Accepted abstracts


Climate Change and Cholera… How can we plan for a cholera-free future in Bangladesh?

Deborah Shackleton

Cholera is endemic in Bangladesh and causes an average of 4500 preventable deaths in the country every year. The country is home to the Ganges Delta, the largest delta in the world, and native homeland of the cholera causing bacteria Vibrio Cholerae. As such, cholera incidence is hugely affected by weather and climate in the region, and is likely to worsen with climate change. This research plans to develop a model which is able to simulate the complex interplay between climatic and human factors and allow policy makers to identify which factors would make most efficient use of resources to best mitigate cholera incidence.


MCDM for water-related hazards modelling

Priscila Barros Ramalho Alves

More than half of the world population currently lives in urban areas, and over 500 cities shelter more than one million people worldwide. While the population growth can be observed in many countries, the degree of problems related to management varies greatly across the world.  Often more, people experience issues related to urban growth, changes in climate and deficiencies in management. Inefficient water systems and occurrence of precipitation extreme events, such as water scarcity and floods hazards, impose even more difficulties to cities planning. This complex context asks for a better cooperation in decision-making process, with an integration of all systems, sectors and parameters, in order to be effectively applied in management. The decision-making is now considered a process that combines different interactions of water resources (mainly the water cycle) with other city development activities (e.g., population density, land use changes, economic development) and climate change. In recent years, there is a trend in applying spatial analysis (also called as Geographic Information Science – GIS) and multiple datasets (or multiple criteria analysis) to better understand those interactions and support decisions in a realistic view.  Those tools are applied in both pre and post-phases of management with a range of objectives, that goes beyond the identification of most efficient strategies for mitigation but also to gain more concrete and truthful insights about interrelationships between datasets. This research applies the multi-criteria analysis in a GIS environment for floods and water scarcity hazards modelling. The study case is Campina Grande – Brazil. The city is inserted at a semi-arid region with dry climate and long water scarcity periods but also with a recurrence of floods cases. The research proposes a mixed-method approach with a combination of elements of quantitative and qualitative styles. ArcMap (version 10.5.1) and Mike URBAN software are being used as environments to model hazards-prone areas and also to support and identify the most efficient strategies for damage mitigation. Results indicate that the stratification of locational, topological and attribute information of the multi-criteria analysis in the GIS environment can be extremely useful for decision-making in real world water-related applications. This research expects to support the mixed-method approach for spatial integrated analysis of extremes events in urban environments.

Keywords: hazards modelling, mixed-method, multi-criteria, integrated analysis, decision-making


Multi-objective mixed H2/H∞ robust load frequency control loop shaping using particle swarm optimization

Deepak Kumar Panda

The frequency of the power system is dependent upon the balance between the load demand and the power generated. Large scale power systems consists of several control areas which exchange power through tie-lines in case of abnormal conditions. The control loop facilitating the balance between the load demand and power generation and the exchange between several areas, is known as load frequency control (LFC). Proportional Integral (PI) controller is generally used in the control loop, which has to be tuned based on several conditions.

An effective power system requires a robust communication infrastructure in order to facilitate decentralized control of multi area power systems and it’s a challenge to incorporate computing, communication and control at appropriate levels along with market signals in the real world power system operation. With the progression of wide area measurement device (WAM) such as phasor measurement units (PMU) transmission of signals to the control centre has become simpler. However the transmission of the signal involves considerable delay, due to the physical distance between the WAM devices and the control centre. The delay has to be taken into account as it can cause system instability, hence the controller parameters are tuned accordingly. Thus the parameters of the PI controller is tuned using an optimal H∞ controller by expressing the constraints as linear matrix inequalities (LMI) [1] which is solved using semi-definite programming methods. This tuning method helps to design better controllers considering communication delay as a significant uncertainty which can degrade system’s performance.

However LFC is a multi-objective control problem, as it is important to regulate the frequency deviation by rejecting the disturbances due to the random nature of load demand and renewable energy generation, thus maintaining tie line interchanges between multiple areas considering generation constraints and time delays as described above. Thus a mixed H2/H∞ control technique [2] will give a better solution to the problem rather than a single norm control technique.

The conflicting objectives of the H2/H∞ is studied by coupling them with multi-objective swarm algorithms. In multi-objective problems, there is a set of trade-offs thus resulting in numerous Pareto Optimal solutions, where all the objectives are simultaneously considered. Particle swarm optimization (PSO) models the social behaviour which guides the swarms of vectors into optimal search areas [3]. Hence, for this problem the H2/H∞ control objectives are summed using weighted conditions, and Pareto optimal points are generated using PSO based on the priori knowledge of the search space. The conflicting control objectives of H2/H∞ is studied based on the Pareto fronts obtained and the resulting solution is utilized in the load frequency control for the optimal multi area power system performance.


