Download protocol.json from ricardofitas/sparse-sensor-thermal-inference: direct link, hf CLI and curl.
- Browser
- Download file 4.1 kB
-
https://huggingface.co/datasets/ricardofitas/sparse-sensor-thermal-inference/resolve/main/protocol.json
- Command line
-
hf download hf://datasets/ricardofitas/sparse-sensor-thermal-inference/protocol.json
-
curl -L -o protocol.json https://huggingface.co/datasets/ricardofitas/sparse-sensor-thermal-inference/resolve/main/protocol.json
4.1 kB
| { | |
| "status": "frozen_before_pilot_and_main_runs", | |
| "research_question": "Can a jointly estimated additive local ambient-temperature offset improve sparse-sensor thermal calibration and conditional transfer, and when does structural mismatch confound it?", | |
| "ambient_definition": "Ta_true(t)=ERA5_Land(t)+delta; Ta_model(t)=ERA5_Land(t)+b_hat; naive fixes b_hat=0; joint estimates b_hat. Delta is a synthetic imposed local offset, not measured Copernicus error.", | |
| "biases_K": [-5,-2,0,2,5], | |
| "noise_sigma_K": [0.25,1.0], | |
| "seeds": [11,23,37,51,71], | |
| "master_sensors_m": [[0.16,0.24],[0.28,0.60],[0.52,0.72],[0.64,0.96],[0.30,0.68]], | |
| "layouts": {"one_hot":[1],"two_clustered":[1,4],"two_spread":[0,1],"four_spread":[0,1,2,3]}, | |
| "calibration_start_dates_utc": ["2022-06-01","2022-07-01"], | |
| "calibration_hours": [0,72], | |
| "within_window_test_hours": [73,96], | |
| "independent_transfer_start_dates_utc": ["2022-07-10","2022-08-15"], | |
| "independent_transfer_test_hours": [1,96], | |
| "truth_regimes": { | |
| "nominal":{"k":12.0,"h":8.0,"winding_source_factor":1.0}, | |
| "alternate":{"k":8.0,"h":12.0,"winding_source_factor":1.0}, | |
| "source_mismatch":{"k":12.0,"h":8.0,"winding_source_factor":1.15} | |
| }, | |
| "model_source_factor": 1.0, | |
| "model_methods": ["naive_kh","joint_khb"], | |
| "reference": {"nx":32,"ny":48,"dt_s":150}, | |
| "calibration_grid": {"nx":12,"ny":18,"dt_s":900}, | |
| "constants": {"Lx_m":0.8,"Ly_m":1.2,"C_J_m3K":1800000,"qcore_W_m3":250,"qwind_W_m3":950,"T0_C":20}, | |
| "bounds": {"k":[3,30],"h":[3,20],"b_K":[-10,10]}, | |
| "optimizer": {"name":"scipy least_squares trf with analytic forward tangent sensitivities","variables":["log(k)","log(h)","b/1K"],"initial_k":10,"initial_h":10,"initial_b":0,"max_nfev":100,"ftol":1e-9,"xtol":1e-9,"gtol":1e-8}, | |
| "primary_metrics": ["field_RMSE_K on fixed common12x18 cell-center evaluation points","peak_RMSE_K on the same evaluation grid","paired joint-minus-naive field RMSE by seed","bias estimation error","k,h relative errors under matched equations"], | |
| "secondary_metrics": ["sensor fit RMSE","parameter bound hits","solver success/nfev","scaled sensitivity singular values/condition number","correlation of scaled parameters"], | |
| "identifiability_subset": {"regime":"nominal","calibration_start":"2022-06-01","bias_K":5,"noise_sigma_K":0.25,"seed":11,"layouts":"all","multistarts":[[6,5,-5],[10,10,0],[20,15,5]],"bias_profile_grid_K":[-10,-8,-6,-4,-2,0,2,4,6,8,10]}, | |
| "pilot": {"regime":"nominal","calibration_start":"2022-06-01","bias_K":[0,5],"noise_sigma_K":0.25,"seed":11,"layout":"four_spread","checks":["discrete energy balance absolute<1e-7 W/m","analytic Jacobian relative central-difference error<1e-4","reference32x48/150 vs48x72/75 RMSE<0.1 K","optimizer objective decreases; no parameter-recovery acceptance criterion"]}, | |
| "full_factorial_fit_count": 2400, | |
| "evaluation_scenarios_per_fit": 3, | |
| "noise_pairing": "For each calibration date/regime/bias/sigma/seed, generate one97x5 master noise array; nested layouts and both methods use exactly its corresponding columns. Exact known initial20C values are uncorrupted. All other timestamps have independent Gaussian noise.", | |
| "forecast_scope": "Conditional numerical integration with known retrospective ambient and prescribed future load. Each transfer scenario begins from the stipulated exact uniform20C field, not an unknown operating asset state.", | |
| "load_design": { | |
| "calibration": "0.85+0.18*sin(2pi(t-6)/24)+0.08*sin(2pi*t/8)+0.30*I[78<=t<90]", | |
| "hot_transfer": "1.00+0.22*sin(2pi(t-4)/24)+0.10*sin(2pi*t/12)+0.35*I[54<=t<66]", | |
| "august_transfer": "0.72+0.15*sin(2pi(t-9)/24)+0.06*sin(2pi*t/6)+0.25*I[30<=t<42]+0.20*I[78<=t<90]" | |
| }, | |
| "interpretation_limits": ["same-PDE synthetic identification despite distinct discretizations","local J rank does not prove global identifiability","source-mismatch inferred parameters are effective/confounded","five noise seeds are not independent physical experiments","bounds are explicit prior restrictions","ERA5-Land historical redistribution, no live data or measured transformer","no GPU execution available"] | |
| } | |