Can we automate the energy exchange of a power trader? To address this challenge, we present the Meta Agent Learner (MAL). The MAL is a tiered and multi-policy energy trader. It comprises data analytics (DA), a deep sequence-to-sequence recurrent neural network (DS2S) and reinforcement learning (RL). The DA phase draws knowledge out of the sheer flow of data. The DS2S phase creates wisdom and provides the intelligence for decision making. The RL phase senses and learns from the market to act strategically. We demonstrate the MAL in a scenario of a price-taker wind farm with a hydro plant. The testbed is real data from the NordPool and East Denmark (DK2). More specifically, electricity consumption, wholesale and balancing prices, cross border energy exchange, and weather conditions. The MAL optimizes the combined production of the wind farm and hydro pumped storage. Runs the hydro plant such that spillage of wind power is avoided or stores cheap market electricity. The performance is benchmarked with three traders.

中文翻译：

---

无补贴的可再生能源交易：一种元代理方法

我们可以自动化交易者的能量交换吗？为了解决这一挑战，我们介绍了Meta Agent Learner（MAL）。MAL是一个分层的多策略能源交易商。它包括数据分析（DA），深度序列到序列递归神经网络（DS2S）和强化学习（RL）。DA阶段从纯粹的数据流中汲取知识。DS2S阶段创造智慧并为决策提供情报。RL阶段感测市场并从中学习以采取战略行动。我们在带有水力发电厂的价格接受风电场的情况下演示了MAL。测试平台是来自NordPool和东丹麦（DK2）的真实数据。更具体地说，电力消耗，批发和平衡价格，跨境能源交换和天气情况。MAL优化了风电场和水力抽水蓄能的结合生产。经营水力发电站，避免风能溢出或存储廉价的市场电力。该性能以三个交易者为基准。