System and method for modeling and forecasting cyclical demand systems with dynamic controls and dynamic incentives

1. A system for modeling a cyclical demand system associated with an electricity grid, comprising:

a metering component in two-way communication with the electricity grid for collecting usage data; and

at least one processing device operatively coupled to a memory in communication with the metering component and configured to:

obtain, over a plurality of demand cycles, fine-grained usage data from the metering component via the two-way communication with the electricity grid over a plurality of fine-grained discrete time intervals, wherein the plurality of fine-grained discrete time intervals have durations that vary based on demand, and wherein the obtained fine-grained usage data comprises first historical data associated with one or more demand measurements over the plurality of demand cycles, second historical data associated with incentive signals over the plurality of demand cycles, and time series data for the first and second historical data;

dynamically transmit the incentive signals to a plurality of customers on the electricity grid, wherein the incentive signals comprise actual electricity pricing data and an indication that a price of electricity will be higher or lower over a given time period;

construct a dynamic linear model using the obtained fine-grained usage data, wherein the dynamic linear model is constructed as a weighted sum of a plurality of basis functions by converting the time series data into coefficients of the plurality of basis functions

wherein the constructing of the dynamic linear model as the weighted sum of the plurality of basis functions reduces dimensionality in connection with performance of computations by the at least one processing device to construct the dynamic linear model based on the plurality of fine-grained discrete time intervals;

wherein a number of the plurality of basis functions is less than a number of the plurality of fine-grained discrete time intervals;

wherein a model for the coefficients of the plurality of basis functions is dependent on the incentive signals, the incentive signals indicating dynamic pricing of the electricity over the given time period to enable the model for coefficients to capture demand shifting effects in response to the dynamic pricing;

wherein the model for the coefficients of the plurality of basis functions comprises a holiday effect;

wherein a holiday indicator is one of a binary variable and a fractional value; and

wherein in converting the time series data, the at least one processing device is configured to convert a first plurality of graphs based on the plurality of fine-grained discrete time intervals to a second plurality of graphs corresponding to the first plurality of graphs, wherein the second plurality of graphs are based on the plurality of basis functions and the number of the second plurality of graphs is less than the number of the first plurality of graphs; and

estimate a forecasted demand cycle over the plurality of fine-grained discrete time intervals by obtaining forecasts and forecast errors using the dynamic linear model;

wherein, in constructing the dynamic linear model, the at least one processing device is further configured to:

specify a state-space model and variance parameters in the state-space model;

estimate one or more unknown variance parameters by using a maximum-likelihood method to generate one or more maximum-likelihood estimates;

replace the one or more unknown variance parameters with the one or more maximum-likelihood estimates; and

generate a load curve from the second plurality of graphs based on the weighted sums of the plurality of basis functions;

wherein the forecasts and forecast errors are obtained based at least in part on the maximum-likelihood estimates.

2. The system according to claim 1 , wherein the first historical data comprises data associated with usage of a resource at regular time intervals over a given period by a plurality of consumers.

3. The system according to claim 1 , wherein the obtained fine-grained usage data further comprises calendar information over the plurality of demand cycles.

4. The system according to claim 1 , wherein the state-space model includes at least one of overall level, day-of-week effects, simultaneous-effects and shifting-effects.

5. The system according to claim 1 , wherein, in constructing the dynamic linear model, the at least one processing device is further configured to specify prior distributions for initial values of a state vector.

6. The system according to claim 1 , wherein the plurality of basis functions comprise a plurality of knots at locations corresponding to peak periods of electricity usage.

7. The system according to claim 6 , wherein the plurality of basis functions comprise periodic cubic-spline basis functions.

8. A method for modeling a cyclical demand system associated with an electricity grid, comprising:

maintaining a metering component in two-way communication with the electricity grid for collecting usage data;

obtaining, over a plurality of demand cycles, fine-grained usage data from the metering component via the two-way communication with the electricity grid over a plurality of fine-grained discrete time intervals, wherein the plurality of fine-grained discrete time intervals have durations that vary based on demand, and wherein the obtained fine-grained usage data comprises first historical data associated with one or more demand measurements over the plurality of demand cycles, second historical data associated with incentive signals over the plurality of demand cycles, and time series data for the first and second historical data;

dynamically transmitting the incentive signals to a plurality of customers on the electricity grid, wherein the incentive signals comprise actual electricity pricing data and an indication that a price of electricity will be higher or lower over a given time period;

constructing a dynamic linear model using the obtained fine-grained usage data, wherein the dynamic linear model is constructed as a weighted sum of a plurality of basis functions by converting the time series data into coefficients of the plurality of basis functions;

wherein the constructing of the dynamic linear model as the weighted sum of the plurality of basis functions reduces dimensionality in connection with performance of computations to construct the dynamic linear model based on the plurality of fine-grained discrete time intervals;

wherein a number of the plurality of basis functions is less than a number of the plurality of fine-grained discrete time intervals;

wherein a model for the coefficients of the plurality of basis functions is dependent on the incentive signals, the incentive signals indicating dynamic pricing of the electricity over the given time period to enable the model for coefficients to capture demand shifting effects in response to the dynamic pricing;

wherein the model for the coefficients of the plurality of basis functions comprises a holiday effect

