摘要： 针对当前短期电价预测方法中变分模态分解（VMD）参数优化效率低、单一预测模型特征表达能力不足及特征冗余等问题，提出一种基于多策略改进冠豪猪优化（MSICPO）算法与深度学习的短期电价预测方法。首先，通过引入莱维飞行策略、周期性种群变异和动态调整参数机制改进冠豪猪优化（CPO）算法，提高其全局搜索能力与收敛速度，并用于优化VMD的模态数量和惩罚因子，以提高信号分解精度；其次，构建融合特征加权的深度学习模型，通过设计动态加权模块抑制噪声干扰并强化关键特征的影响，结合sLSTM的长期依赖捕捉能力与Transformer的并行计算优势，实现多尺度特征的协同优化处理；最后，构建MSICPO-VMD-WF-sLSTM-Transformer混合预测模型，以进行电价预测。实验结果表明，相较原始CPO算法和其他传统优化算法，MSICPO算法实现了VMD参数优化中最优解精度和优化效率的精细化平衡，MSICPO-VMD-WF-sLSTM-Transformer模型在预测精度方面表现良好，拟合度系数（R²）达到了0.95。此外，跨区域数据预测实验也进一步验证了该模型在不同区域电力市场中的适用性和可泛化性。所提方法不仅为智能优化算法的改进与多特征预测技术的发展提供了理论参考，还为复杂电力市场下的短期电价预测提供了高精度、强泛化的解决方案。

关键词: 电价预测, 特征加权, 冠豪猪优化算法, 变分模态分解, 深度学习

Abstract: To address the issues of low efficiency in Variational Mode Decomposition (VMD) parameter optimization, insufficient feature expression capability of single prediction models, and feature redundancy in current short-term electricity price prediction methods, a short-term electricity price prediction method based on a Multi-Strategy Improved Crested Porcupine Optimization (MSICPO) algorithm and deep learning is proposed. First, the Crested Porcupine Optimization (CPO) algorithm is improved by introducing a Lévy flight strategy, periodic population mutation, and dynamic parameter adjustment mechanism to enhance its global search capability and convergence speed. The improved algorithm is then used to optimize the number of VMD modes and penalty factor, thereby improving the signal decomposition accuracy. Second, a deep learning model incorporating feature weighting is constructed. By designing a dynamic weighting module, noise interference is suppressed and the influence of key features is strengthened. The long-term dependency capture capability of sLSTM is combined with the parallel computing advantage of Transformer to achieve collaborative multi-scale feature optimization. Finally, an MSICPO-VMD-WF-sLSTM-Transformer hybrid prediction model is built for electricity price prediction. Experimental results show that, compared with the original CPO algorithm and other traditional optimization algorithms, the MSICPO algorithm achieves a refined balance between solution accuracy and optimization efficiency in VMD parameter optimization. The MSICPO-VMD-WF-sLSTM-Transformer model demonstrates strong predictive performance, with a coefficient of determination (R²) of 0.95. Cross-regional data prediction experiments further validate the applicability and generalizability of the model across different regional electricity markets. The proposed method not only provides theoretical references for the improvement of intelligent optimization algorithms and the development of multi-feature prediction techniques but also offers a high-precision, strongly generalizable solution for short-term electricity price prediction in complex electricity markets.

Key words: electricity price prediction, feature weighting, Crested Porcupine Optimization (CPO) algorithm, Variational Mode Decomposition (VMD), deep learning

中图分类号: 

• TP181