针对脑机交互系统发展中数据不足的问题，通过神经质量模型合成事件相关去同步（ERD）和事件相关同步（ERS）特征，节省模型训练时间，避免数据过拟合。引入了基于脑同侧运动感觉区μ/β节律的ROI神经元群模型，调整幅值的加减常数后，生成模拟ERD/ERS信号。实验证明，模拟信号与真实信号在共空间模式特征上相似，滤波和共空间模式特征提取后的机器学习分类准确率接近真实数据。混合不同比例的模拟和真实数据，对分类准确率的影响不大，验证了基于神经质量模型的模拟信号对ERD/ERS信号进行数据增强的有效性。这一方法有望在小样本数据集下用于算法创新和检验，同时可以缩短实验时间，为脑机交互系统的发展提供有力支持。

For the challenge of insufficient data in brain-machine interface （BMI） systems，this study employs a neural mass model to synthesize event-related desynchronization/event-related synchronization （ERD/ERS） features to augment limited training samples of electroencephalogram （EEG） and enhance the decoding performance.A region of interest （ROI） neural ensemble model， based on the μ/β rhythms within the motor cortex，is introduced to adjust amplitude parameters through precise constants，thereby generating simulated ERD/ERS signals.Experimental results demonstrate the resemblance between simulated and authentic signals in terms of common spatial pattern （CSP） features.The machine learning classification accuracy， post-filtering and CSP feature extraction，closely approximates that derived from authentic data.The negligible impact on classification accuracy as blending varying proportions of simulated and authentic data validates the efficacy of the simulated signals based on the neural mass model in enhancing ERD/ERS signals. This methodology holds promise for algorithm.