Optimization of photovoltaic soiling model parameters
A method for optimizing a parameter set for a photovoltaic (PV) soiling model including executing the PV soiling model with a first plurality of parameter sets using as input wash data to generate first predicted energy loss data, calculating first losses based on a difference between the first predicted energy loss data and time-series sensor data, executing a parameter-tuning model using as input the first plurality of parameter sets to generate a second plurality of parameter sets, executing the PV soiling model with the second plurality of parameter sets using as input the wash data to generate second predicted energy loss data, calculating second losses based on the second predicted energy loss data and the time-series sensor data, and, in response to a loss of the second losses being below a threshold, generating a data structure including a parameter set corresponding to the loss.
1 . A method comprising:
executing a photovoltaic (PV) soiling model with a first plurality of parameter sets using as input wash data to generate first predicted energy loss data;
calculating a first set of losses based on a difference between the first predicted energy loss data and time-series sensor data;
ranking the first plurality of parameter sets based on the first calculated set of losses;
selecting a subset of the first plurality of parameter sets based on the ranking of the first plurality of parameter sets;
executing a parameter-tuning model using as input the subset of the first plurality of parameter sets to generate a second plurality of parameter sets;
executing the PV soiling model with the second plurality of parameter sets using as input the wash data to generate second predicted energy loss data;
calculating a second set of losses based on a difference between the second predicted energy loss data and the time-series sensor data;
generating a data structure including a selected parameter set of the second plurality of parameter sets;
generating a recommendation for one or more manual cleaning events by executing the PV soiling model with the selected parameter set using as input candidate manual cleaning data; and
causing implementation of the recommendation for the one or more manual cleaning events to improve PV output.
2 . The method of claim 1 , further comprising initializing the first plurality of parameter sets by generating pseudo-random numbers.
3 . The method of claim 1 , wherein the loss between the first predicted energy loss data and the time-series sensor data includes an average absolute difference between a predicted percent energy loss and a measured percent energy loss.
4 . The method of claim 1 , wherein the selected parameter set includes at least one of soiling buildup rates, a rain cleaning threshold, a rain accumulation period, and a grace period.
5 . The method of claim 4 , further comprising adjusting the soiling buildup rates based on an average soiling buildup rate for a time period.
6 . The method of claim 1 , wherein the wash data includes rainfall data and manual cleaning data.
7 . The method of claim 1 , wherein the parameter-tuning model comprises a derivative-free optimization algorithm.
8 . The method of claim 1 , further comprising:
executing the PV soiling model with the selected parameter set using as input second wash data to generate third predicted energy loss data; and
applying the third predicted energy loss data to a prediction of PV array output.
9 . The method of claim 1 , wherein the input for the parameter-tuning model to generate the second plurality of parameter sets includes the subset of the first plurality of parameter sets and the ranking of the first plurality of parameter sets.
10 . The method of claim 1 , further comprising generating soiling model parameters that characterize soiling behavior at a specific photovoltaic (PV) site, wherein the soiling model parameters are compatible with Typical Meteorological Year (TMY) weather data files for subsequent simulation or performance modeling.
11 . A non-transitory, computer-readable medium including instructions which, when executed by one or more processors, cause the one or more processors to:
execute a photovoltaic (PV) soiling model with a first plurality of parameter sets using as input wash data to generate first predicted energy loss data;
calculate a first set of losses based on a difference between the first predicted energy loss data and time-series sensor data;
rank the first plurality of parameter sets based on the first calculated set of losses;
select a subset of the first plurality of parameter sets based on the ranking of the first plurality of parameter sets;
execute a parameter-tuning model using as input the subset of the first plurality of parameter sets to generate a second plurality of parameter sets;
execute the PV soiling model with the second plurality of parameter sets using as input the wash data to generate second predicted energy loss data;
calculate a second set of losses based on a difference between the second predicted energy loss data and the time-series sensor data;
generate a data structure including a selected parameter set of the second plurality of parameter sets;
generate a recommendation for one or more manual cleaning events by executing the PV soiling model with the selected parameter set using as input candidate manual cleaning data; and
cause implementation of the recommendation for the one or more manual cleaning events to improve PV output.
12 . The non-transitory, computer-readable medium of claim 11 , wherein the instructions cause the one or more processors to initialize the first plurality of parameter sets by generating pseudo-random numbers.
13 . The non-transitory, computer-readable medium of claim 11 , wherein the loss between the first predicted energy loss data and the time-series sensor data includes an average absolute difference between a predicted percent energy loss and a measured percent energy loss.
14 . The non-transitory, computer-readable medium of claim 11 , wherein the selected parameter set includes at least one of soiling buildup rates, a rain cleaning threshold, a rain accumulation period, and a grace period.
15 . The non-transitory, computer-readable medium of claim 14 , wherein the instructions cause the one or more processors to adjust the soiling buildup rates based on an average soiling buildup rate for a time period.
16 . The non-transitory, computer-readable medium of claim 11 , wherein the wash data includes rainfall data and manual cleaning data.
17 . The non-transitory, computer-readable medium of claim 11 , wherein the parameter-tuning model comprises a derivative-free optimization algorithm.
18 . The non-transitory, computer-readable medium of claim 11 , wherein the instructions cause the one or more processors to:
execute the PV soiling model with the selected parameter set using as input second wash data to generate third predicted energy loss data; and
apply the third predicted energy loss data to a prediction of PV array output.
19 . The non-transitory, computer-readable medium of claim 11 , wherein the input for the parameter-tuning model to generate the second plurality of parameter sets includes the subset of the first plurality of parameter sets and the ranking of the first plurality of parameter sets.
20 . The non-transitory, computer-readable medium of claim 11 , wherein the instructions cause the one or more processors to generate soiling model parameters that characterize soiling behavior at a specific photovoltaic (PV) site, wherein the soiling model parameters are compatible with Typical Meteorological Year (TMY) weather data files for subsequent simulation or performance modeling.