Schedule
Wednesday 26 June
14:00-17:00 Registration and Tutorial in Hugh Robson Building (George Square) Theatre G.04
18:30-21:00 Welcome drinks at The Pear Tree
We have booked 30 seats at tables in the yard. If it rains, we'll find space inside
The pub serves respectable food
Thursday 27 June
09:00-12:30 Registration and Presentations in Adam House (Chambers Street) Basement Theatre
09:30 Julian Hall
HiGHS: Turning gradware into software and impact
10:10 Jacek Gondzio
Linear Programming: From the Simplex Method to Interior Point Method and Back
10:40 Filippo Zanetti
A new parallel interior point solver for HiGHS
11:00 Coffee/tea break: none provided
11:30 Ryan O'Neil
Symphonic HiGHS: Operationalizing next moves with DecisionOps
12:30-14:00 Lunch (DIY)
14:00-17:00 Presentations in Adam House (Chambers Street) Basement Theatre
14:00 Luke Marshall
Resource Management for Microsoft Azure Database-as-a-Service
14:25 Franz Wesselmann
Reaching (new) HiGHS with MathWorks and Optimization Toolbox
14:50 Filipe Brandão
Enhanced Solver Interfaces: Automatic Reformulations in AMPL’s New MP Solver Interface Library
15:15 Coffee/tea: none provided
15:45 Edwin Reynolds and Fotios Katsigiannis
Optimising Retail Operations using Mixed Integer Programming Solvers
16:10 Sam Barrett
Using HiGHS for max SAT problems
16:35 Antony Phillips
Optimising global pharmaceutical supply chains at 7bridges with HiGHS
18:30-21:00 Workshop dinner at Scotts Kitchen: Aperitif (18:30); Dinner (19:15).
Numbers at Scotts are limited to 36, so dinner for students will be at a venue to be arranged by local PhD students attending the workshop.
Due to late registrations and the number of true guest bookings being uncertain, we may not be able to offer places at the dinner to anyone registering after 14 June.
An after-dinner venue for all will be arranged.
Friday 28 June
09:30-12:00 Presentations in Adam House (Chambers Street) Basement Theatre
09:30 Fabian Neumann
Using HiGHS for open energy system planning and operation with PyPSA
09:55 Maximilian Parzen
Need for speed! The Role of Open-Source Solvers for Energy System Planning
10:20 Anders N. Andersen
The energy system analysis challenge for HiGHS
10:45 Coffee/tea: none provided
11:30 Josh Fogg
Robust Portfolio Optimization for Genetic Selection
11:55 Harley Mackenzie
Energising Renewables with Smart Optimisation with HiGHS
12:20 Lukas Biton
Optimal schedules for BESS trading in the long and short run
12:45-13:45 Lunch (DIY)
13:45-15:50 Presentations in Adam House (Chambers Street) Basement Theatre
13:45 Ailbhe Mitchell
Fish friendly(er) pumping: A MILP to optimize water allocation to the world’s largest pumps
14:10 William Pettersson
HiGHS benchmark results on a real-world problem
14:35 Andreas Lundell
Solving MINLP problems with HiGHS and SHOT
15:00 Wim Vanroose
15:25 Julian Hall
Finding irreducible infeasibility systems in HiGHS: what do users want?
16:00 The Pear Tree
Abstracts
Anders N. Andersen
The energy system analysis challenge for HiGHS
energyPRO is one of the software packages offered by EMD International, for planning investments in large energy plants, as wind farms, PV plants, P2X and district heating.
The basic task for energyPRO is to make digital twins of energy plants and optimizing the operation in the chosen planning period (typically 1 year) of the production units at the plants, taking into account e.g. the hourly electricity prices, the sizes of the production units and the energy storages (e.g. hydrogen storages and thermal storages), furthermore restrictions in the transmission capacities between the locations of the different components.
These digital twins are then used for analyzing the investment in e.g. a new heat pump, electrolyzer, hydrogen storage.
In this presentation are shown three examples for the challenge for energyPRO making digital twins of energy plants. Furthermore, in a presented film the manager of Hvide Sande district energy plant in Denmark tells how complex the daily market optimization of the plants is. The MILP challenge for HiGHS in these examples may quickly exceed 200.000. independent variables, half of these being integers.
Based on these examples optimized in energyPRO, show that HiGHS optimize operation in half the time as CBC, and Gurobi optimize operation in half the time as HiGHS. When it comes to the most complex energy system, CBC are not able to find a solution when optimizing 720 hours (one week), whereas Gurobi and HiGHS find a solution.
