Method for predicting outputs of photovoltaic devices based on two-dimensional fourier analysis and seasonal auto-regression
AssigneeMitsubishi Electric Research Laboratories, Inc.
InventorsWeihong Zhang, Daniel Nikolaev Nikovski
Life of this patent
Estimating the term…
Abstract
An output of a photovoltaic (PV) device is predicted by applying Fourier analysis to historical data to obtain frequencies and a mean of the frequencies in the data. Regression analysis is applied to the data to obtain a regression coefficient. Then, the prediction is a sum of the mean at the time step and a deviation from the mean at a previous time step, wherein the means are represented and approximated by selected frequencies, and the deviation for the previous time step is weighted by the regression coefficient.
Claims14
A method for predicting an output of a photovoltaic (PV) device, comprising the steps of:
- aligning data in two dimensions according to days in a year, and hours in a day;
- applying a two-dimensional Fourier analysis to the data to obtain frequencies and a mean of the frequencies in the data, wherein the data are time series of a historical outputs of the PV device;
- applying regression analysis to the data to obtain a regression coefficient;
- and
- predicting the PV output at a time step by summing the mean at the time step and a deviation from the mean at a previous time step, wherein the means are approximated by selected frequencies, and wherein the deviation for the previous time step is weighted by the regression coefficient, and wherein the steps of the method are performed in a processor of a control module generating a control strategy for operating the PV device.
A method for predicting an output of a photovoltaic (PV) device, comprising the steps of:
- applying Fourier analysis to data to obtain frequencies and a mean of the frequencies in the data, wherein the data are time series of a historical outputs of the PV device, wherein the data are aligned as
- [
- x
- (
- 1
- )
- x
- (
- 2
- )
- …
- x
- (
- l
- )
- ]
- =
- [
- x
- 1
- x
- 2
- …
- x
- MN
- x
- MN
- +
- 1
- x
- MN
- +
- 2
- …
- x
- 2
-
-
- MN
- …
- …
- …
- …
- x
- (
- I
- -
- 1
- )
-
- MN
- +
- 1
- x
- (
- I
- -
- 1
- )
-
- MN
- +
- 2
- …
- x
- IMN
- ]
- in a matrix form, where M indicates a number of days in a year, N a number of hours in the day, and I is the number of years that is spanned by the data;
- applying regression analysis to the data to obtain a regression coefficient;
- and
- predicting the PV output at a time step by summing the mean at the time step and a deviation from the mean at a previous time step, wherein the means are approximated by selected frequencies, and wherein the deviation for the previous time step is weighted by the regression coefficient, and wherein the steps of the method are performed in a processor of a control module generating a control strategy for operating the PV device.
A controller for predicting an output of a photovoltaic (PV) device and generating a control strategy for operating the PV device, the controller comprising a processor for
- applying Fourier analysis to data to obtain frequencies and a mean of the frequencies in the data, wherein the data are time series of a historical outputs of the PV device;
- applying regression analysis to the data to obtain a regression coefficient;
- and
- predicting the PV output at a time step by summing the mean at the time step and a deviation from the mean at a previous time step, wherein the means are approximated by selected frequencies, and wherein the deviation for the previous time step is weighted by the regression coefficient, and wherein the predicted output for the next time step t is
- {circumflex over (x)}t(mt,nt)=μt(mt,nt)+ρ×(xt−1(mt−1,nt−1)−μt−1(mt−1,nt−1)),
- where mt represents a day in a year, and nt represents an hour in the day, μt is the mean represented by an approximate Fourier series, and ρ is the regression coefficient.
Dependencies are inferred from the claim text.