Ransome_poster_2022

MLFM FITTING OUTDOOR MATRICES : pr_dc(G, t_mod) Fit data non-weighted or “weighted by occurrence” [Gantner #78 c-Si] normalised efficiency pr_dc DRAFT FOR CFV Do not distribute MLFM COEFFICIENTS ARE INDEPENDENT FOR UNIQUE MATRIX FITS • Alter each of the mlfm4+ coefficients (c_c, c_t, c_lg, c_g) separately • Show sensitivity : shape and magnitude of apparent performance change (red arrows) • Changes are independent meaning there’s a unique best fit 1) c_c vs. 10% rise Constant change Fits: “meas/nameplate” 4) c_g vs. 50% rise Changes most at high g Fits: r_series loss 3) c_lg vs. 20% rise Changes most at low g Fits: v_oc, r_shunt drop 2) c_t vs. 20% rise Change ∝ (t_mod-25C) Fits: gamma, beta etc. STC DEFINITIONS # 18 PVPMC 23-24 Aug 2022 Salt Lake City, USA IMPROVING ANALYSIS METHODS FOR IEC 61853 MATRIX MEASUREMENTS Steve Ransome1 (SRCL) and Juergen Sutterlueti (Gantner Instruments) 1 steve@steveransome.com : www.steveransome.com MLFM FITTING v_oc, v_mp, pr_dc : INDOOR vs. OUTDOOR Indoor: CFV IEC61853 Outdoor: Gantner Tempe AZ, 1year Module #5 Canadian Solar : rmse Module #78 Solar World : wrmse INTRODUCTION • IEC 61853 “Matrix method” defines ~28 dc measurements at up to 7 irradiances (g=0.1 – 1.1 kW/m2) × 4 temperatures (t_mod=15, 25, 50, 75C) • 4 independent coefficients are needed to uniquely fit a performance matrix : 1) c_c : measured/nameplate performance at STC 2) c_t : temperature coefficient (=1/X × dX/dt_mod [1/K] ) X=p_mp, v_oc . 3) c_lg : low light drop (caused by v_oc drop or r_shunt loss increasing at low g) 4) c_g : high light drop (caused by r_series as loss ~ I2.r_series ~ g2.r_series) • Matrices of pr_dc, v_oc, v_mp etc. can be fitted easily with a mechanistic model “MLFM4” (with ~50% of the fit errors of SAPM or PVGIS as neither of them model r_series correctly, it needs a c_g term [PVSC-49] ) SUMMARY References : www.steveransome.com email : steve@steveransome.com [PVSC 49] http://www.steveransome.com/pubs/2206_PVSC49_philadelphia_4_presented.pdf PVPMC/PVLIB : https://pvpmc.sandia.gov/ https://github.com/pvlib/pvlib-python Acknowledgements : Gantner Instruments and CFV for measurement data https://pvpmc.sandia.gov/download/7701/ Dots = Measured Smooth lines = Fits • MLFM is better than SAPM or PVGIS fitting matrices for all parameters with only 50% of their rmse (they don’t model r_series) [see PVSC49] • MLFM has optimised fits to indoor measurements and fits good outdoor measurements well • Weighting outdoor measurements by occurrence mean infrequent extreme or transient data don’t affect the fits • The MLFM matrix fit c_t parameter is an accurate temperature coefficient (without needing extra measurements at 1000W/m2 ) # GLOSSARY : nomenclature and definitions [unit] # g = measured poa irradiance ~(0.1 – 1.1) [kW/m^2] # t_mod = measured module temperature ~(15,25,50,75) [C] g_stc = 1 # [kW/m^2] t_stc = 25 # [C] dt = t_mod – 25 # [C] t_k = t_mod + 273.15 # [K] t_stc_k = 298.15 # [K] # normalise data for easier fitting and understanding) # NAMING PREFIXES meas(ured) norm(alised) fit(ted), stc, lic, noct etc. norm_i_sc = meas_i_sc / stc_i_sc / g # [%] norm_v_oc = meas_v_oc / stc_v_oc # [%] norm_pr_dc = meas_p_mp / stc_p_mp / g # [%] norm_i_mp = meas_i_mp / stc_i_mp / g # [%] norm_v_mp = meas_v_mp / stc_v_mp # [%] # MLFM4+: 4 meaningful, normalised coefficients # 1 const 2 temp coeff 3 low light improvement 4 high light norm_param = c_c + c_t*(t_mod–25) + c_lg * log10(g)*(t_k/t_stc_k) + c_g*g Occurrence of external data at Tempe, AZ, 1m each h for 1 yr Most frequent region (g ~0.8-1.0kW/m2, t_mod ~40-65C) Can ignore least frequent and any ‘outliers’ (0.1%) A BETTER METHOD TO FIND TEMPERATURE COEFFICIENTS Temperature coefficients can be more simply and accurately derived using c_t from mlfm matrix fits without needing extra measurements and trend fits as used in IEC 61853 Fitting good indoor vs good outdoor data : • Weight outdoor data by occurrence • Outdoor weighted v_mp and v_oc fits can be as good as indoor! • Higher pr_dc variability outdoors (soiling, aoi, beam fraction and spectrum affect i_sc) • MLFM fits matrices well v_oc c_t ~ -0.32%/K wrmse ~0.17% v_oc c_t ~ -0.30%/K rmse ~0.35% v_mp c_t ~ -0.39%/K rmse ~ 0.30% pr_dc c_t ~ -0.40%/K rmse ~0.18% Non-weighted pr_dc fit Weighted pr_dc fit STC LIC STC MLFM fits both weighted and unweighted well 4 independent terms are needed to model matrix behaviour IV curve terms IEC 61853 values and linear trend fits Not needed IEC 61853 TREND values and residual from MLFM fits c_t = -0.46%/K (gamma) rmse ~0.99% wrmse ~0.61% c_t = -0.45%/K (gamma) rmse ~1.05% wrmse ~ 0.55% STC v_mp c_t ~ -0.40%/K wrmse ~0.20% pr_dc c_t ~ -0.45%/K wrmse ~0.55% Tmod (C) Gi (W/m2) Weighted rmse (wrmse) is ~50% the rmse for good outdoor data as it tends to fit well behaved, bright, hot conditions and not cool, dull, infrequent and/or outliers