Funded Research Projects
Grid-customer Integrated Resilience Assessment and Enhancement for Modern Power Systems
Principle Investigator: Jimmy Chih-Hsien Peng | Grantor: National Research Foundation | 2018-2019
Singapore’s grid is undergoing rapid rise of distributed solar photovoltaics, which are intermittent energy resources, and aging of critical assets. These facts have raised serious concerns on the grid’s resilience against large disturbances and natural disasters. Meanwhile, modern grids offer consumer-centric services such as demand response (DR) that utilize consumers' participation and their innate demand flexibility to increase the grid resilience. However, the implementation of such consumer-oriented programs opens up a new avenue to attackers who seek to manipulate the power system: the very consumers who utilize the grid.
In this research, we seek to answer two overarching questions: (1) could the grid stability be influenced by a malicious entity attacking the DR infrastructure, and (2) develop a systematic and comprehensive resilience assessment of the power grid that utilizes price incentives for consumers to increase resilience during large physical grid disturbances. Regarding the first question, we would assess the impact of load patterns on the residential distribution system that arise as a result of an attack manipulating the consumers’ behaviour using the DR messaging service. This requires a high resolution and scalable bottom-up model for the consumers’ response to a DR event. Our preliminary simulations demonstrate that significant undesirable network effects could indeed be caused, including but not limited to voltage and power flow violations. Towards the latter question, we aim to develop a data-driven resilience assessment methodology that bridges the end-user social behaviour (customer level), the critical electric assets (component level), and the system operation (grid level).
This research would enable us to collect end-use customer data, perform preliminary analytics, design and implement surveys to capture consumers’ behavioural patterns that would allow us to populate our models and develop attack scenarios. It would further enable us to develop systematic methodologies for power system resilience assessment and reinforcement at the full proposal stage.
Design of Future Residential Apartment Microgrids
Principle Investigator: Jimmy Chih-Hsien Peng | Grantor: Ministry of Education | 2016-2019
This project studied the operation of residential apartments as microgrids are capable of operating either in isolation, or in connection to the power grid. Operating such microgrids could potentially increase the efficiency of residential energy consumption. By intelligently coordinating the various local generation and storage, while incentivizing consumers to flexibly use certain appliances such as washing machines, dryers, and dishwashers, the microgrid operator could ensure lowering of energy costs, and reduce overall emissions from the power grid.
Specifically, to allow for residential buildings to be operated as microgrids, local generation sources such as solar photovoltaic generation, battery storage, and loads in the residences must be controlled in a coordinated fashion. This project focused on developing topologies and the associated control philosophies for ensuring that such operation is possible in a stable manner. Mathematical modelling and studies were performed in both the Matlab/Simulink and RTDS platforms to validate that coordination between these resources is possible while ensuring a low cost of infrastructure, and that stability of the system is maintained at all times. To ensure that the proposed control is feasible even when scaled to larger systems, a decentralized approach was adopted, wherein each agent in the system operates only based on locally available information such as voltage and current measurements at its own terminals. Such a design is further resilient to communication failures.
Next, consumer-behavior models were developed to generate the residents’ load profiles under business-as-usual conditions. These models were subsequently applied to assess the available flexibility in shifting certain loads in order to improve the consumption efficiency. Novel baseline estimation algorithms were then developed to allow us to assess how much a resident contributes to this increase in efficiency by changing his/her energy consumption behaviors.
Finally, when multiple microgrids are connected together—like multiple buildings in a block connected to each other—theories were developed to understand how such a system may lose stability when disturbances occur in the system. To quantify the impact of the critical parameters on the stability, a scalable sensitivity criterion was derived, which has the advantage of low computational complexity, while retaining its accuracy. Based on this criterion, a method to regain the stability of the system in case of real-time contingencies was designed and demonstrated in this project.
To test the accuracy of our mathematical models and control method, a scaled-down hardware test bed was developed in our laboratory. Experiments showed the close correspondence between the theory and experimentation, proving the accuracy of the modelling process and envisioned performance.
