Feature Engineering (per hour t)

Normalize load

Causal roll means

Time encodings

Calendar features

Feature vector

Supervised Windowing

Build window

Target

Forecaster

QK-LSTM (kernelized gates) — or — LSTM

Train with MAE Early stopping, fixed seeds

Predict: ŷ_{t+1}

Stage 2: 1-Hour-Ahead DR LP (at time t → hour t+1)

Inputs

Decision variables

Constraints

Objective

Solve LP and update

Data & Preprocessing

Raw data

Chronological sort

Remove outliers

Train/Validation split

Min-max scaling

Optional Day-Ahead LP

Data Collection and database of historical data

Energy Generation

Residential Consumers Database of historical data

Energy Generation Forecast

Energy Consumption Forecast

Forecasting Methods

QLSTM Forecasting

Classical LSTM Forecasting

Load Shedding Management

Electric Grid