Lukas Biton
Optimal schedules for BESS trading in the long and short run
Zenobe operates large battery energy storage systems (BESS) across the UK. To help with this process, we use linear programs in two major ways:
1. Before the BESS is built, deciding whether the project is financially viable 15-20 years ahead, with a long term revenue optimiser
2. After the BESS is built, optimally trading energy across all markets with a high level of granularity, with a short term revenue optimiser
We cast both problems as MILPs, written in python with the pyomo framework and using HiGHS as our solver. The two main reason we like to use HiGHS are its permissive license, allowing us to easily use multiprocessing in python to radically reduce our run time, and its run speed, allowing us to solve these optimisation problems much faster than through a grid search.
Filipe Brandão
Enhanced Solver Interfaces: Automatic Reformulations in AMPL’s New MP Solver Interface Library
The ideal of model-based optimization is to describe your problem the way you think about it, and then let the computer do the work of getting a solution. Recent enhancements aim to bring the AMPL modeling language and system closer to this ideal. Using a variety of modeling language extensions, common formulations are described more naturally, with the AMPL translator, the AMPL-solver interface, or the solver itself doing most of the needed transformations.
Extensions described in this presentation include quadratic expressions, logical operators and constraints, simple near-linear and nonlinear functions, and combinations of these together with linear terms. All are supported by a new C++ AMPL-solver interface library that can be adapted to handle the multiple detection and transformation strategies required by large-scale solvers.
New solver drivers built with this interface include HiGHS, Gurobi, CPLEX, Xpress, COPT, MOSEK, CBC, SCIP, and GCG. The new interfaces improve solver switchability by providing a more uniform support of modeling constructs (e.g., logical constraints), options, and features (e.g., multi-objective support). This is achieved by adjusting the way the problem is communicated to solvers according to what works best with each solver.
Josh Fogg
Robust Portfolio Optimization for Genetic Selection
Developed in the 1950s, Markowitz' Modern Portfolio Theory has been employed to explore problems in finance for decades, and remains a cornerstone of contemporary theory. It examines the bi-objective optimization problem,
min x'μ, max x'Σx subject to x ≥ 0, x'e = 1, l ≤ x ≤ u
helping investors balance risk and return. In this talk we demonstrate how the algorithm can be adapted to solve a similar problem that arises in genetics in the context of selective breeding. There we seek to maximize genetic merit through the selection of desirable traits, while also minimizing inbreeding and its associated risks.
We are interested in the robust selection problem, which requires the solution of a quadratic programming problem with a conic constraint. This can be achieved with a bespoke SQP algorithm that, when implemented using HiGHS, has the potential to significantly out-perform Gurobi.
Jacek Gondzio
Linear Programming: From the Simplex Method to Interior Point Method and Back
A brief overview of two key LP solution techniques, that is, the simplex method and the interior point method will be given. The relations between the two approaches will be discussed and areas where techniques developed for one of them area source of inspiration for the other will be presented.
Julian Hall
HiGHS: Turning gradware into software and impact
HiGHS is open-source optimization software for linear programming (LP), mixed-integer programming (MIP) and quadratic programming (QP). Most of the solvers were originally written as “gradware” by PhD students, and this talk will give an account of how HiGHS was created from them. Independent benchmark results will be given to justify the claim that HiGHS is the world’s best open-source linear optimization software. Non-academic impact is an increasingly important measure when assessing UK academic departments, so the impact of HiGHS will be discussed. The themes of this talk will allow members of the HiGHS team and some workshop participants to be introduced personally.
Julian Hall
Finding irreducible infeasibility systems in HiGHS: what do users want?
For solver developers, it is important to identify infeasible problems accurately, particularly in the context of MIP solving, and the Farkas proof follows naturally from the detection of unboundedness in the dual simplex algorithm. However, when it occurs in a model, identifying an irreducible infeasibility system (IIS) is an important aid to a modeller wishing to identify the source of the "error". IIS calculation in HiGHS is currently being implemented, and this talk will discuss the process, with the aim of informing the variants that will be made available.
