Data-enhanced applications for power systems analysis

data mining;cluster analysis;power systems analysis;sensitivity analysis;equivalent estimation;parameter estimation;bad data detection;reduced network modeling

Rogers, Katherine

关键词: data mining;    cluster analysis;    power systems analysis;    sensitivity analysis;    equivalent estimation;    parameter estimation;    bad data detection;    reduced network modeling;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/29740/Rogers_Katherine.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

As power system data becomes more abundant, there is an increasing need for applications which can make appropriate use of this data to perform power system analysis tasks.Data provides opportunities for performing event detection, diagnostics, and forensics on the system in a way which has not previously been possible.The integration of data and information systems remains a key challenge in taking full advantage of the smart grid. There is a need to “tap into the data” to discover relationships and to draw attention to patterns.Opportunities are provided by the data to develop and perform diagnostics.New functionality and applications in the smart grid can best be enabled by taking full advantage of available data.If not resolved, the lack of suitable advanced data analysis capability threatens to become a barrier to developing new intelligent grid operation and control paradigms. This need motivates the work in this thesis.There remains a great deal of opportunity to advance the state of the art, especially for developing suitable techniques to perform automated analysis and event detection based on the data.The work is presented with the anticipation of encouraging future extensions, especially with respect to the identification and classification of patterns in power system data.This thesis examines data mining and advanced data analysis techniques in the context of a number of specific power system applications.In particular, these applications concern the use of model data (sensitivities) to identify relationships, the data-enhanced estimation of network models, event identification from oscillation monitoring data, and dealing with the challenges of real-world data and data quality.Several important contributions of this work are the following.Analysis and results show that sensitivity and model data can be leveraged via correlation and cluster analysis to gain information about the expected or model-supported relationships and patterns in power systems. In particular, results exemplify these benefits for the areas of market power potential identification, coordinated control of resources, and in the creation of structure-preserving network reductions. Results also show that a space of network reductions which satisfy power flow solution equivalence exists and can be further explored by choice of desirable constraints. Another key contribution is the realization that the expected or model-supported patterns can be ascertained from actual data, which is done in the form of estimating equivalent models and network model parameters.Solutions are presented which highlight the benefit of data-over-time for estimating the structure of a reduced network model in the form of a matrix representing the system structure, and at a lower-level, for estimating parameters on individual transmission lines from historical data. The realization is that while data facilitates these advanced data-centric applications, there are also barriers to the progress of such applications. These limitations are a function of the data itself, and arise both with respect to the noise and error qualities of the data as well as from a lack of adequate representation of true characteristics present in the data, such as load fluctuations.Thus, it is imperative to understand, improve, and enhance the quality of real data.It is concluded that the ability to use and analyze real data is critical to implementing and advancing cutting-edge data mining solutions in power systems.The second half of the work in this thesis focuses on addressing these issues.Real data is shown to be useful for diagnosing problems. An oscillation monitoring event detection framework is presented, where results indicate that with information from nominal modes, it may be possible to correctly classify ringdowns to an originating event. The data quality (availability and integrity) issues are found to be critical to this line of work, and it is concluded that this is probably the most pressing area for immediate attention.Without trusting the data, one cannot trust decisions based upon the data.A key contribution is to relate the data and cyber dependencies of power systems and demonstrate how it is possible to take advantage of available information (including from cyber network monitors) to flag suspicious data sensors and measurement values and then take appropriate action. Results show that bad data from malicious sources can be detected without relying upon traditional residual-based schemes.The analysis is performed before such data is available to corrupt applications such as state estimation.In summary, this work presents a number of contributions for the enhancement of power system applications through the use of data and data mining techniques.