Event-Driven Methods for Demand Response in Electric Grids
Principle Investigator: Jimmy Chih-Hsien Peng | Grantor: NUS-HU Berlin Joint Program | 2017-2018
The traditional approaches for implementing event-based Demand Response (DR) have been static, and do not involve feedback to the consumers regardless of their performance during the DR event. This may however lead to an incomplete system-wide response, thereby forcing the utility to employ direct load control to achieve the required response, or buy additional generation reserve from the spot market. To mitigate this inefficiency, we propose closing the loop through an incentive control for residential DR participants enabled using event stream monitoring. By realizing the latter in an adaptive and distributed manner, we reduce the data communication and computation overhead involved in the decision making process. Using simple assumptions, we demonstrate that methods for event stream processing could be used to ensure that the scheduled DR is achieved completely. Therefore, the proposed implementation allows for scalable, effective, privacy-preserving, and robust implementation of incentive-based residential DR that ensures full overall compliance of the flexible resources to the DR task.
Past M.Sc. Research Projects
Generative adversarial networks for evaluating power system stability
Student: Cao Xilei | 2018-2019
In islanded systems with droop-controlled sources, the droop coefficients need to be tuned in real-time using supervisory control to maintain asymptotic stability. In contrast to offline tuning methods, online domain-of-stability estimation yields non-conservative droop gains in real-time, ensuring good power sharing performance as the operating point varies. The challenge in the conventional online domain-of stability estimation process is its unscalability and high computational complexity. In this paper, an efficient alternative using conditional Generative Adversarial Networks (cGANs) is described. We demonstrate that the notion of power system stability can be learned by such deep neural networks, and that they can offer a scalable alternative to conventional domain-of-stability estimation methods in islanded distribution systems. The implementation of cGANs-based stability assessment is described for an LV distribution test case and its advantages demonstrated. The arXiv link to the full conference paper can be found here.
EV charging optimization based on day-ahead pricing incorporating consumer behavior
Student: Zhang Qun | 2018-2019
With the increasing penetration of electric vehicles (EVs) into the automotive market, the electricity peak demand would increase significantly due to home-EV-charging. This paper tackles this problem by defining an ‘ideal’ EV consumption profile, from which a day-ahead pricing model is derived. Based on historical residential EV-use data ranging over a year, we demonstrate that the proposed optimization process results in a pricing profile that achieves a dual objective of minimizing the total electricity cost, as well as the peak aggregate system demand. Importantly, the proposed formulation is simple, and accounts for the trade-off between consumer convenience in terms of the number of available charging slots during a day and the reduction in the total electricity cost. This technique is demonstrated to be scalable with respect to the size of the community whose EV charging demands are being optimized.
Relationship of media bias and public opinion on climate change
Student: Liu Zhuolong | 2018-2019
It is widely known that climate change, also referred to as global warming, is unambiguously happening. And, as the IPCC concluded in 2013, "It is extremely likely that human influence has been the dominant cause of the observed warming since the mid-20th century." However, debate about the causes and the necessity of government action has never stopped. Research shows that it is related with the political leaning of the media outlet. In this report, we collected and analyzed more than 5000 news articles from 9 newspaper agencies with different political leanings using data mining techniques, natural language processing (NLP) techniques, and machine learning (ML) algorithms. The results show that political leaning has a limiting impact on this topic. Most people in US believe that climate change is happening, and we need to tackle it.
