【摘要】 为减小高速磁浮系统谐波对牵引供电网产生的影响，解决谐波治理存在的时滞性问题，通常需要对谐波电流进行预测。对此，采用融合深度学习算法的组合模型相比于传统算法的表现更加出色。文章提出一种新的融合注意力机制的多智能体磁浮谐波预测算法，该算法通过麻雀搜索算法（SSA）优化变分模态分解（VMD）的相关参数，并利用该优化参数将原始电流信号分解为多个不同中心频率的谐波分量，将各个分量分别输入融合注意力机制的长短时记忆网络（LSTM）中进行时序预测，形成多个独立预测智能体，再对各个智能体预测结果进行重构，从而实现对高速磁浮谐波电流预测。在此基础上引入误差修正机制，进一步提高模型的预测精度。对上海高速磁浮牵引系统进行理论分析与仿真，采集网侧电流数据，并用所提出的算法模型对此数据进行试验和分析。结果表明，与其他模型相比，所提的预测模型在磁浮谐波电流预测方面具有较好的性能，并可使时滞性得到进一步改善。更多还原

【Abstract】 To mitigate the impact of high-speed maglev harmonics on the traction power supply network and address the time lag in harmonic control, predicting harmonic current is commonly necessary. In this context, a superior alternative to traditional algorithms is a combined model that integrates deep learning algorithms. In this paper, a new multi-agent maglev harmonic prediction algorithm with integrated attention mechanism was proposed. The algorithm optimized the parameters of variational mode decomposition(VMD) by sparrow search algorithm(SSA), and used these optimized parameters to decompose the original current signal into harmonic components with different center frequencies. Subsequently, all components were input into the long short-term memory(LSTM) network of the integrated attention mechanism for time series prediction, and multiple independent prediction agents were formed, followed by the reconstruction of each agent’s prediction results, so as to predict the harmonic current of the high-speed maglev. On this basis, the error correction mechanism was introduced to enhance the prediction accuracy of the model. Theoretical analysis and simulations of Shanghai highspeed maglev traction system were conducted, with the grid-side current data collected for experiment and analysis using the proposed algorithm model. The results indicate that, compared with other models, the proposed prediction model demonstrated superior performance in predicting maglev harmonic current, with the potential for further reducing the time lag.更多还原