wherein a holiday indicator is one of a binary variable and a fractional value; and

wherein converting the time series data comprises converting a first plurality of graphs based on the plurality of fine-grained discrete time intervals to a second plurality of graphs corresponding to the first plurality of graphs, wherein the second plurality of graphs are based on the plurality of basis functions and the number of the second plurality of graphs is less than the number of the first plurality of graphs; and

estimating a forecasted demand cycle over the plurality of fine-grained discrete time intervals by obtaining forecasts and forecast errors using the dynamic linear model;

wherein constructing the dynamic linear model comprises:

specifying a state-space model and variance parameters in the state-space model;

estimating one or more unknown variance parameters by using a maximum-likelihood method to generate one or more maximum-likelihood estimates;

replacing the one or more unknown variance parameters with the one or more maximum-likelihood estimates; and

generating a load curve from the second plurality of graphs based on the weighted sums of the plurality of basis functions;

wherein the forecasts and forecast errors are obtained based at least in part on the maximum-likelihood estimates; and

wherein the steps of the method are performed by at least one processing device operatively coupled to a memory.

9. The method according to claim 8 , wherein the first historical data comprises data associated with usage of a resource at regular time intervals over a given period by a plurality of consumers.

10. The method according to claim 8 , wherein the obtained fine-grained usage data further comprises calendar information over the plurality of demand cycles.

11. The method according to claim 8 , wherein the state-space model includes at least one of overall level, day-of-week effects, simultaneous-effects and shifting-effects.

12. The method according to claim 8 , wherein constructing the dynamic linear model further comprises specifying prior distributions for initial values of a state vector.

13. The method according to claim 8 , wherein the plurality of basis functions comprise a plurality of knots at locations corresponding to peak periods of electricity usage.

14. The method according to claim 13 , wherein the plurality of basis functions comprise periodic cubic-spline basis functions.

15. An article of manufacture comprising a non-transitory processor readable storage medium comprising program code tangibly embodied thereon, which when executed by a processor, performs steps for modeling a cyclical demand system associated with an electricity grid, the steps comprising:

maintaining a metering component in two-way communication with the electricity grid for collecting usage data;

obtaining, over a plurality of demand cycles, fine-grained usage data from the metering component via the two-way communication with the electricity grid over a plurality of fine-grained discrete time intervals, wherein the plurality of fine-grained discrete time intervals have durations that vary based on demand, and wherein the obtained fine-grained usage data comprises first historical data associated with one or more demand measurements over the plurality of demand cycles, second historical data associated with incentive signals over the plurality of demand cycles, and time series data for the first and second historical data;

dynamically transmitting the incentive signals to a plurality of customers on the electricity grid, wherein the incentive signals comprise actual electricity pricing data and an indication that a price of electricity will be higher or lower over a given time period;

constructing a dynamic linear model using the obtained fine-grained usage data, wherein the dynamic linear model is constructed as a weighted sum of a plurality of basis functions by converting the time series data into coefficients of the plurality of basis functions;

wherein the constructing of the dynamic linear model as the weighted sum of the plurality of basis functions reduces dimensionality in connection with performance of computations by the processor to construct the dynamic linear model based on the plurality of fine-grained discrete time intervals;

wherein a number of the plurality of basis functions is less than a number of the plurality of fine-grained discrete time intervals;

wherein a model for the coefficients of the plurality of basis functions is dependent on the incentive signals, the incentive signals indicating dynamic pricing of the electricity over the given time period to enable the model for coefficients to capture demand shifting effects in response to the dynamic pricing;

wherein the model for the coefficients of the plurality of basis functions comprises a holiday effect

wherein a holiday indicator is one of a binary variable and a fractional value; and

wherein converting the time series data comprises converting a first plurality of graphs based on the plurality of fine-grained discrete time intervals to a second plurality of graphs corresponding to the first plurality of graphs, wherein the second plurality of graphs are based on the plurality of basis functions and the number of the second plurality of graphs is less than the number of the first plurality of graphs; and

estimating a forecasted demand cycle over the plurality of fine-grained discrete time intervals by obtaining forecasts and forecast errors using the dynamic linear model;

wherein constructing the dynamic linear model comprises:

specifying a state-space model and variance parameters in the state-space model;

estimating one or more unknown variance parameters by using a maximum-likelihood method to generate one or more maximum-likelihood estimates;

replacing the one or more unknown variance parameters with the one or more maximum-likelihood estimates; and

generating a load curve from the second plurality of graphs based on the weighted sums of the plurality of basis functions;

wherein the forecasts and forecast errors are obtained based at least in part on the maximum-likelihood estimates.

16. The article according to claim 15 , wherein the first historical data comprises data associated with usage of a resource at regular time intervals over a given period by a plurality of consumers.

17. The article according to claim 15 , wherein the obtained fine-grained usage data further comprises calendar information over the plurality of demand cycles.

18. The article according to claim 15 , wherein the state-space model includes at least one of overall level, day-of-week effects, simultaneous-effects and shifting-effects.

19. The article according to claim 15 , wherein constructing the dynamic linear model further comprises specifying prior distributions for initial values of a state vector.

20. The article according to claim 15 , wherein the plurality of basis functions comprise a plurality of knots at locations corresponding to peak periods of electricity usage.