[1] Yu, Xiaofeng, and Kevin Tomsovic. “Application of linear matrix inequalities for load frequency control with communication delays.” IEEE transactions on power systems19, no. 3 (2004): 1508-1515.

[2] Khargonekar, Pramod P., and Mario A. Rotea. “Mixed H2/H∞ control: a convex optimization approach.” IEEE Transactions on Automatic Control 36, no. 7 (1991): 824-837.

[3] Hu, X. and Eberhart, R., 2002. Multiobjective optimization using dynamic neighborhood particle swarm optimization. In Proceedings of the 2002 Congress on Evolutionary Computation. CEC’02 (Cat. No. 02TH8600) (Vol. 2, pp. 1677-1681). IEEE.


EMOCS: Evolutionary Multi-objective Optimisation for Clinical Scorecard Generation

Diane P. Fraser, Edward Keedwell, Stephen L. Michell and Ray Sheridan

Clinical scorecards are an efficient method of assisting clinicians to stratify patients according to risk using evidence drawn from datasets of previous patient outcomes. A scorecard contains a (usually small) set of clinical measurements that are assigned a score, with higher scores usually indicating increased risk of mortality, or severity of disease. To be effective, scorecards need to be accurate. They also need to be simple to implement by clinicians, to allow for accurate manual scoring in a high pressure clinical environment. As such, the scorecard should contain as few variables as possible for clinicians to measure and score, whilst providing the highest accuracy possible in terms of predicting risk or mortality, to allow for differential allocation of limited treatment resources.

The creation of scorecards is usually accomplished manually, using various computational tools. A dataset of patient variables (features) coupled with disease severity or mortality outcomes is analysed to determine feature importance within the dataset. A scoring mechanism is then applied and the scorecard assessed for accuracy by calculating an ROC curve with each scoring point evaluated for false positive and false negative rates. The area under this curve (AUC) is the standard accuracy reporting measure. This study develops an automated tool or framework for the derivation of scorecards from clinical data. It demonstrates the first use of multi-objective evolutionary algorithms to develop a Pareto-front of potential scorecards for clinical experts to select from. The methodology could be applied to any appropriate disease with sufficient clinical data, however, the system is demonstrated here on real-world C.Difficile data as a case study.

Three automated methods are presented which improve on previous manually derived scorecards. These employ the NSGA-II Multi-objective Optimization Genetic Algorithm (GA) to optimize the Pareto-front of two clinically-relevant scorecard objectives, size and accuracy. The first method is a hybrid algorithm which uses the GA for feature selection and a decision tree for scorecard generation. In the second, the GA generates the full scorecard. The third is an extended full scoring system in which the GA also generates the scores allocated to each scorecard feature. In each case, combinations of features and thresholds for each scorecard point are selected by the relevent algorithm and the evolutionary process is used to discover near-optimal Pareto-fronts of scorecards for further exploration by expert decision makers. Although AUC results from the full scoring system show the greatest improvement over those of the other methods, through the use of differential scores for each feature, the additional level of complexity may not be suitable for all clinical situations. Nonetheless, the ability of these methods to offer a selection of scorecards to choose from, allows the clinical expert to make an informed choice based on clinical priorities.


Multi-objective LQR with optimum weight selection to design FOPID controllers for delayed fractional order processes

Saptarshi Das

An optimal trade-off design for fractional order (FO)-PID controller is proposed with a Linear Quadratic Regulator (LQR) based technique using two conflicting time domain objectives. A class of delayed FO systems with single non-integer order element, exhibiting both sluggish and oscillatory open loop responses, have been controlled here. The FO time delay processes are handled within a multi-objective optimization (MOO) formalism of LQR based FOPID design. A comparison is made between two contemporary approaches of stabilizing time- delay systems within LQR. The MOO control design methodology yields the Pareto optimal trade-off solutions between the tracking performance and total Variation (TV) of the control signal. Tuning rules are formed for the optimal LQR-FOPID controller parameters, using median of the non-dominated Pareto solutions to handle delayed FO processes.


Multi-objective optimisation with ensembles

Ivascu, Carina

Ensembles are collections of predictors which are often more accurate than single predictors. Two key features in the performance of an ensemble are accuracy and diversity. Since these two objectives are usually in conflict with each other, a multi-objective optimisation approach is used.