Optimization models, methods and algorithms applied to hydropower operations planning

加拿大水力发电装机容量为78359兆瓦，潜力是实际装机容量的两倍多。在这个气候变化的时代，各国都在关注清洁能源发电，水电就是其中之一。重要的是要有效地管理实际到位的水电系统，以最大限度地利用可用水的能源生产。目前的研究提案旨在制定新的战略，以便在运营的基础上有效地管理水电站。传统上，中期优化模型被用来确定一周内的储水量或总的预期能源产量。短期优化模型每天用于确定水电系统中水轮机和发电厂之间的准确水量分配。优化问题通常是随机的，因为在做出决策时，水库的流入、需求和/或能源价格是未知的。 这一研究计划的长期目标是建立数学模型和开发优化方法，以尽可能精确地解决中短期水电问题，并在这些模型的数值实现方面取得新的进展。此外，所开发的模型将在全球不同的水电系统上进行测试，以进一步改进所提出的模型和求解方法。 在较短的时间范围内，本提案分为三个目标，旨在培养两名博士生和两名硕士研究生。最近的趋势，如机器学习，是水电管理的有希望的途径，因为水电生产商可以获得大量数据。因此，该建议的第一个目标是从机器学习中获得工具，如神经网络，以生成流入场景作为随机优化问题的输入。第二个目标侧重于竞价概念，以加强短期水电问题的制定。大多数水电生产商在一个放松管制的市场中发展，竞相以高价出售能源，以较低的价格购买能源。将探索新的问题公式和求解方法，以改进现有的优化模型。使用涡轮机组合的建模方法很有趣，因为它们减少了决策变量的数量。最后，第三个目标涉及到计算发展，以解决水电问题，减少计算时间，以及所需的基础设施。水电生产商每天更新他们的流入预测，并要求他们一天多次做出有效管理系统的决定。需要迅速提供解决方案，这一目标涉及计算机集群上的不同并行化技术以及使用图形处理单元进行计算。

Canada has 78,359 megawatts of installed capacity of hydropower generated energy, and the potential is more than double the actual capacity. In this era of climate change, countries are focusing on clean energy generation and hydropower is one of them. It is important to efficiently manage the hydropower systems actually in place to maximize the energy production with the available water. The present research proposal aims at developing new strategies to efficiently manage hydropower plants on an operational basis. Traditionally, mid-term optimization models are used to determine reservoir volumes or total expected energy production throughout a week. Short-term optimization models are used on a daily basis to determine the exact dispatch of water between the turbines and power plants of the hydropower system. The optimization problems are usually stochastic, since inflows in the reservoirs, demand and/or energy prices are unknown at the time of making a decision. The long term objectives of this research program is the formulation of mathematical models and development of optimization methods to solve the mid- and short-term hydropower problems as precisely as possible and new developments in the numerical implementations of these models. Furthermore, the models developed are to be tested on different hydropower systems across the globe, to further improve the proposed models and solution methods. On a shorter time scale, the present proposal is divided into three objectives and aims at training two doctoral students and two master's students. Recent trends such as machine learning are promising avenues for hydropower management since the hydropower producers have access to large amounts of data. Therefore, the first objective of this proposal is to derive tools from machine learning, such as neural networks to generate inflow scenarios used as input to the stochastic optimization problems. The second objective focuses on bidding concepts to enhance formulations of the short-term hydropower problem. Most of the hydropower producers evolve in a deregulated market, competing to sell energy at high prices and to buy at lower prices. Novel formulations of the problem and solution methods will be explored to improve the current optimization models. Modeling approaches that use combinations of turbines are interesting as they reduce the number of decision variables. Finally, the third objective is concerned with computational developments to solve hydropower problems, to reduce computing time, but also the required infrastructure. Hydropower producers update their inflow forecasts daily and decisions to manage their systems efficiently are required multiple times a day. Solutions need to be available quickly and this objective addresses different parallelization techniques on computer clusters as well as the use of graphical processing units to perform calculations.