Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1543
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-16T06:54:39.780275Z
status: published
lens: g_space
sector: climate
headline: Multi-objective optimized signal decomposition with explainable and reliable hybrid models for multistep significant wave height forecasting in ocean wave energy applications
dek: Accurate forecasting of significant wave height (Hs) is vital for marine safety, coastal engineering, and renewable energy production. This study presents an effective integrated model system to forecast Hs for two distinct sites in Australia, Gladstone and Palm Beach, over various time horizons from 30-min to 24-h ahead during the period of 2018 to 2022. In the first stage, a novel multi-objective Optuna is developed to optimize the decomposition parameters using two criteria of compactness, bandwidth, and peak…
gain_title: Using SHAP to cut 37 to 48 decomposed inputs down to one dominant predictor kept accuracy nearly unchanged and lowered computational cost, supporting wave energy and coastal hazard planning.
problem_title: (none)
trace_subject: (none)
gain_reading: Using SHAP to cut 37 to 48 decomposed inputs down to one dominant predictor kept accuracy nearly unchanged and lowered computational cost, supporting wave energy and coastal hazard planning.
gain_evidence: reduced the input variables from 37 to 48 decomposed components to a single dominant predictor while maintaining nearly identical predictive accuracy
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers built a staged forecasting system for significant wave height at Gladstone and Palm Beach in Australia, covering horizons from 30 minutes to 24 hours using data from 2018 to 2022. A multi-objective Optuna search tuned the signal decomposition methods, SHAP ranked the most influential components, and LSTM, deep RVFL and XGBoost models produced the forecasts.

The work matters because wave height predictions inform marine safety, coastal engineering and renewable energy output, and a cheaper model that stays accurate is easier to use operationally. The paper reports results for only two sites, and accuracy varies widely across stations and horizons, so how well it transfers elsewhere is not established.
limitation: The lower bound of R2 (0.49) shows accuracy dropped at some stations or horizons, and reliability above 88% is only reported for forecasts up to 6 hours ahead, with only two sites tested.
tag: Evidence-backed gain
key_points: SSA-RVFL was the best model, with lower computational cost and the highest accuracy across all horizons. | Reported ranges were Index of Agreement 0.81-1, R2 0.49-0.98 and RMSE 0.04-0.33 across stations and horizons. | Monte Carlo simulation quantified uncertainty and showed operational reliability above 88% for forecasts up to 6 hours ahead. | Multi-objective Optuna removed the need for manual selection of decomposition parameters for SSA and D-MVMD.
rundown: The pipeline has three stages: optimized decomposition of wave signals using SSA and De-mixing Multivariate Variational Mode Decomposition, SHAP-based analysis of which components matter most, and Monte Carlo reliability analysis. The optimization used criteria tied to the compactness, bandwidth and peakedness of the signal's frequency spectrum.

The decomposed, interpretable signals fed three standalone forecasting models: LSTM, deep Random Vector Functional Link and XGBoost. Among them the SSA-RVFL combination performed best at both Gladstone and Palm Beach.

The authors conclude that the fully optimized SSA with SHAP-RVFL-MCS offers a robust framework for decision-making in coastal hazard and wave energy planning.
sources:
- peer_reviewed | Engineering Applications of Artificial Intelligence | https://doi.org/10.1016/j.engappai.2026.116454 | 2026-10-14
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

9df9a6126853e8b3307bf6f61049d57c4f35bc9b507799d83c2408efb482b59c

previous

0000000000000000000000000000000000000000000000000000000000000000