Akihiro Kishimoto

A Lagrangian decomposition approach for the pump scheduling problem in water networks

Dynamic pricing has become a common form of electricity tariff, where the price of electricity varies in real time based on the realized electricity supply and demand. Hence, optimizing industrial operations to benefit from periods with low electricity prices is vital to maximizing the benefits of dynamic pricing. In the case of water networks, energy consumed by pumping is a substantial cost for water utilities, and optimizing pump schedules to accommodate for the changing price of energy while ensuring a continuous supply of water is essential. In this paper, a Mixed-Integer Non-linear Programming (MINLP) formulation of the optimal pump scheduling problem is presented. Due to the non-linearities, the typical size of water networks, and the discretization of the planning horizon, the problem is not solvable within reasonable time using standard optimization software. We present a Lagrangian decomposition approach that exploits the structure of the problem leading to smaller problems that are solved independently. The Lagrangian decomposition is coupled with a simulation-based, improved limited discrepancy search algorithm that is capable of finding high quality feasible solutions. The proposed approach finds solutions with guaranteed upper and lower bounds. These solutions are compared to those found by a mixed-integer linear programming approach, which uses a piecewise-linearization of the non-linear constraints to find a global optimal solution of the relaxation. Numerical testing is conducted on two real water networks and the results illustrate the significant costs savings due to optimizing pump schedules.

Distribution Loss Minimization With Guaranteed Error Bound

Determining loss minimum configuration in a distribution network is a hard discrete optimization problem involving many variables. Since more and more dispersed generators are installed on the demand side of power systems and they are reconfigured frequently, developing automatic approaches is indispensable for effectively managing a large-scale distribution network. Existing fast methods employ local updates that gradually improve the loss to solve such an optimization problem. However, they eventually get stuck at local minima, resulting in arbitrarily poor results. In contrast, this paper presents a novel optimization method that provides an error bound on the solution quality. Thus, the obtained solution quality can be evaluated in comparison to the global optimal solution. Instead of using local updates, we construct a highly compressed search space using a binary decision diagram and reduce the optimization problem to a shortest path-finding problem. Our method was shown to be not only accurate but also remarkably efficient; optimization of a large-scale model network with 468 switches was solved in three hours with 1.56% relative error bound.

Game-Tree Search Using Proof Numbers: The First Twenty Years

Solving games is a challenging and attractive task in the domain of Artificial Intelligence. Despite enormous progress, solving increasingly difficult games or game positions continues to pose hard technical challenges. Over the last twenty years, algorithms based on the concept of proof and disproof numbers have become dominating techniques for game solving. Prominent examples include solving the game of checkers to be a draw, and developing checkmate solvers for shogi, which can find mates that take over a thousand moves. This article provides an overview of the research on Proof-Number Search and its many variants and enhancements.

Active Learning for Multi-relational Data Construction

Knowledge on the Web relies heavily on multi-relational representations, such as RDF and Schema.org. Automatically extracting knowledge from documents and linking existing databases are common approaches to construct multi-relational data. Complementary to such approaches, there is still a strong demand for manually encoding human expert knowledge. For example, human annotation is necessary for constructing a common-sense knowledge base, which stores facts implicitly shared in a community, because such knowledge rarely appears in documents. As human annotation is both tedious and costly, an important research challenge is how to best use limited human resources, whiles maximizing the quality of the resulting dataset. In this paper, we formalize the problem of dataset construction as active learning problems and present the Active Multi-relational Data Construction (AMDC) method. AMDC repeatedly interleaves multi-relational learning and expert input acquisition, allowing us to acquire helpful labels for data construction. Experiments on real datasets demonstrate that our solution increases the number of positive triples by a factor of 2.28 to 17.0, and that the predictive performance of the multi-relational model in AMDC achieves the highest or comparable to the best performance throughout the data construction process.

