Abstract The rapid development of distributed generators and demand response management programs are transforming the traditional consumers to emerging prosumers. While, it is difficult to manage these prosumers because different types of energy are locally generated and consumed with the autonomous operations. For this purpose, this paper proposes a multi-energy forecasting framework based on deep learning methodology to simultaneously predict the electrical, thermal and gas net load of integrated local energy systems. First, the inherent multi-energy load and generation features of heterogeneous prosumers are qualitatively analyzed, and a hierarchical clustering framework is formulated to classify these prosumers into various aggregations to facilitate the multi-energy forecasting model. Then, a deep belief network based forecasting method is developed to extract the hidden features in multi-energy time series, thereby achieving the net-load prediction of numerous prosumers. Finally, the proposed multi-energy net load forecasting methodology is extensively and comprehensively validated using the real data from household-scale prosumers. The comparative results demonstrate the superiority and high forecast accuracy of the proposed methodology, and confirm its capability to cope with the multi-prosumer prediction problem with multi-energy carriers.

中文翻译：

---

具有异构产消者的集成本地能源系统的多能源净负荷预测

摘要 分布式发电机和需求响应管理程序的快速发展正在将传统消费者转变为新兴的产消者。然而，管理这些产消者是很困难的，因为不同类型的能源是在本地产生和在自主操作中消耗的。为此，本文提出了一种基于深度学习方法的多能源预测框架，以同时预测综合本地能源系统的电力、热力和燃气净负荷。首先，定性分析异构产消者固有的多能源负荷和发电特征，并制定层次聚类框架将这些产消者分类为各种聚合，以促进多能源预测模型。然后，开发了一种基于深度信念网络的预测方法来提取多能量时间序列中的隐藏特征，从而实现众多产消者的净负荷预测。最后，所提出的多能源净负荷预测方法使用来自家庭规模的产消者的真实数据进行了广泛而全面的验证。比较结果证明了该方法的优越性和较高的预测精度，并证实了其处理多能量载体的多产消费者预测问题的能力。