pv-estimator-tight-v1-int8

lelloman/pv-estimator-tight-v1-int8

pv-estimator-tight-v1-int8 is a machine learning model from lelloman. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for onnx. The card lists the license as apache-2.0.

This repository contains the tight-v1 INT8 ONNX source-model bundle used by PV Estimator local v1.

Downloads · 30 days

Access

Public

Updated Jun 6, 2026

Repo size

28.9 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.onnx28.9 MB · 100%

At a glance

Library: onnx
License: apache-2.0
Access: Public
Created: Jun 6, 2026
Updated: Jun 6, 2026
SHA: 9e0120db

Model card

From the Hugging Face model README

PV Estimator Tight-v1 INT8 Source-Model Ensemble

This repository contains the tight-v1 INT8 ONNX source-model bundle used by PV Estimator local v1.

PV Estimator uses this bundle to estimate photovoltaic production locally from coordinates and simple system parameters. The default GitHub build embeds these same files into the pv CLI and pv-tui terminal UI.

Files

File	Purpose
`source-model-artifacts.json`	Runtime manifest, normalization stats, source list, and coverage rules.
`nasa_power.onnx`	NASA POWER climate-normal source model.
`pvgis_era5.onnx`	PVGIS ERA5 source model.
`pvgis_sarah3.onnx`	PVGIS SARAH3 source model.
`coverage/pvgis_sarah3_empirical_grid_mask.json`	Empirical SARAH3 applicability mask.
`SHA256SUMS`	Artifact checksums.

Manifest Snapshot

Field	Value
Schema version	`1`
Model family	`monthly-hourly climate-normal residual MLP`
Input features	`66`
Temporal bins	`288`
Output targets	`10` climate-normal targets

Targets:

ghi_mean_w_m2
dni_mean_w_m2
dhi_mean_w_m2
temp_mean_c
wind_mean_m_s
ghi_std_w_m2
dni_std_w_m2
dhi_std_w_m2
temp_std_c
wind_std_m_s

Sources and Coverage

Source ID	Label	Coverage rule
`nasa_power`	NASA POWER	Global
`pvgis_era5`	PVGIS-ERA5	Global land PVGIS gateway
`pvgis_sarah3`	PVGIS-SARAH3	Empirical grid mask

The runtime uses only applicable sources for each coordinate. PV Estimator JSON output reports coverage.applicable_sources and coverage.pvgis_sarah3_applicable.

Evaluation Snapshot

The tight-v1 INT8 comparison used the same pv-cli estimate runtime path on a 120-location regional benchmark.

Reference	Count	Tight v1 MBE	Tight v1 MAE	Tight v1 RMSE
PVGIS-ERA5	120	-0.442%	2.470%	3.201%
PVGIS-SARAH3	100	2.374%	3.218%	4.418%

Distribution snapshot:

Reference	p90 abs error	max abs error
PVGIS-ERA5	4.892%	11.623%
PVGIS-SARAH3	6.469%	19.456%

Mean source disagreement spread was 4.785% in the tight-v1 comparison.

Use With PV Estimator

After downloading this repository, pass the directory to pv estimate:

pv estimate \
  --lat 40.4168 \
  --lon=-3.7038 \
  --model-dir . \
  --format json

The GitHub project embeds this same bundle by default, so --model-dir is optional when using the release binary.

Limitations

Estimates are climate-normal predictions, not measured site data.
This is not a substitute for a bankability study, engineering design, permitting review, or site inspection.
Local shading, horizon obstructions, soiling, inverter clipping, curtailment, snow, and detailed equipment behavior are not fully modeled in local v1.
Accuracy can degrade in locations or climates underrepresented by training and evaluation data.
Uncertainty bands are based on source-model disagreement and a multiplier, not a fully calibrated probabilistic forecast for every coordinate.

Project

PV Estimator GitHub repository: https://github.com/lelloman/pv-estimator

The source project is licensed under MIT OR Apache-2.0. This model repository is published under Apache-2.0.

Library: onnx
License: apache-2.0
Created: Jun 6, 2026
Updated: Jun 6, 2026