Multi-criteria journey aware housing recommender system

Recommender systems can be employed to assist users in complex decision making processes. This paper presents a multi-criteria housing recommender system which takes into account not just features of a home, such as rent, but also the transportation links to user specified locations. First, we describe an efficient multi-hop journey time calculator. Second, we introduce a mechanism to find the optimal solutions for multi-criteria evaluation, where a balanced trade-off between the target goals is found. Finally, we present a user study to demonstrate the potential of such a system.

Succinct interval-splitting tree for scalable similarity search of compound-protein pairs with property constraints

Analyzing functional interactions between small compounds and proteins is indispensable in genomic drug discovery. Since rich information on various compound-protein inter- actions is available in recent molecular databases, strong demands for making best use of such databases require to in- vent powerful methods to help us find new functional compound-protein pairs on a large scale. We present the succinct interval-splitting tree algorithm (SITA) that efficiently per- forms similarity search in databases for compound-protein pairs with respect to both binary fingerprints and real-valued properties. SITA achieves both time and space efficiency by developing the data structure called interval-splitting trees, which enables to efficiently prune the useless portions of search space, and by incorporating the ideas behind wavelet tree, a succinct data structure to compactly represent trees. We experimentally test SITA on the ability to retrieve similar compound-protein pairs/substrate-product pairs for a query from large databases with over 200 million compound- protein pairs/substrate-product pairs and show that SITA performs better than other possible approaches.

Chemical Reactant Recommendation Using a Network of Organic Chemistry

This paper focuses on the overall task of recommending to the chemist candidate molecules (reactants) necessary to synthesize a given target molecule (product), which is a novel application as well as an important step for the chemist to find a synthesis route to generate the product. We formulate this task as a link-prediction problem over a so-called Network of Organic Chemistry (NOC) that we have constructed from 8 million chemical reactions described in the US patent literature between 1976 and 2013. We leverage state-of-the-art factorization algorithms for recommender systems to solve this task. Our empirical evaluation demonstrates that Factorization Machines, trained with chemistry-specific knowledge, outperforms current methods based on similarity of chemical structures.

Pump Scheduling for Uncertain Electricity Prices

Water utilities have optimized pump schedules to take advantage of day/night electricity pricing plans for several decades. As intermittent renewable energy sources such as solar and wind power provide an increasingly large share of the available electricity, energy providers are moving to dynamic pricing schemes where the electricity price is forecast 24 hours in advance on 30-minute time steps. The customer only knows the actual price several days after the electricity is used. Water utilities are uniquely positioned to take advantage of these dynamic prices by using their existing infrastructure for pumping and storage to respond to changing costs for power. This work develops an operational technique for generating pump schedules and quantifying the uncertainty in the cost of these schedules. With information about the pumping schedules and the distribution of possible costs, a system operator can pump according to her desired level of risk. To develop this information, a representative sample of electricity price forecasts covering nearly the full range of possible price curves must be created. Forecasts from the energy supplier and historical data on actual prices are used to condition stochastic sampling of daily energy price trajectories using covariance decomposition methods. From this ensemble of realizations, electricity price profiles are classified into a handful of scenario classes. The optimal pumping schedule for each price class is then computed. Once the pumping schedule is known, the price of that schedule is evaluated against all other price classes to determine the robustness of the schedule. The method is applied on a simple real-world network in Ireland. In this application, electricity prices vary every half hour and range from 5 to 262 €/mWh. Optimizing the pumping schedule proved to be the slowest step in the process so selection of proper price scenarios on which to generate the schedule was critical to obtaining results in an operational time-frame.

Parallel A* for State-Space Search

A* is a best-first search algorithm for finding optimal-cost paths in graphs. A* benefits significantly from parallelism because in many applications, A* is limited by memory usage, so distributed memory implementations of A* that use all of the aggregate memory on the cluster enable us to solve problems that can not be solved by serial, single-machine implementations. We survey approaches to parallel A*, focusing on decentralized approaches to A* which partition the state space among processors. We also survey approaches to parallel, limited-memory variants of A* such as parallel IDA*.