thermodynamic state.
17 . The non-transitory computer-readable storage medium of claim 16 , wherein the first neuron and the second neuron are arranged in the neural network to correspond spatially to structural components of the building.
18 . The non-transitory computer-readable storage medium of claim 16 , wherein a transient state value of at least one neuron is captured as output.
19 . The non-transitory computer-readable storage medium of claim 16 , wherein the activation functions in different neurons represent different sets of equations.
20 . The non-transitory computer-readable storage medium of claim 19 , wherein the activation functions in different neurons model different materials.
21 . The non-transitory computer-readable storage medium of claim 16 , wherein the neurons further comprise permanent state values, and wherein the permanent state values model physical features of an object represented by the neuron.