Description
#409 Suggests Bishop [1988] as a truly explicit method to calculate i-from-v. Bishop's method is not subject to Newton convergence failure or to LambertW argument "blow up", handles ideal device parameters without extra logic, has no dependency on scipy, and naturally vectorizes via standard numpy functions' vectoization. The trade-off is that the explicit device terminal current computation occurs at the diode junction voltage (V_d := V + I R_s), not the device terminal voltage. (The device terminal voltage is "back-calculated" as V = V_d - I Rs.) Thus, additional computation work must be done to determine currents at specified terminal voltages, inc. I_sc @ V_sc=0, I_mp @ V_mp, and I_oc=0 @ V_oc.
The Bishop method is a very good candidate for a "gold" function implementation for testing the accuracy and speed of various implementations of i_from_v() that pvlib may wish to offer users or refactor to. By inverting the dataset, it could also be used to test v_from_i() implementations.
singlediode(photocurrent, saturation_current, resistance_series, resistance_shunt, nNsVth, ivcurve_pnts=None)I think that after this bishop88 function is implemented (presumably by #409), we should create and commit a corresponding "gold" dataset that we can use for better testing, refactoring, and optimization of the singlediode() function. I think we should try to cover a large number (all?) of the devices in the CEC database over a combination of temperature and irradiance ranges (perhaps the IEC 61853-1 matrix?). This dataset should also be valuable to the greater PV modeling community who wish to test/contribute new computational algorithms.
Outstanding questions:
(1) What should the voltage range be? Specifically, the diode voltages will give some terminal voltages less than zero, and how much beyond V_oc should we try to extend?
(2) Can/should we also include I_sc, I_x, I_mp, I_xx, and V_oc points, and what about the slope at I_sc and slope at V_oc?
(3) Does python have an interval analysis package up to the task of verifying our computations, e.g., intPy?
(4) Is it feasible to run such a comprehensive test suite on every PR?