【作者】 刘晓宁；

【导师】 石访； 闫红华；

【作者基本信息】 山东大学 ， 电气工程（专业学位）， 2024， 硕士

【摘要】 随着高比例新能源并入电网,电力系统动态响应发生显著变化。系统等效惯量下降,短路故障等大扰动事件下电网功角失稳风险增加。紧急控制是防御扰动造成大停电事故的第二道关键防线,常用的紧急切负荷手段对用户影响较大,大规模粗放切除方式社会容忍度较低。随着传统机组出力空间被压缩,同步机不再具备大容量紧急控制支撑能力,因此电网安全稳定控制迫切需要开发新型快速柔性调节资源。随着通讯技术发展与终端智能设备投入,广泛接入电网的风光电源具有巨大调节潜力。将风光电源聚合纳入电网管控架构,不仅可提高紧急控制措施的精细化程度,更能增强电网安全稳定防御能力。充分挖掘风光机组的灵活快速响应能力,将海量异构电源高效聚合为可控整体参与电网互动,可丰富电力系统安全稳定控制手段。然而风光聚合集群参与紧急控制在特性分析、能力评估以及控制策略方面鲜有研究。在上述背景下,本文对规模化分布式风光电源参与电网紧急控制优化匹配方法开展了研究,主要研究内容如下:(1)研究了基于长短期记忆网络和堆叠自编码模型相结合的机组状态监测方法。通过删除停机数据和识别离群异常点来构造机组正常状态运行数据集。利用LSTM神经层替换堆叠自编码模型的隐藏层,并拼接下层模型输入数据、隐藏层输出信息和解码层重构结果来构造堆叠模型的上层输入,以上层重构误差作为反映机组性能异常程度的多维指标。在此基础上利用指数加权移动平均法和马氏距离将重构误差映射为一维监测指标。同时利用参数优化后的支持向量回归模型实时生成自适应动态阈值,依据监测指标与阈值的关系反映机组实时状态。(2)研究了考虑新能源聚合等效集群紧急控制调节能力的系统级功角稳定控制策略生成方法。通过分析不同单体风光电源响应模式和调节参数的差异化,提出了集群紧急调节能力等效方法;针对风光机组参与调节时需要考虑的功率预测误差和响应误差,定义了集群参与响应的非线性代价因子。为避免大规模风光电源复杂建模问题,从网络能量角度定义支路势能函数来反映系统稳定裕度,进而构建系统层面功角稳定控制优化模型。为高效求解所提高维度非线性复杂模型,依据电气距离对模型边界条件进行约束,采用并行改进鲸鱼优化算法结合机电暂态仿真进行求解,缩短了紧急控制策略刷新周期。(3)研究了风光设备集群参与电网紧急控制的控制架构以及功率分配策略。基于风光机组物理外特性分析,提出以惯性函数和阶跃函数分别表征机组不同动态响应模式。同时利用概率分布生成群体设备的物理参数,并通过构建集群紧急控制架构将海量控制特性差异化的终端电源纳入管理。针对风电机组和光伏机组响应过程,提出了反映机组实时调节性能的机械损耗、运行状态以及响应比例代价等多方面指标。依据所提指标和功率预测结果对单体机组进行周期性调节能力与控制代价评估,进而为集群指令优化模型提供求解边界。电网进入紧急控制状态后,利用成熟商业求解器求解所提混合线性规划模型,并将分配结果迅速下发至终端电源完成响应。更多还原

【Abstract】 With the high proportion of new energy sources integrated into the grid,the dynamic response of the power system has undergone significant changes.The system equivalent inertia decreases and the risk of grid power angle instability increases under large disturbance events such as short circuit faults.Emergency control is the second key line of defense against major outages caused by disturbances.Conventional load shedding has a significant impact on users,and large-scale extensive removal methods have lower social tolerance.With the compression of traditional unit output space,synchronous machine no longer has the ability to support largecapacity emergency control.Therefore,the development of fast and flexible regulation resources is urgently needed for the safe and stable control of the power grid.With the development of communication technology and the deployment of terminal intelligent devices,wind and solar power sources widely connected to the grid have tremendous regulatory potential.Incorporating wind and solar power aggregation into the grid control structure not only improves the refinement of emergency control measures,but also enhances the defense capability of power grid security and stability.Fully tapping into the flexible and rapid response capabilities of wind turbines and photovoltaic arrays,and efficiently aggregating massive heterogeneous power sources into a controllable whole to participate in grid interaction can enrich the means of power system security and stability control.However,the participation of new energy clusters in emergency control has rarely been studied in terms of characteristics analysis,capacity assessment and control strategies.In the context mentioned above,this paper conducted research on the optimization and matching methods for large-scale distributed wind and solar power sources participating in grid emergency control,with the following main research content:(1)A unit state monitoring method based on the combination of long-short term memory network and stacked auto-encoder model is studied.The unit normal state operation dataset is constructed by removing shutdown data and identifying outliers through anomaly detection algorithms.The LSTM neural layer is used to replace the hidden layer of the stacked autoencoder model,and the upper input layer of the stacked model is constructed by concatenating the lower model input data,the hidden layer output data,and the decoding layer reconstruction information.The upper model reconstruction error is used as the multidimensional indicator reflecting the degree of abnormal unit performance.On this basis the reconstruction error is mapped to a one-dimensional monitoring metric using exponential weighted moving average and Mahalanobis distance.Meanwhile,the parameter-optimized support vector regression model is used to generate adaptive dynamic thresholds in real time,and reflect the real-time status of the unit based on the relationship between the monitoring indicators and the thresholds.(2)A system-level power-angle stability control strategy generation method considering the equivalent emergency control regulation capability of new energy cluster aggregation is studied.By analyzing the differentiation of response modes and adjustment parameters of different individual wind and solar power sources,a cluster emergency regulation capability assessment method is proposed.The power prediction error and response error that need to be taken into account when new energy units are involved in regulation,so the nonlinear cost factor of cluster response is defined.In order to avoid the complex modelling problem of large-scale new energy units,the branch potential energy function is defined from the perspective of network energy to reflect the system stability margin,and then the optimization model of power angle stability control is constructed.In order to efficiently solve the increased dimensional nonlinear complex model,the model boundary conditions are constrained based on electrical distances,and a parallel improved whale optimization algorithm combined with electromechanical transient simulation is used to solve the model,which shortens the emergency control strategy refresh cycle.(3)The power allocation strategy and control architecture for wind and solar equipment clusters participating in grid emergency control are studied.Based on the analysis of the physical and external characteristics of the wind and solar units,the different dynamic response modes of the turbine are characterized by inertial regulation and step function,respectively.The probability distribution is used to generate the physical parameters of the group equipment,and the mass-controlled characteristics of differentiated terminal power sources are incorporated into the management by constructing a cluster emergency control architecture.By analyzing the response process of wind turbine and photovoltaic(PV)turbine,various indicators such as mechanical loss,operation status and response ratio cost reflecting the real-time regulation performance of the turbine are proposed.Based on the proposed metrics and power prediction results,the regulation capability and control cost of individual units are evaluated to provide solution boundaries for the cluster command optimization model.When the grid enters an emergency control state,the proposed hybrid linear programming model is solved using a proven commercial solver,and the allocation results are quickly sent down to the terminal power sources to complete the response.更多还原