Determining demand curves from comfort curves

Abstract

The amount of state over time (demand curves) that needs to be injected into a structure over time to achieve desired state values over time (desired comfort curves) at locations are determined by using a neural network that models the structure. Possibly random demand curves are fed into the neural network model at areas, such as the outside, state source locations (such as heaters), and are fed forward though the model, diffusing the state throughout the model. Comfort curves at chosen locations within the neural net representing physical locations are output. The comfort curves are compared with the desired comfort curves using cost function. Machine-learning methods are used to incrementally improve the demand curves until the output comfort curves are sufficiently close the desired state values.

We claim: 
     
         1 . A method of determining a demand curve implemented by one or more computers comprising:
 receiving a neural network of a plurality of controlled building zones;   receiving a desired comfort curve for at least one of the plurality of controlled building zones;   performing a machine learning process to run the neural network using a simulated demand curve as input and receiving a simulated comfort curve as output;   computing a cost function using the simulated comfort curve and the desired comfort curve;   using the cost function to determine a new simulated demand curve;   iteratively executing the performing, computing, and using steps until a goal state is reached; and   determining that the new simulated demand curve is the demand curve upon the goal state being reached.   
     
     
         2 . The method of  claim 1 , wherein the simulated demand curve is a time series of zone energy inputs and the simulated comfort curve is a time series of zone state values. 
     
     
         3 . The method of  claim 2 , wherein computing the cost function further comprises determining difference between the desired comfort curve and the simulated comfort curve. 
     
     
         4 . The method of  claim 2 , wherein performing the machine learning process further comprises performing automatic differentiation backward through the neural network producing a new simulated demand curve. 
     
     
         5 . The method of  claim 1 , 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. 
     
     
         6 . The method of  claim 5 , wherein the neural network comprises multiple activation functions within its neurons. 
     
     
         7 . The method of  claim 6 , wherein performing the machine learning process comprises computing a gradient of the neural network by using automatic differentiation. 
     
     
         8 . The method of  claim 7 , wherein performing the machine learning process further comprises optimizing the simulated demand curve using gradient descent, stochastic gradient descent, or mini-batch gradient descent. 
     
     
         9 . The method of  claim 1 , wherein the neural network is a heterogeneous neural network. 
     
     
         10 . The method of  claim 9 , wherein any neuron may be an output neuron. 
     
     
         11 . A demand curve creation system, the system comprising: a processor, a memory in operable communication with the processor, and demand curve creation code residing in memory which comprises: receiving a neural network of a plurality of controlled building zones;
 receiving a ground truth comfort curve for at least one of the plurality of controlled building zones;   performing a machine learning process to run the neural network using a simulated demand curve as input and receiving a simulated comfort curve as output;   computing a cost function using the simulated comfort curve and the ground truth comfort curve;   using the cost function to determine a new simulated demand curve;   iteratively executing the performing the machine learning process, computing the cost function, and using the cost function steps until a goal state is reached; and   determining that the simulated demand curve is the demand curve upon the goal state being reached.   
     
     
         12 . The demand curve creation system of  claim 11 , wherein the machine learning process comprises using backpropagation that computes a cost function gradient for values in the neural network, and then uses an optimizer to update the demand curve. 
     
     
         13 . The demand curve creation system of  claim 12 , wherein the backpropagation that computes a cost function gradient uses automatic differentiation. 
     
     
         14 . The demand curve creation system of  claim 12 , wherein the optimizer uses stochastic gradient descent or mini-batch gradient descent to minimize the cost function. 
     
     
         15 . The demand curve creation system of  claim 11 , wherein the neural network is a heterogenous neural network. 
     
     
         16 . The demand curve creation system of  claim 15 , wherein the heterogenous neural network comprises neurons, any of which may be inputs. 
     
     
         17 . The demand curve creation system of  claim 16 , wherein any of the neurons may be outputs. 
     
     
         18 . A computer-readable storage medium configured with executable instructions to perform a method for creation of a demand curve upon receipt of a comfort curve, the method comprising:
 receiving a ground truth comfort curve for at least one of a plurality of controlled building zones;   performing a machine learning process to run a heterogenous neural network using a simulated demand curve as input and receiving a simulated comfort curve as output;   computing a cost function using the simulated comfort curve and the ground truth comfort curve;   using the cost function to determine a new simulated demand curve;   iteratively executing the performing the machine learning process, computing the cost function, and using the cost function steps until a goal state is reached; and   determining that the simulated demand curve is the demand curve upon the goal state being reached.   
     
     
         19 . The computer-readable storage medium of  claim 18 , wherein the machine learning process comprises using automatic differentiation to perform backpropagation. 
     
     
         20 . The computer-readable storage medium of  claim 19 , wherein the machine learning process comprises using a gradient descent method to perform incremental optimization of the simulated demand curve.