The electric power system is one of the largest human-made artificial systems and provides a quality of life to humans never available before. Since the electrical energy is an integral part of our daily lives, prolonged power outages have significant negative social and economic impacts. To mitigate the consequences of power system malfunctions and serious incidents, various protective measures are implemented. One of these measures, which is related to the frequency stability of the power system, is under-frequency load shedding. The main objective of under-frequency load shedding is to restore the balance between the generated and consumed active power after a major deficit, which is achieved by disconnecting loads.
Decades of experiences with under-frequency protection show that existing under-frequency load shedding schemes, while fast and robust, are unable to adjust their response to volatile operating conditions due to static settings. This results in inadequately sized load shedding in situations for which conventional protection was not designed, which can lead to the power system blackout under extreme conditions. Over the years, this issue is becoming more and more evident as operating conditions in the modern power system are constantly subject to rapid and unpredictable changes caused by newly connected converter-interfaced generation. Thus, there is a need for a solution that enables the under-frequency load shedding protection to more accurately control the active power deficit under different operating conditions.
This dissertation deals with the decentralized (local) type of under-frequency load shedding scheme, which is activated based on the detection of typical frequency patterns and adjusts its operation to the prevailing operating conditions using short-term frequency prediction.
The first part presents current problems and challenges in ensuring frequency stability of the power system. The main challenge is to overcome the problems related to converter-interfaced generation.
The second part presents the theoretical background of the response of synchronous generators in case of a sudden active power imbalance in the power system. In fact, frequency is a local parameter whose behaviour depends strongly on the inertia constants of the individual generators, the synchronization coefficients and the transients. Transients at different locations in the power system cause different fluctuations that can have a negative impact on the operation of the under-frequency load shedding protection.
The third part carries out a comprehensive analysis of the existing literature in the field of under-frequency load shedding. Due to the large number of publications, it is not easy to find development trends and all potential opportunities for research. Therefore, this dissertation also presented the methodology to systematically evaluate the literature. The methodology is based on the use of dimensionality reduction techniques of data and clustering methods. For this purpose, 381 publications in the field of the under-frequency load shedding were evaluated with 15 general and 39 specific features. The former describe the type of under-frequency load shedding scheme, while the latter describe the methods and tools used to try to ensure the best possible operation of a particular type of scheme. It has been shown that based on the analysis of features, we can identify trends from different time periods and gaps for further research.
The fourth part proposes an innovative solution that can be classified as a local and hybrid (combination of passive and active) types of schemes, as its main purpose is to improve the functioning of the conventional (local and passive) scheme. The proposed solution maintains the existing settings of the under-frequency protection relays (i.e. frequency thresholds) and introduces in parallel an additional dynamic under-frequency load shedding stage using a small and specialized group of intelligent electronic devices equipped with machine learning and prediction functionality. The operation is based on four main steps: i) detection of (eight) typical frequency patterns by principal component analysis and k-nearest neighbours, ii) short-term prediction of the frequency response by fitting the frequency response model of the power system to locally acquired frequency measurements in a least-squares sense, iii) use of the prediction to determine the time and frequency band, until the next static stage of the conventional under-frequency load shedding protection is violated, and iv) automatic adjustment of the settings of the dynamic stage based on the information provided by the estimated time and frequency band in the take-off characteristics.
The fifth part presents an analytical model of conventional and proposed under-frequency load shedding scheme. Six indicators were analytically derived: the frequency-time response of the power system, the first frequency-time derivative, the steady-state frequency deviation, the minimum transient frequency deviation, the timing and number of load shedding stages and the take-off characteristic. The analytical models of the schemes and the derived indicators were then used to compare performance.
The sixth part tests the novel solution with dynamic simulations on a smaller (western part of the Slovenian power system) and a larger (Slovenian power system) test model. The criteria for operational efficiency were the total amount of disconnected loads, the minimum and maximum frequency reached and the value of the first frequency-time derivative immediately after the last intervention of the under-frequency load shedding. A comparison was made with the conventional scheme currently used in the Slovenian power system and a comparison with another advanced solution that has similar objectives. In addition, the impact of the deployment of the intelligent electronic devices, the number of stages and the operational uncertainties in power generation were analysed. The operation of the proposal in the case of cascading power plant outages was also demonstrated.
The seventh part presents the tools used and the computational and software requirements necessary for the practical implementation of the proposed algorithm.
The results show that the proposed solution maintains the speed and robustness of conventional under-frequency load shedding scheme while providing the desired flexibility. A conventional scheme combined with a dynamic stage with three sub-stages was found to be equivalent to a conventional scheme with dozens of static stages. A comparison of the efficiency with the conventional scheme showed that the improvement is most significant when the proposed solution is used in networks with low inertia. Moreover, the intervention of the proposed solution leads to less aggressive interference in the operation of the power system, which is reflected in a more continuous frequency response and the absence of over-shedding. The main advantage compared to other existing methods is the simpler and more cost-effective implementation in the existing power system and the fact that no communication is required.
|