Domain of stability characterization for hybrid microgrids considering different power sharing conditions
Student: Aderibole Adedayo | 2015-2017
Renewable energy resources have become an increasingly popular choice for meeting the world’s growing energy demand. More attention is being paid to green-house gases (GHGs) emission and their impact on climate and the environment. The most common method of integrating renewable energy resources is by connecting them to the distribution network of existing power systems, and as such, they are generally referred to as distributed generation (DG). The rise in the participation of distributed energy resources (DERs) in electricity production has led to the development of “microgrids”. A major challenge in all power systems is to ensure stability during steady-state and transient conditions. Since the interconnection of large conventional power systems in the 1950s, power systems oscillations are a well-known phenomena studied by researchers over the years. Microgrid reliability and security are being improved by interconnecting microgrids to form multi-microgrid (MMG) structures. Because of the similarity between conventional power systems and microgrids, oscillations are expected to be present in microgrids and MMGs. However, due to diversity of DGs which make up MMGs, the nature of oscillations inherent to MMGs differ from conventional power system oscillations. For microgrids comprising of inverter-based and synchronous-based distributed generation, the impact of equal and unequal active power sharing conditions on the dynamic stability is analyzed. Small-signal stability analysis is employed to characterize the stability of the microgrid. The sensitivity of low-frequency eigenvalues to changes in active power droop gains is studied and a “domain of stability” region is proposed to assess the stability of a microgrid under different power sharing conditions. The different operating regions within the domain of stability are defined. Based on the sensitivity analysis and the domain of stability, the optimal droop gain for the diesel generator to improve the stability of the microgrid is determined. For MMGs, the small-signal stability of a MMG consisting of inverter-based and synchronous-based distributed generation is evaluated. Local and inter-microgrid oscillatory modes present in the MMG are characterized by modal analysis. Local and inter-microgrid oscillatory modes similar to conventional interconnected power systems oscillations are observed. In addition, a different “dominant” oscillatory mode caused by interactions between the synchronous generators and inverters in the MMG is identified. Parametric studies show that MMG oscillations exhibit undesirable damping characteristics for certain conditions and trade-offs must be made between achieving good transient and dynamic performances. Consequently, power system stabilizers (PSS) are developed for the DGs to improve the transient and dynamic performance of the MMG. From the studies carried out in this work, it is concluded that low-frequency oscillatory modes arise in microgrids and MMGs as a result of the interactions between the DGs
Optimal sliding-mode load frequency control with high penetration of variable distributed energy resources
Student: Maksymilian Klimontowicz | 2013-2015
Frequency stability in the conventional electrical power systems is highly dependent on the active power management. During each second of power system operation active power demand changes in a random way. It means that some loads are connected and at the same time some others are disconnected what eventually creates mismatch between generated and con- sumed power. From the load point of view, frequency sensitive devices like engines or clocks require quasi constant frequency to work properly. In the transmission system, power losses like hysteresis and eddy currents in transformers are frequency dependent. At the generation side, it is essential to prevent rotor angle from exceeding maximal threshold value; thus avoiding generators becoming out-of-step. There is also third aspect, which is dictated by interconnection between power system areas. Tie line power flow is determined by contracts between utilities so within the interest of both sides it is crucial to maintain tie-line power flows as scheduled. In reality it comes down to redress power outputs from generating units within strictly determined time. Many control techniques were proposed to solve the load frequency control problem. Moreover, current optimization methods and high performance computers allow engineers to optimize complicated linear and nonlinear problems. Among optimization techniques, heuristic based genetic algorithm optimization is utilized in this work. This thesis presents a comparative study of conventional and sliding mode control approaches towards load frequency problem. Optimal controllers are applied to the three area generalized system model and two-area non-linear power system model to validate designed control strategies. And finally distributed energy resources are introduced to show that proposed structure outperforms conventional techniques.
An enhanced state estimator for bad data detection using PMU measurements
Student: Abubeker Alamin | 2013-2015
The deployment of phasor measurement units (PMUs) by power electric utilities to enhance the operating capabilities of state estimators is an emerging trend around the world. In the literature, several publications introduced PMU measurements into the current state estimators to boost their estimation accuracy. However, they could not replace all the existing conventional measurements due to their high cost. Instead, PMUs are being deployed gradually in few numbers by many utilities. One of the essential functions of a state estimator that could potentially benefit from PMU technology is bad data detection and identification. Bad data are gross errors coming from flawed measurement devices and has the potential to affect the estimation results leading to incorrect information of the system status. Therefore, state estimators are required to be equipped with advanced bad data detection techniques. One of the most commonly used bad data detection techniques is the largest normalized residual test (LNRT). However, it is known to fail with certain measurements known as critical measurements.
In this thesis, a state estimator (SE) based on weighted least squares (WLS) was developed and evaluated against Iterative Kalman Filter (IEKF). For bad data detection, largest normalized residual test (LNRT) was integrated into the WLS estimator, and a detailed analysis of its detection threshold was evaluated. Finally, the LNRT capability was enhanced by incorporating PMU measurements at certain locations to enable it to detect bad data within critical measurements. An IEEE 14 Bus test system was simulated in Matlab under various scenarios. The results demonstrated the enhancement of LNRT detection capability by introducing few PMUs into the system and strategically placing them to eliminate critical measurements. This made LNRT detect and identify any bad data irrespective of its location. Moreover, variation of the value of detection threshold with redundancy ratio dictating against fixing the threshold was observed. Generally, the proposed technique showed an improvement on the bad data detection capability and state estimation results accuracy.