Andreas Lundell
Solving MINLP problems with HiGHS and SHOT
The Supporting Hyperplane Optimization Toolkit (SHOT) is a state-of-the-art open source solver for convex mixed-integer nonlinear programming (MINLP). It combines a dual strategy based on polyhedral outer approximation with a primal strategy based mainly on heuristics, e.g., solving integer-fixed nonlinear programming (NLP) problems. Convergence to the global optimal solution is guaranteed (theoretically) for convex MINLP problems (including convex mixed-integer quadratic programming and mixed-integer quadratically constrained quadratic programming, i.e., MIQP and MIQCQP, problems). For general nonconvex problems, SHOT primarily acts as a local solver. However, certain special cases can also be solved to global optimality by leveraging reformulations and other strategic approaches.
In the dual strategy in SHOT, iteratively improving approximations of the nonlinear feasible set is generated and solved in the form of MILP problems. Recently support for HiGHS as a subsolver was added to SHOT. In this presentation, we demonstrate the integration of SHOT with HiGHS, showcasing how SHOT can now serve as an extension of HiGHS to support solving MIQP, MIQCQP, and MINLP problems.
Harley Mackenzie
Energising Renewables with Smart Optimisation with HiGHS
An introduction to OptiGen, a next-generation optimiser recently developed for renewable energy generation and battery management within the Australian National Electricity Market (NEM). This 20-minute talk will explore the innovative capabilities of OptiGen, designed to optimise utility-scale batteries and DC coupled wind and solar plants, tailored for both scheduled and non-scheduled generation. Discover how OptiGen not only enhances revenue and reduces operational risk, but leverages the power of the HiGHS linear optimiser and Jumps to create a scalable and cost-effective solution for the NEM and potentially other international energy applications.
Luke Marshall
Resource Management for Microsoft Azure Database-as-a-Service
In Database-as-a-Service (DBaaS) clusters, resource management is a complex optimization problem that assigns tenants to nodes, subject to various constraints and objectives. Tenants share resources within a node; however, their resource demands can change over time and exhibit high variance. As tenants may accumulate large state, moving them to a different node becomes disruptive, making intelligent placement decisions crucial to avoid service disruption. Placement decisions need to account for dynamic changes in tenant resource demands, different causes of service disruption, and various placement constraints, giving rise to a complex search space. We show how Integer Programming solvers perform on this problem, with the objective of minimizing service disruption as an optimization problem amenable to fast solutions. We implemented our approach in the Service Fabric cluster manager codebase. Experiments show significant reductions in constraint violations and tenant moves, compared to the previous state-of-the-art, including the unmodified Service Fabric cluster manager, as well as recent research on DBaaS tenant placement.
Ailbhe Mitchell
Fish friendly(er) pumping: A MILP to optimize water allocation to the world’s largest pumps
The pumping station of the IJmuiden lock complex is of great importance for the water management of western Netherlands. It is used to regulate the level on the North Sea Canal, for example to ensure that the water can be drained adequately, but also that the water level in the North Sea Canal and the Amsterdam-Rhine Canal remains sufficiently high so that the ships (approx. 45,000 per year) can sail smoothly and safely (source: Sluizencomplex IJmuiden (rijkswaterstaat.nl)). A (Delft-FEWS) system is in place for the operational water management of the whole system.
Approximately 4.6 billion m3 of water is discharged annually, divided over the drain with 7 large drain pipes with a maximum capacity of 700 m3/s and the pumping station with a total of six pumps with a capacity of 260 m3/s (source: Sluizencomplex IJmuiden (rijkswaterstaat.nl)). These are super pumps, as they enjoy a listing in the Guinness Book of Records for its size (source: Pentair - News | Pentair Fairbanks Nijhuis). But unfortunately it has been known since 2008 that the current pumps are very harmful for fish and especially for the silver eels. The estimated direct mortality in silver eels of 46 percent, rising to 96 percent due to injuries. About 10,000 silver eels die directly each year, with up to double due to injuries. (source: Nieuwe dodelijke superpomp: visveiligheid vraagt om wetgeving (ravon.nl))
Deltares develop an MILP optimization routine for determining how the water can best be distributed among the pumps where the HiGHS solver is employed via the RTC-Tools software package. This includes an option to use the pumps in a way to reduce fish mortality or minimize energy consumption of the pumps.
Fabian Neumann
Using HiGHS for open energy system planning and operation with PyPSA
In this talk, we will demonstrate the use of HiGHS with PyPSA for modeling renewable energy systems with open-source software. PyPSA is a Python library for the planning and operational dispatch of energy systems and is widely used to find cost-effective pathways to reach climate neutrality by mid-century. HiGHS has recently become the default solver in PyPSA, reducing the reliance on commercial solvers. We will showcase several example applications where HiGHS has been beneficial, such as public online interfaces and tutorials for new users.
