| With the development of large-scale power grids and large-capacity units, adoption of ultra high voltage, introduction of market competition mechanism and energy reduction policy, the scheduling and control of power systems become more complex. In addition to the enhancement of power quality, it is necessary to perform the system coordination control. The Energy Management System (EMS) faces greater challenges.The forecasting, estimation and forecasting-estimation of system operation characteristics (nodal loads and system states) are important and indispensable tasks in EMS. Under the new situation, new ideas and methods could be introduced into EMS, then function for system operation characteristics analysis can be used for the current development of power systems. Therefore, the study on this subject is necessary and has practical significance.The analysis on system operating characteristics and its variation trend is an efficient method for solving the above problems. Load Forecasting (LF), Dynamic State Estimation (DSE) and State Forecasting-Estimation (SFE) are interrelated and most basic problems to be resolved. The model and algorithm of LF are fundamental and pivotal. The extended issue of DSE is resolved considering states variation trend. Concepts and theories of SFE are further studied. Finally new ideas are brought forward in connotation for the relation between SFE and scheduling decisions.LF is relatively a more traditional research topic. In recent years, with the extensive application of neural networks, fuzzy math, genetic algorithm, rough sets, and support vector machines, and increased requirements of power system operations and control, study on LF has attracted much attention once again. In order to complete scheduling decisions on the premise of foreseeing ability, modern EMS need not only forecasting message of aggregate load but also nodal active and reactive loads. This thsis maks a further study on an adaptive and dynamic multi-node active and reactive ultra-short term loads forecasting on the basis of former researches. LF is recognized as a complicated issue including space, time and attribute. By now, the equivalent concept and the overall situation thought are very important.Based on the analysis on load variation regularity and interactional relationship, it is found that different voltage levels divide the system into the many hierarchies. Different hierarchies reflect different load variation rules. The entire electrical power system is made up of multi-layers according to voltage level from high to low. Of course, load variation regularity is implicated in this hierarchical multilayer system and studies on multi-node load forecasting can be performed by multi-orientation information, such as variation regularity between layers, and self variation regularity of each node. Variation regularity of the lower layer system infiltrating into the higher layer system makes the higher layer system possess special regularity. Using the description of the dynamic behavior characteristics of nodal parameters, regularity of the lower layer can be reduced from regularity of the higher layer. Hence, no matter which level is to be analyzed, the first task is to grasp the regularity of aggregate active load and correlative information.In this thesis, aggregate load forecasting is implemented using RLS-SVM and the dynamic forecasting model of multi-node parameters (nodal active load distribution factor and nodal power factor) is solved by the Kalman filter technique. Finally a T-S fuzzy controller is introduced to implement the error correction.On the study of ultra-short term multi-node load forecasting, this thesis performs an in-deep analysis on state estimation, an important part of the EMS, which includes DSE and static state estimation. DSE considers statistical characters of system state variables in past periods. It has feature on state estimation and forecasting, posses predominance that static estimation lacks in terms of theory. DSE has already caught academia's attention.The thesis further analyzes DSE theory based on the Extend Kalman Filter (EKF) and points out two existent principal problems. Firstly, complex nonlinear properties or strong non-Gaussian of states would exist in the actual power system, so Gaussian filtering method with linear approximation may result in local suboptimal Bayesian filtering estimation. Furthermore large approximate errors may lead to large accumulative estimation errors. Secondly, linear method based on Taylor expansion is easily influenced by reference points. If current estimation values vary greatly with the true values, the deviation would further lead to linearization errors and imprecise filtering correction.The model and algorithm for self-adapting dynamic estimator are presented here. The new ideas embody two aspects. In the forecasting model, considering control action of nodal power states and self-regulation of system states, the integrated model for state forecasting is used to improve the prediction accuracy. In the filtering part, using Least Square Support Vector Machines (LSSVM) technology, self-adapting dynamic filter is formed with limited memory to increase estimation capability and computing speed.However, to construct the modern electrical network energy management system construction, the information about future load variation and the current states, which have obvious limitations, are unable to meet the needs of modern EMS. The existing state estimation, short of a forward-looking ability, is used to provide the reliable real-time database for the control center.Nodal load forecasting is relatively weak comparing with aggregate load forecasting. In the same way, reactive forecasting is weak comparing with active power forecasting. If forecasting only thinks that systemic variation tendency is completely stochastic and independent, not consider predictable scheduling and control rule and be away from system operation self physics regulation, then it is not comprehensive enough to analyze system states regulation regulation. Hence, timely, reasonable and accurately grasping the future power system running states appears more important. This task should be achieved by SFE.It is obvious that SFE has the close relation with power flow (PF), State Estimation (SE), LF, active frequency adjustment and reactive voltage control. With regard to function, PF can be used to analyze the appointed operation mode of history, current as well as future. SE can be used to analyze current system states. LF is used to assure future load demand characteristics. Active and reactive adjustments are used to deal with random imbalance between supply and demand. SFE can be viewed an as Extended comprehensive study of PF and SE in the time axis, as well as LF and generation regulation in the space axis. Therefore on the basis of the existing research, this thesis makes a further study on the model and algorithm of SFE and constructs the overall framework considering the nodal perturbation power. Using the idea of tracking estimation with state estimation results as starting values, the method accounts total changes caused by disturbance of quasi-steady process and considers nodal disturbance power as measurement induced by randomness of load demand, and the self and performing uncertainty of generators scheduling control plan. Weighted least squares and high order correct technology is employed to forecast and estimate variation trend of system state.On the basis of the above work, the thesis points out that forecasting and system decision-making are regarded by most researchers as two isolated individuals, which keep a stiff and mechanical weak link in the physical aspect. Under the new situation, their metaphysical concepts and ideas will likely be improved. How to make the forecasting serve system decision-making better will be one of the issues which related researchers and user face.Without a doubt, users have their special application experience on uncertainty forecasting to raise the benefit and reduce the risk, and also know how to forecast may serve them well. This kind of experience and the demand can help forecaster improve forecasting from the point of users to provide users more valuable forecasting information.The thesis adopts a combined forecasting model, and uses indexed and quantificational decision-making effect of automatic voltage control as reference to revise model parameters. Hidden interdependence relations between them are then found and mutual cooperation forecasting model is constructed in succession to realize the basic idea that forecasting leads decision-making and decision-making guide forecasting. In conclusion, this thesis rightly combines LF, DSE and SFE together, and makes a further and general study on operation variation regulation of power systems. It proposes a series of models and algorithms which are tested using the Shandong power grid. This work has achieved significant progress in this field. Of course, there still exit several problems to be investigated for the enrichment, development and perfectin of study, both theory and practice.
|
PDF Full Text Request