Robust damping control strategy for mitigating inter-area oscillations
Student: Younes Isbeih | 2013-2015
Robust control theories and filtering techniques provide promising solutions for damping low-frequency inter-area oscillations. The objective of this research was to develop a systematic procedure of designing a distributed two-stage damping control system for power grid inter-area oscillations by applying separation and identification filter accompanied by robust control techniques. Practical considerations were emphasized in the proposed control design.
The first practical consideration was to select the control stabilizing signals and control site locations. Residual analysis were used to study the candidate signals and to evaluate their comparative strength. The most effective control signals were found to be line power flows and injected currents.
The second consideration was the robustness of the designed controller. System identification and separation were used to decouple the system in presence of faults and disturbances into three variations; fault and disturbance-free, fault-dependent and disturbance-dependent. This approach minimized the stringent trade-offs inherent in the control design objectives such as reference tracking, disturbance rejection and noise attenuation. As a result, the control design could be formulated into two main objectives. The first goal was related to reference tracking and oscillation damping which was achieved using the fault-free subsystem. The second aim referred to fault and disturbance suppression which was achieved through fault and disturbance dependent subsystems. The control synthesis was formulated as mixed sensitivity H∞ output feedback problem with pole placement. Linear Matrix Inequality (LMI) approach was used to formulate the problem and to solve for a stabilizing controller.
The design procedure of two-stage robust control damping system was illustrated by two study systems. The first study was a single machine to infinite bus system. The second one was a two-area four-machine system.
The proposed architecture considerably minimized the overshoot when pole placement was not applied. For example, the overshoot was reduced by 41.98% in the first study case and by 17.60% in the second case study. If pole placement was required to achieve the minimum damping ratio, the improvement introduced by two-stage controller was less. For example, the overshoot was reduced by 0.4% only in the second case study. These results were drawn in com- parison with the conventional centralized robust control. The proposed architecture successfully suppressed the impact of disturbances and faults on the operation of the system.
Improved off-nominal accuracy in Phasor Measurement Unit
Student: Salish Maharjan | 2013-2015
Phasor Measurement Units (PMU) provide time-synchronized measurements of magnitude, phase angle, frequency, and rate of change of frequency at higher reporting rate in comparison with the conventional remote terminal unit (RTU) used by SCADA. Discrete Fourier Transformation (DFT) used by PMUs, usually designed at system frequency for phasor computation. As a result, leakage effect increases at large off-nominal frequency of the grid, which degrades the estimation from DFT. In order to enhance the accuracy of DFT during wide frequency deviations, two types of method based on frequency information were proposed in this research. The ﬁrst approach is known as variable sampling time (VST) and the second is sample value adjustment (SVA). For frequency information, two estimators such as Biased Jacobsen and recursive least square (RLS), were used in VST and SVA methods, respectively. The combination of frequency and phasor estimators constitute an enhanced version of DFT, suitable for PMUs.
The MATLAB model of the proposed PMU designs were tested with various types of input signals as speciﬁed in the IEEE C37.118.1-2011 standard. The compliance of the designs with the standard were evaluated for steady and dynamic state by help of indices such as Total vector error (TVE), frequency error (FE), and rate of change of frequency error (RFE). Both the design demonstrated TVEs less than 1% and 3% at steady and dynamic states, respectively. Furthermore, the second design using SVA performed better than VST in terms of TVE and reporting rate. Because of this, the second design was implemented in prototype PMU and the measurements from it were visualized and recorded by LabVIEW. The recorded measurements were validated by determining the TVE and were in compliance with the standard.
Damping control loops for mitigating power oscillation with support from wind farms
Student: Abdulla Al Shammari | 2013-2015
The increasing wind penetration in transmission grids alters the existing system stability. To maintain grid stability, wind parks are required to provide damping performance to mitigate electromechanical oscillations caused by their intermittent generatioan profiles. The corodination of wind parks and synchronous generators to damp oscillations is investigated. The proposed control scheme could enhance the rotor-angle stability under various generation and load diversity.