Ryan O'Neil
Symphonic HiGHS: Operationalizing next moves with DecisionOps
An individual may develop a decision model. But operationalizing it for real-world impact involves collaboration across teams and symphony of tooling. Many of us began our operations research journey with open source projects such as HiGHS. How can we accelerate their impact?
From model build to testing and evaluation, deployment, management, and iteration, DecisionOps delivers a standardized workflow for algorithm developers to realize the value of optimization while seamlessly obtaining stakeholder buy-in throughout the process.
In this session, we’ll explore how to use a price optimization model built using HiGHS. We’ll look at how to stand up a custom API endpoint backed by scalable compute, use scenario testing to help with model iteration, run acceptance tests to set defined expectations, hook into Git-based CI/CD workflows, and discuss the role of open source solvers in the growth of optimization as a discipline.
Maximilian Parzen
Need for speed! The Role of Open-Source Solvers for Energy System Planning
Energy system planning impacts operation and investment decisions amounting to several billion euros annually. Open-source (OS) solvers play a crucial role in this planning ecosystem, often serving as the key speed bottlenecks for generating practical and useful insights. This talk will introduce the concept of open modelling, showcase real-life use cases of energy planning studies, and explain why optimizers are typically the primary speed bottleneck. It will highlight the importance of high-quality OS solvers for a successful energy transition. Finally, a new benchmark project will be introduced, offering benefits to both the energy modelling and solver communities.
William Pettersson
HiGHS benchmark results on a real-world problem
Kidney exchange programmes increase the rate of living-donor kidney transplants, and they do so in part by solving integer linear programmes. In this talk, I will explain two of the more popular models used in kidney exchange programmes, and give some benchmark results not only comparing these two models, but also comparing HiGHS against other popular ILP solvers on each of these models. This gives a real-world demonstration of the impact of operations research techniques in general, and an insight into why various ILP solvers are used.
Antony Phillips
Optimising global pharmaceutical supply chains at 7bridges with HiGHS
7bridges builds an AI-powered logistics platform, internally solving numerous optimisation problems across the spectrum of supply chain management. This talk will focus on one specific supply chain optimisation problem designed for the global pharmaceutical sector in 2024. We will cover how strategic and tactical decisions can be made while modelling efficiency and resilience at the operational level. This system and optimisation model have been developed using real-world data and are powered by HiGHS.
Cristina Radu
Creative Solutions in Supply Chain Optimization
I propose a discussion about how to create more space for creativity into a highly technical area such as supply chain optimization software. I give a few examples of certain approaches we can use to find simple solutions despite the complexity. Finally, I would like to present the concept of a drag-and-drop equations editor.
Edwin Reynolds and Fotios Katsigiannis
Optimising Retail Operations using Mixed Integer Programming Solvers
Data Science underpins the technology that is crucial for Tesco to operate efficiently and deliver great value to our customers. In this talk, we will describe our approach to solving optimisation problems at large scale in a retail environment, the kinds of problems we have found MIP solvers to be suitable for, and how MIP solvers such as HiGHS can contribute towards successful projects.
Wim Vanroose
Residual simplex method
We propose a novel method for solving large-scale linear programming problems based on ideas from Krylov subspace methods. Each iteration we solve a projected problem where the linear constraints are projected on the basis of the residuals from the previous iterations. This results in a series for small projected LP problems that increase in size. But each problem can be warm-started with the basic set and of solution of the previous problem. We discuss the convergence, the implementation with HiGHS and the performance on some large-scale network problems.
Franz Wesselmann
Reaching (new) HiGHS with MathWorks and Optimization Toolbox
Starting with MATLAB release R2024a Optimization Toolbox has been using HiGHS as the default solver for linear and mixed-integer programming problems (LPs and MIPs). In this talk, we report on our experience with integrating HiGHS and opportunities for expanding its usage. We also highlight areas where our team and MathWorks could assist in developing HiGHS further. Finally, we discuss early customer feedback and our wish list for HiGHS.
Filippo Zanetti
A new parallel interior point solver for HiGHS
We present the latest developments concerning the new HiGHS interior point solver. We show the advantages and disadvantages of solving the Newton system using the normal equations or the augmented system approaches, and we highlight the use of regularization to avoid pivoting. We show some preliminary serial results and we present future plans for parallelizing the code.