為區分 BESS 與預測輔助排程之影響，本研究設計無 BESS、僅加入 BESS，以及結合預測輔助設備排程與 BESS 等控制情境。結果顯示，BESS 可降低住宅購電成本，而在相同儲能條件下，導入預測輔助排程後可進一步改善成本表現。系統實作方面，本研究將模型部署於 NVIDIA Jetson Xavier NX，並透過 HEMS Device Simulator 與 Wireshark 驗證 ECHONET Lite 控制封包之建立、傳送、解析與狀態回應流程。整體而言，本研究完成由負載預測、HEMS 排程、BESS 成本分析至邊緣端設備控制之整合架構。;With the increasing deployment of smart meters， renewable energy resources， and residential energy storage systems， a home energy management system must integrate short-term load forecasting， appliance scheduling， and edge-side control. This study proposes a residential short-term load forecasting framework that combines Temporal Fusion Transformer (TFT) with conditional TimeGAN (cTimeGAN). The forecasting results are further connected to a Home Energy Management System (HEMS) for schedulable-appliance arrangement， Battery Energy Storage System (BESS) cost analysis， and ECHONET Lite-based edge-side control.
The household energy data are divided into non-schedulable base load， schedulable load， renewable energy output， and storage-related information. The non-schedulable base load is selected as the forecasting target. cTimeGAN is used to increase the diversity of the training data， while TFT performs short-term load and quantile forecasting. The experimental results show that the augmented model improves several error metrics and provides more stable time-series tracking together with prediction intervals for subsequent energy-management decisions.
To distinguish the influence of BESS from that of forecast-assisted scheduling， this study compares control cases without BESS， with BESS only， and with forecast-assisted appliance scheduling combined with BESS. The results indicate that BESS reduces electricity costs， while forecast-assisted scheduling provides additional improvement under the same storage conditions. For system implementation， the trained model is deployed on an NVIDIA Jetson Xavier NX. The HEMS Device Simulator and Wireshark are used to verify ECHONET Lite packet construction， transmission， parsing， and device-state responses. Overall， the proposed framework integrates residential load forecasting， HEMS scheduling， BESS cost analysis， and edge-side device control.
