Research Of Security Assessment For Power Systems Based On Data And Mechanism

The online security assessment of power system based on the rapid, accurate and effective information acquisition system is of great importance for preventing security incidents, which is one of the new front subjects in power system development. For the rapid increasing operating data in power systems, fully utilizing it, fast and efficiently analyzing, processing and refining it to discover useful knowledge, which also belong to one of the key problems faced by electric power industry. Under the above dual backgrounds, based on the related theories and methods of knowledge discovery in massive data and combined with the mechanism of power system security problem, this dissertation has researched online security assessment approaches for power systems. The research includes four aspects:static voltage stability, inter-area oscillation stability, pre-fault transient stability assessment and post-fault transient stability prediction. This dissertation mainly finished the following works:(1) According to the deficiencies of the conventional methods for online static voltage stability assessment in the practical application, this dissertation proposes an approach for online static voltage stability assessment based on connotative relationships exploration (RE). First of all, the relationships between the power system operation variables and static voltage stability margin are given scores by maximal information coefficient (MIC) and Pearson correlation coefficient (PCC); Then the highly scored relationships are selected to build the assessment model, the input features of the model are corresponding to the highly scored relationships; Finally, when the real-time data of the selected variables are obtained, the voltage stability margin can be online assessed based on the model. The approach is tested and analyzed in a 21-bus system and a 1648-bus system, and it is compared with conventional methods. The influences of training set size, number and type of selected relationships, rank of selected relationships, measurement error, type conversion of PV to PQ bus, topology change on the approach are analyzed; The data processing speed of algorithm is analyzed; The approach is applied to a practical system. The approach has high accuracy. Compared with conventional methods, the approach has the following advantages:handling the problems of PMU data loss better, avoiding tedious debugging and the problem of over fitting; as a kind of transparent tool, the approach can provide a clearer relationship between the operation variables and the start of an instability event than black-box tools. The approach provides a lot of freedom for the selection of input features, which means that it will leave more selection space of input features for other monitored indices in power systems and can reduce the economic investment. Moreover, this dissertation gives a strategy for the approach and conventional data mining methods to hand the increased assessment error in the condition of system network topology change. The strategy is building a database of model and updating it in online application, which can avoid the failure of methods and achieve seamless online assessment. In addition, the approach has good robustness to PMU measurement and can accommodate the type conversion of PV to PQ bus well.(2) The conventional methods of online inter-area oscillation stability assessment have the deficiency in accuracy and incomplete considerations of variation factors in power system operation condition. This dissertation focuses on those problems and proposes a credible integrated approach for online inter-area oscillatory stability assessment based on connotative RE. First of all, the approach uses MIC and PCC to explore the connotative relationships between the power system operation variables and inter-area oscillation stability margin; Then the explored relationships are selected to build the assessment model, and credible decision-making rules for connotative RE are designed in the model; Finally, the model and the measured values of the corresponding variables of connotative relationships to achieve credible online inter-area oscillation stability assessment. The approach takes into account various factors in system operation condition (topology change, variation of power distribution among generators/loads, and variation of peak load/minimum load) in the model update stage. The approach is tested and analyzed in IEEE 39 system and a 1648-bus system, and it is compared with conventional methods. The influences of training set size, number and type of selected relationships, and system operation condition on the approach are analyzed. The data processing speed of algorithm is analyzed. Compared with conventional methods, the approach has the following advantages:detecting the potential error in regression and classification assessments based on credible decision-making rules, and avoiding the use of such unreliable results to achieve higher assessment accuracy. The approach has good robustness to the variation of operation condition.(3) According to the deficiencies of the conventional methods for online pre-fault TSA in accuracy, calculation speed and intuitive presentation of TSA results, this dissertation proposes a credible and visual approach for online pre-fault TSA, which is based on multiple ELMs. The approach uses credible decision-making rules for multiple ELMs; It uses integrated learning of multiple ELMs:randomly selecting features, training instances, hidden nodes, and activation function in the training, which can improve generalization capacity of the model. It gives a visualization processing for online TSA results of possible fault location. The approach is tested and analyzed in a 1648-bus system, and it is compared with conventional methods. The data processing speed of algorithm is analyzed; The approach is applied to a practical system. Compared with conventional methods, the approach has the following advantages:giving more accurate TSA results of regression and classification assessments; presenting visual TSA results according to diversity of fault location, which is convenient for the system operators to obtain a better view of transient stability level including geographically distributed information; having the characteristic of rapid training. In addition, the approach has a certain adaptability for the variation of operation condition, and it will be have good robustness to the variation if it uses model update.(4) The conventional methods for online post-fault transient stability prediction usually have not taken into account the probability distribution of various factors comprehensively. This dissertation focuses on this problem and proposes a generic probabilistic evaluation approach for comprehensively evaluating the accuracy of data mining models in online post-fault transient stability prediction. The approach is based on data mining model, and it takes into account probability distribution of five factors, including:fault type, fault location, fault clearing time, load level and network topology. It comprehensively evaluates the accuracy of a data mining model in post-fault transient stability prediction. Combined with decision tree (DT), the approach is tested and analyzed in IEEE 39 system. Compared with previous researches, the approach has the following advantages:it can evaluate the accuracy of a data mining model more comprehensively and objectively by taking into account basic factors of influencing transient stability and their probability distribution; it is independent of used data mining model and generic, it allows fair comparison of the accuracies of different data mining models. In addition, when the approach is applied to DT model, it can give accurate prediction results in a very short time with taking into account probability distribution of various factors.