Scholarlink

Efficient planning and dispatch of electric power require accurate forecast of future energy demand. This is a challenging problem: given the historical load profile, weather variable and the time index, we predict the accurate future load. It has been shown that a joint modeling approach, using Long Short-Term Memory (LSTM) networks together with the Light Gradient Boosting Machine (LightGBM) algorithm, have contributed to improve prediction and inference capabilities. Although LSTM is good at learning sequential dependencies on time series data, LightGBM is specialized in modeling complex, and non-linear patterns. For the forecasting study, short-term day-ahead load prediction is carried out and is analysed using the data gathered from 25th Dec 2023 to 30th April 2025. The dataset, which retrieved from Chhattisgarh Load Dispatch Centre (CLDC), split in training and testing sets. It is composed of eight features, with the last one being the target variable and the first seven being input variables. In order to assess the performance of model, some commonly used performance indicators including RMSE, MAE and MAPE are calculated and compared with the corresponding research results of other existing techniques.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Leveraging Hybrid LSTM-LightGBM Models for Predicting Short-Term Load in Chhattisgarh State

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Suruchi Shrivastava,
  • Anamika Yadav,
  • Shubhrata Gupta

摘要

Efficient planning and dispatch of electric power require accurate forecast of future energy demand. This is a challenging problem: given the historical load profile, weather variable and the time index, we predict the accurate future load. It has been shown that a joint modeling approach, using Long Short-Term Memory (LSTM) networks together with the Light Gradient Boosting Machine (LightGBM) algorithm, have contributed to improve prediction and inference capabilities. Although LSTM is good at learning sequential dependencies on time series data, LightGBM is specialized in modeling complex, and non-linear patterns. For the forecasting study, short-term day-ahead load prediction is carried out and is analysed using the data gathered from 25th Dec 2023 to 30th April 2025. The dataset, which retrieved from Chhattisgarh Load Dispatch Centre (CLDC), split in training and testing sets. It is composed of eight features, with the last one being the target variable and the first seven being input variables. In order to assess the performance of model, some commonly used performance indicators including RMSE, MAE and MAPE are calculated and compared with the corresponding research results of other existing techniques.