梯级水电站群长期优化调度混合智能算法及并行方法研究

发布时间：2018-03-09 21:52

本文选题：梯级水电站群　切入点：优化调度　出处：《大连理工大学》2014年博士论文　论文类型：学位论文

【摘要】：近年来,我国水电事业的发展速度和规模突飞猛进,已跃居世界水电大国首位,并形成了或将形成一大批具有电站级数多、装机容量大、输送电范围广等特点的大规模梯级水电站群。这类大规模梯级水电站群的优化调度问题具有非常显著的高维数、非线性、多阶段、约束强耦合等特点,动辄十几级甚至几十级电站规模使“维数灾”问题将变得尤为突出,求解难度越来越大,计算效率越来越低。采用传统优化方法求解呈现一定的局限性,已无法满足电网和水电公司的精细化调度需求,探索合理高效的求解算法是水电调度工作亟待解决的重要课题。因此,针对上述问题,本文以西南地区大型流域红水河和澜沧江梯级水电站群为工程背景,在深入分析梯级水电站群长期优化调度特性的基础上,开展不同智能算法相结合的混合智能算法研究,并结合近年来兴起的多核并行技术提出了动态规划及其改进算法的并行求解方法。主要成果如下： (1)针对粒子群算法早熟收敛、易陷入局部最优、停滞等不足,提出了梯级水电站群长期优化调度自适应混合粒子群进化算法。该算法引入混沌优化的遍历性生成初始种群,提高初始种群的个体质量；通过粒子能量、粒子能量阈值、粒子相似度和粒子相似度阈值来描述算法的自适应变化以及群体进化程度,同时结合遗传变异思想进行粒子操作,提高初期全局搜索以及后期局部精化能力,抑制算法早熟；采用一种基于邻域的随机贪心搜索策略解决算法后期进化速度慢的缺点。红水河梯级水电站群研究实例结果表明,自适应混合粒子群进化算法能够获得优于粒子群算法的计算结果,与逐步优化方法相近,但计算时间优于逐步优化方法。因此,自适应混合粒子群进化算法是求解大规模梯级水电站群优化调度可供选择的一种求解方法。 (2)针对遗传算法易早熟且局部搜索能力差等缺点,提出了梯级水电站群长期优化调度自适应混沌整体退火遗传算法。该算法通过混沌优化的遍历性特点随机生成初始种群,提高初始种群的个体质量；从种群多样性对算法的影响考虑,利用父代参与竞争的整体退火选择方式选择杂交母体,避免种群早熟及过早收敛；同时,采用自适应交叉、变异算子提高算法局部搜索能力及收敛速度,避免算法走向局部最优。红水河梯级水电站群研究实例结果表明,自适应混沌整体退火遗传算法的计算结果优于遗传算法与逐步优化方法,且计算耗时较逐步优化方法更少。因此,自适应混沌整体退火遗传算法为大规模水电系统优化调度提供了一种合理有效的求解途径。 (3)针对动态规划方法求解大规模水电系统优化调度易造成计算耗时过多以及求解效率较低的情况,以离散微分动态规划(DDDP)确定性模型和随机动态规划(SDP)模型的并行化求解为例,在详细分析递推公式并行性基础上,分别设计DDDP多阶段和SDP单阶段内所有离散变量组合在递推公式中的求解运算作为并行计算父任务,利用Fork/Join多核并行框架,提出了梯级水电站群长期优化调度多核并行动态规划方法。以澜沧江下游梯级水电站群优化调度为应用实例,在两台不同多核资源配置中展开方法验证。计算结果表明,并行方法可充分利用多核资源,大幅缩减计算耗时,提高求解效率。而且,提出的并行求解设计思路可为其他动态规划方法的并行化提供参考和借鉴。 (4)针对极端干旱气候易造成梯级水电站群出力破坏深度剧烈从而影响电网安全稳定运行的情况,提出了考虑系统发电保证率和破坏深度的梯级水电站群长期优化调度多核并行机会约束动态规划方法。该方法将系统发电保证率要求作为机会约束纳入模型中,采用惩罚函数法处理出力破坏深度控制要求作为递推方程惩罚项。在逐时段递推最优决策过程中,应用Fork/Join多核并行框架实现单时段内所有变量组合的递推求解并行化。根据动态更新的拉格朗日乘子和惩罚系数迭代生成多组优化调度规则,通过模拟调度的系统发电保证率和最大破坏深度进行筛选,再以多年平均发电量最大为准则选取最优调度规则。澜沧江下游梯级水电站群优化调度应用实例结果表明,所提方法得到的优化调度规则可兼顾系统发电保证率和出力破坏深度控制,并实现系统发电量最大,能够有效响应极端干旱气候对水电系统的影响,为电网安全稳定运行提供了强有力的技术保障；方法并行化可大幅削减计算耗时,为更深入了解梯级水电站群调度运行规律提供了更高效的手段。最后对全文做了总结,并对有待于进一步研究的问题进行了展望。
[Abstract]:In recent years, the development speed and the scale of China's hydropower industry by leaps and bounds, has been ranked the world's first large hydropower, and the formation or the formation of a large number of power series, large capacity, wide range of transmission characteristics of large-scale cascade hydropower stations. The large-scale optimization problem of cascade hydropower station group has the high dimensionality, very significant nonlinear, multi stage, characteristics of the constrained strong coupling, at ten level or even dozens of hydropower station scale makes the "dimension disaster" problem will become more prominent, is more and more difficult, the computational efficiency is more and more low. By using the traditional optimization method has certain limitations, has been unable to meet the grid and fine scheduling needs of hydropower company, explore the reasonable and efficient algorithm is an important task of hydropower dispatching to be solved. Therefore, in view of the above problems, the large area south of West Red River Basin and Lancang River cascade hydropower stations as the engineering background, in-depth analysis of cascade hydropower station group based on the characteristics of long-term optimal scheduling, hybrid intelligent algorithm research of different combination of intelligent algorithms, combined with multi-core technology in recent years the rise of a parallel solving method of dynamic programming algorithm and its improvement the main results are as follows:
(1) according to the particle swarm algorithm premature convergence, easy to fall into local optimum, stagnation, the long-term optimal operation of multi adaptive hybrid particle swarm optimization algorithm of cascade hydropower station. The initial population generation the ergodicity of chaos optimization algorithm, improve the individual quality of the initial population; by particle energy, particle energy threshold, particle similarity and the particle similarity threshold to describe adaptive algorithm and swarm evolution degree and particle manipulation combined with genetic variation thought, improve the early and late local refinement of the global search ability, suppression algorithm precocious; uses a neighborhood random greedy search strategy to solve the shortcomings of the algorithm at the end of slow evolution. Based on the Red River cascade hydropower research example the station group showed that the calculation of adaptive hybrid particle swarm optimization algorithm can obtain better results of particle swarm algorithm, and by The step optimization method is similar, but the computation time is better than the stepwise optimization method. Therefore, adaptive hybrid particle swarm optimization algorithm is an alternative solution for large-scale cascade hydropower stations.
