Unsupervised solar energy disaggregation system and method
A computer-implemented method for solar energy disaggregation at a node of a distribution system includes obtaining first data including samples of averaged net power measured by a meter installed at the node and second data including samples of solar irradiation local to the node averaged over intervals temporally correlated with the first data. A physics-based model of the node is trained using the first and second data, to optimize a set of model parameters. The physics-based model is defined such that an average net power includes a composite of an average solar generation and an average power consumption at the node and the average solar generation is modeled as a function of an average solar irradiation. An aggregate solar generation at the node is estimated/predicted from actual irradiation data/irradiation forecast data local to the node using a model parameter optimized by the training.
1 . A computer-implemented method for estimating or predicting an aggregate solar generation at a node of a distribution system, the method comprising:
obtaining, by a computing system comprising one or more processors, first data pertaining to a net power measured by a meter installed at the node, the first data comprising a number of samples of net power averaged over a defined interval,
obtaining, by the computing system, second data pertaining to a solar irradiation local to the node, the second data comprising a number of samples of solar irradiation averaged over intervals temporally correlated with the first data,
training, by the computing system, a physics-based model of the node using the first and second data, to optimize a set of model parameters, the physics-based model being defined such that an average net power includes a composite of an average solar generation and an average power consumption at the node and the average solar generation is modeled as a function of an average solar irradiation, and
estimating or predicting, by the computing system, an aggregate solar generation at the node from an input respectively comprising actual irradiation data or irradiation forecast data local to the node, using a model parameter optimized by the training.
2 . The method according to claim 1 , wherein the interval for averaging includes a one-day interval or a multiple thereof.
3 . The method according to claim 1 , wherein the actual irradiation data or irradiation forecast data, and respectively the estimated solar generation or predicted solar generation at the node, comprise time series data having higher resolution than the averaged data used for training the physics-based model.
4 . The method according to claim 1 ,
wherein the average power consumption in the physics-based model is modeled as a function of a base power consumption,
wherein the physics-based model comprises a regression model and the set of model parameters optimized comprises a first model parameter including the base power consumption and a second model parameter including a coefficient defining a linear relationship between the average solar irradiation and the average solar generation at the node, and
wherein the optimized second model parameter is used for estimating or predicting the aggregate solar generation at the node from the input comprising respectively the actual irradiation data or the irradiation forecast data local to the node.
5 . The method according to claim 4 , wherein the second parameter comprises multiple components assigned to respectively multiple intervals of average irradiation over a day.
6 . The method according to claim 4 , wherein the second parameter comprises multiple components assigned to respectively multiple types of irradiation, the multiple types of irradiation selected from the group consisting of: direct irradiation, diffuse irradiation and reflected irradiation.
7 . The method according to claim 1 , further comprising:
obtaining, by the computing system, third data pertaining to a temperature local to the node, the third data comprising a number of samples of temperature averaged over intervals temporally correlated with the first data, and
training, by the computing system, the physics-based model using the first, second and third data, the physics-based model being defined such that the average power consumption at the node is modeled as a function of an average temperature.
8 . The method according to claim 7 ,
wherein the average power consumption at the node is modeled as a sum of a base power consumption and an average temperature dependent power consumption at the node,
wherein the physics-based model comprises a regression model and the set of model parameters optimized comprises a first model parameter including the base power consumption, a second model parameter including a coefficient defining a linear relationship between the average solar irradiation and the average solar generation at the node, and a third model parameter including a coefficient defining a relationship between the average temperature and the average temperature dependent power consumption, and
wherein the optimized second model parameter is used for estimating or predicting the aggregate solar generation at the node from the input comprising respectively the actual irradiation data or the irradiation forecast data local to the node.
9 . The method according to claim 8 , wherein the third model parameter defines a linear relationship between the average temperature and the average temperature dependent power consumption.
10 . The method according to claim 9 , wherein the third model parameter comprises a first component defining a relationship between the average temperature and the average temperature dependent power consumption when the average temperature is above a threshold temperature, and a second component defining a relationship between the average temperature and the average temperature dependent power consumption when the average temperature is below the threshold temperature.
11 . A non-transitory computer-readable storage medium including instructions that, when processed by a computing system, configure the computing system to perform the method according to claim 1 .
12 . A computing system for estimating or predicting an aggregate solar generation at a node of a distribution system, the computing system comprising:
one or more processors,
a non-transitory memory in communication with the one or more processors, the non-transitory memory including algorithmic modules executable by the one or more processors, the algorithmic modules comprising:
a model trainer configured to:
obtain first data pertaining to a net power measured by a meter installed at the node, the first data comprising a number of samples of net power averaged over a defined interval,
obtain second data pertaining to a solar irradiation local to the node, the second data comprising a number of samples of solar irradiation averaged over intervals temporally correlated with the first data, and
train a physics-based model of the node using the first and second data, to optimize a set of model parameters, the physics-based model being defined such that an average net power includes a composite of an average solar generation and an average power consumption at the node and the average solar generation is modeled as a function of an average solar irradiation, and,
a solar generation estimator/predictor configured to estimate or predict an aggregate solar generation at the node from an input respectively comprising actual irradiation data or irradiation forecast data local to the node, using a model parameter optimized by the training.