# An Hour-Specific Hybrid DNN–SVR Framework for National-Scale Short-Term Load Forecasting

## TL;DR

A new hybrid model combining deep learning and support vector regression improves short-term electricity load forecasting accuracy by over 13% compared to existing methods.

## Contribution

Proposes an hour-specific hybrid DNN–SVR framework validated on a 17-year national dataset with global weather integration.

## Key findings

- Hour-specific hybrid models reduce forecasting error by 13.5% compared to PatchTFT baselines.
- Global weather data integration cuts error by 29.2% over local measurements alone.
- Hybrid DNN–SVR outperforms standalone deep learning models in national-scale load forecasting.

## Abstract

What are the main findings?

The hour-specific hybrid DNN–SVR framework achieves a 13.5% lower forecasting error than the PatchTFT baseline.

Integrating global numerical weather predictions reduces error by 29.2% compared to using local measurements alone.

What are the implication of the main findings?

Decomposing daily prediction into hour-specific submodels proves more effective than unified modeling for capturing diverse intraday load regimes.

Validated on a 17-year dataset, this hybrid architecture offers a robust, high-accuracy alternative to pure deep learning baselines for operational grid management.

Short-term load forecasting (STLF) underpins the efficient and secure operation of power systems. This study develops and evaluates a hybrid architecture that couples deep neural networks (DNNs) with support vector regression (SVR) for national-scale day-ahead STLF using Croatian load data from 2006 to 2022. The approach employs an hour-specific framework of 24 hybrid models: each DNN learns a compact nonlinear representation for a given hour, while an SVR trained on the penultimate layer activations performs the final regression. Gradient-boosting-based feature selection yields compact, informative inputs shared across all model variants. To overcome limitations of historical local measurements, the framework integrates global numerical weather prediction data from the TIGGE archive with load and local meteorological observations in an operationally realistic setup. In the held-out test year 2022, the proposed hybrid consistently reduced forecasting error relative to standalone DNN-, LSTM- and Transformer-based baselines, while preserving a reproducible pipeline. Beyond using SVR as an alternative output layer, the contributions are as follows: addressing a 17-year STLF task, proposing an hour-specific hybrid DNN–SVR framework, providing a systematic comparison with deep learning baselines under a unified protocol, and integrating global weather forecasts into a practical day-ahead STLF solution for a real power system.