(2) the genetic algorithm premature and poor local search ability and other shortcomings, puts forward the long-term optimal scheduling of cascade hydropower stations adaptive chaotic whole annealing genetic algorithm. The algorithm through the ergodicity of the randomly generated initial population chaos optimization, improve the individual quality of the initial population; considering the effect of population diversity of the algorithm, the use of the parent to participate in the competition of the whole annealing selection of hybrid matrix, avoid premature convergence and premature convergence; at the same time, adaptive crossover, mutation operator to improve the algorithm of local search ability and convergence speed, avoid the algorithm to a local optimum. The Red River Cascade Hydropower Station group case study results show that the calculation results of chaos adaptive whole annealing genetic algorithm better than genetic algorithm and stepwise optimization method, and the computation time is gradually less. So the optimization method, adaptive whole annealing genetic chaos The transmission algorithm provides a reasonable and effective solution to the optimal scheduling of large-scale hydropower systems.
(3) according to the dynamic programming method to solve large-scale hydropower scheduling caused by too much computational time and computational efficiency are lower, the discrete differential dynamic programming (DDDP) model of deterministic and stochastic dynamic programming (SDP) parallel solving model as an example, the detailed analysis of the recursive formula of parallelism based on design computation for multi stage DDDP and SDP single stage all discrete variable combination in recursive formula in parallel computing as the parent task, using Fork/Join multi core parallel framework, proposes a multi core group of long-term optimal scheduling of cascade hydropower stations parallel dynamic programming method. In order to lower Lancang optimal operation of cascade hydropower stations as an example in two, expansion method to verify the different multi-core resource allocation. The calculation results show that it can take full advantage of multi-core parallel method, greatly reduce the computation time and improve the efficiency of the algorithm. Moreover,. The parallel solution design idea can provide reference and reference for the parallelization of other dynamic programming methods.
(4) in the extreme arid climate caused by the cascade hydropower stations to output failure depth of intense affect the safe and stable operation of power grid, considering the power system to ensure the long-term optimal scheduling of cascade hydropower stations in multi core rate and failure depth of parallel chance constrained dynamic planning method. This method will ensure system of power requirements as an opportunity constraints are incorporated into the model, using penalty function method to deal with the power failure depth control requirements as punishment. In the recursive equation by period recursive optimal decision process, the application of Fork/Join multi core parallel framework for the realization of all the variables in the combination of single period recursive parallelization. According to the dynamic update of the Lagrange multipliers and penalty coefficient iteration multiple sets of optimal scheduling rules, the simulation system of power dispatching guarantee rate and the maximum failure depth were selected, with annual average generating capacity of the Select the optimal scheduling rule as the criterion. The lower reaches of Lancang River cascade hydropower stations optimal scheduling application results show that the proposed method has optimal scheduling rules can both guarantee rate and output power system failure depth control, and realizes the maximum power output, can effective impact of extreme drought climate on hydropower system, provides the technology a strong guarantee for the safe and stable operation of power grid; parallel method can greatly reduce the computation time, provides a more efficient means for more in-depth understanding of scheduling of cascade hydropower station group operation rules.
Finally, the full text is summarized, and the problems to be further studied are prospected.

【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TV737

【相似文献】