【摘要】 针对燃煤机组存在碳排放数据时滞、有效数据相对稀少和化石能源消费数据难以准确获取的问题，提出了一种基于虚拟样本生成的燃煤电厂碳排放预测方法，首先利用原始样本，构建基于随机森林的初始燃煤电厂碳排放预测模型；然后，利用t分布随机邻域嵌入(t-SNE)算法和初始燃煤电厂碳排放预测模型生成虚拟样本；最后将这些虚拟样本与原始训练样本相结合，对预测模型进行训练，以验证优化效果。将集成不同虚拟样本数目下的预测模型精度进行比较，发现集成虚拟样本数目为140时预测模型的精度最高，与基于原始样本的燃碳排放预测模型相比，均方根误差(RMSE)平均降低3.87%,平均绝对百分比误差(MAPE)平均降低10.6%。结果表明，通过集成虚拟样本提高了燃煤电厂月度碳排放量的预测精度。更多还原

【Abstract】 Aiming at the problems of time lag of carbon emission data, relative scarcity of effective data and difficulty in accurately obtaining fossil energy consumption data in coal-fired units, a carbon emission prediction model for coal-fired power plants based on virtual sample generation is proposed. Firstly, the original sample is used to construct the carbon emission prediction model of initial coal-fired power plants based on Random Forest. Then, the t-distributed stochastic neighborhood embedding(t-SNE) algorithm and the initial coal-fired power plant carbon emission prediction model are used to generate virtual samples. Finally, these virtual samples are combined with the original training samples to train the prediction model to verify the optimization effect. The accuracy of the prediction model under different number of integrated virtual samples is compared. The results show that the accuracy of the prediction model is the highest when the number of integrated virtual samples is 140. Compared with the prediction model of carbon emissions based on the original samples, the root mean square error(RMSE) is reduced by 3.87% on average, and the mean absolute percentage error(MAPE) is reduced by 10.6% on average. That means the prediction accuracy of monthly carbon emissions of coal-fired power plants is improved by integrating virtual samples.更多还原