随着新能源发电与电力市场化改革的发展,电价波动性显著增强,传统预测方法在捕捉非线性与长时依赖方面存在局限。为应对这一问题,我基于公开电价与相关因素数据集,构建了电价预测模型,旨在探索机器学习与深度学习方法在电力市场短期与中长期价格预测中的适用性与效果差异。项目不仅聚焦于提高预测精度,也尝试通过模型对比与结果可视化,为电力市场价格建模提供可解释性参考
With the increasing penetration of renewable energy and the ongoing liberalization of electricity markets, price volatility has become more pronounced. Traditional forecasting methods face limitations in capturing nonlinear dynamics and long-term dependencies. To address this challenge, I developed an electricity price forecasting model using publicly available price and related feature datasets. The project aimed to evaluate the effectiveness of machine learning and deep learning approaches (e.g., LSTM, Transformer) in both short-term and long-term forecasting. Beyond accuracy, the project emphasized model benchmark
电价预测模型 — 独立项目
时间:2024年暑期
基于时间序列与深度学习模型(LSTM、Transformer)构建电价预测系统,探索短期与中长期预测的差异化效果;
实现数据预处理与特征工程(缺失值处理、归一化、滞后特征构造、节假日与天气特征融合),提升模型鲁棒性;
在实验中对比了ARIMA、XGBoost与改进的Transformer结构,并以MAPE、RMSE为指标进行量化评估;
使用PyTorch + sklearn 完成模型搭建与训练,并基于可视化分析解释预测结果,撰写完整技术报告;
项目代码与报告开源至GitHub,积累实践经验并初步形成科研表达能力
Summer 2024
Developed a time-series forecasting system for electricity prices using LSTM and Transformer-based models, comparing short-term and long-term forecasting performance.
Designed and implemented data preprocessing & feature engineering, including missing value imputation, normalization, lag features, and integration of holiday/weather factors.
Conducted benchmarking with ARIMA, XGBoost, and enhanced Transformer variants, evaluated via MAPE and RMSE metrics.
Built and trained models using PyTorch and scikit-learn, with visualization for interpretability and a full technical report.
Open-sourced project on GitHub, gaining hands-on experience in applied machine learning and technical reportin
数据处理:收集并清洗电价及相关特征数据,完成缺失值填补、归一化、时间窗口切分,并构造滞后特征与节假日、天气等外生变量;
模型搭建:实现 ARIMA、XGBoost、LSTM、改进版 Transformer 等多种模型,采用 PyTorch + scikit-learn 完成训练与调参;
实验设计:基于时间序列交叉验证,分别进行短期(日级、周级)与中长期(月级)预测对比;
结果分析:使用 MAPE、RMSE 等指标评估模型性能,并通过可视化图表分析预测误差与价格波动规律;
工具链:Python、Pandas、NumPy、Matplotlib、PyTorch、scikit-learn;项目代码及报告已开源至 GitHub