Slavko Krajcar

Load forecasting using a multivariate meta-learning system

Although over a thousand scientific papers address the topic of load forecasting every year, only a few are dedicated to finding a general framework for load forecasting that improves the performance, without depending on the unique characteristics of a certain task such as geographical location. Meta-learning, a powerful approach for algorithm selection has so far been demonstrated only on univariate time-series forecasting. Multivariate time-series forecasting is known to have better performance in load forecasting. In this paper we propose a meta-learning system for multivariate time-series forecasting as a general framework for load forecasting model selection. We show that a meta-learning system built on 65 load forecasting tasks returns lower forecasting error than 10 well-known forecasting algorithms on 4 load forecasting tasks for a recurrent real-life simulation. We introduce new metafeatures of fickleness, traversity, granularity and highest ACF. The meta-learning framework is parallelized, component-based and easily extendable.

Genetic Algorithm Approach for Multiple Depot Capacitated Vehicle Routing Problem Solving With Heuristic Improvements

Abstract The paper proposes a Genetic Algorithm (GA) in conjunction with handy heuristic techniques to solve the non-fixed destination Multiple Depot Capacitated Vehicle Routing Problem (MDCVRP). New heuristic techniques are added in order to prevent converging to local optima and to speed up the convergence of the algorithm through a reduction of the search space domain. The proposed GA approach has been tested on several instances of practical longterm link distribution network design problems which can be easily correlated with the non-fixed destination MDCVRP. Test results reveal that the features of easy implementation, fast convergence, and a near optimal solution in solving the MDCVRP can be achieved with the proposed GA approach.

Darwin meets computers: new approach to multiple depot capacitated vehicle routing problem

We present a study of using genetic algorithms (GAs) to solve non-fixed destination multiple-depot capacitated vehicle routing problem. The genetic algorithm was developed on the basis of experiences in solving the travelling salesman problem and the single depot capacitated vehicle routing problem. Heuristic improvements in population initialization and crossover operators are made to prevent converging to local optima and to reduce the search space domain. To deal effectively with the constraints of the problem, and to prune the search space of GA in advance, the difficult capacity and supply reliability constraints are embedded in the decimal strings that are coded to represent the vehicle routes between depots. Computational results carried out on several instances indicate that the total distance travelled can be reduced significantly when such method is used.

A heuristic modification of genetic algorithm used for solving the single depot capacited vehicle routing problem

The paper proposes a genetic algorithm (GA) in conjunction with handy heuristic techniques to solve the single depot capacited vehicle routing problem (CVRP). The genetic algorithm is developed on the basis of experiences in solving the travelling salesman problem (TSP). Few heuristic improvements are added in order to prevent converging to local optima and to reduce the search space domain. To deal effectively with constraints of the problem and prune the search space of the GA in advance, the constraints are embedded in decimal coding of the problem. The proposed GA approach has been tested on several CVRP with different number of consumer nodes and different control parameters. The results of 144 node problem used for sensitivity analysis, reveal that the features of easy implementation, fast convergence, and near optimal solution in solving the CVRP can be achieved by the proposed GA approach.

Genetic algorithm and GIS enhanced long term planning of large link structured distribution systems

In order to enhance the serviceability in the distribution system genetic algorithm and GIS based method is proposed in this article for planning the link distribution networks. All practical issues such as cost parameters (investments, line losses, maintenance), and technical constraints (voltage drop, thermal limit, reliability) as well as physical routing constraints (obstacles, high cost passages, existing line sections) are taken into consideration. Fuzzy set concept and scenario representation (tree of futures) to model uncertainties, as well as decision making guided by a paradigm of multi-criteria risk analysis are discussed. The merits of the approach are discussed by analyzing its application to a study case based on a real case in a Croatian utility.

The genetic algorithm method for multiple depot capacitated vehicle routing problem solving

Many organizations face the problem of delivering goods from a certain number of warehouses to a number of retail sites using a fleet of vehicles. The multiple depot capacitated vehicle routing problem is mathematical model that closely approximates the problem faced by many of these organizations. Because the problem is NP-hard, requiring excessive time to be solved exactly, we develop a heuristic based on a genetic algorithm that finds high quality solutions in a reasonable amount of computer time. Basic GA procedures adapted to a given problem are presented and six versions of crossover operators are compared. The test results reveal that the method is able to produce results of a kind that are not easily obtained, namely in terms of the amount of information about the solutions and the solution space.

An efficient implementation of genetic algorithms for constrained vehicle routing problem

We propose a genetic algorithm based heuristic for solving the problem of open loop distribution network planning. The goal of power distribution system planning is to satisfy the growth and changing system load demand during the planning period and within operational constraints, with minimal costs. Although the algorithm was developed for specific real world problems, the method is quite general and can be encountered in many planning contexts that can be correlated with the well known Capacitated Vehicle Routing problem (CVRP). For the CVRP problem, the influences of the respective control parameters were examined. Also the issues regarding the usage of different selection parameters are examined, in order to observe their impact on the optimization procedure. The results of experiments testing the solution procedures are reported.

Dynamic planning of medium voltage open-loop distribution networks under uncertainty

An efficient method to address the multistage planning of open loop structured mv distribution networks under uncertainty, taking into account distributed generation connected to distribution system, has been proposed. The fuzzy model can cope with important features implicit in planning studies such as time-phased representation, consideration of conflicting objectives and uncertainty in loads, distributed generation and economic data. Using two evolutionary algorithms simultaneous optimization of costs and the reliability is achieved. Thus, in addition to optimal radial layout along several stages in time, the algorithm can determine the optimal locations of reserve feeders that achieve the best network reliability with the lowest expansion and operational costs. The model and evolutionary algorithms have been applied intensively to real life power distribution systems showing its potential applicability to significantly larger systems than those frequently found in literature about dynamic distribution networks planning. Results have illustrated the significant influence of the uncertainties in the optimal distribution network planning mainly in terms of topology and supply capacity of the resulting optimal distribution system

Profit maximization of a hydro producer in a day-ahead energy market and ancillary service markets

Water is scarce resource with uncertain availability. Hence finding an optimal production schedule for hydro producer is usually a complex task and it is necessary to carefully balance the timing of water use. This paper addresses the self-scheduling problem for a price-taker hydro producer. The goal is maximizing the profit of a hydro producer through participating in the day-ahead energy and ancillary service markets. The spinning reserve and regulation market are considered as ancillary service markets. The self-scheduling problem of hydro producer is formulated and solved as a mixed integer linear programming problem. Model is deterministic and three-dimensional relationship between the power produced, the water discharged, and the head of associated reservoir is accounted. Real hydropower system Vinodol is considered.

New genetic algorithm for link distribution system design

The ability to supply consumers of an urban area without any longer interruption during a feeder segment or substation transformer outage is assured by a link network configuration. The goal of method proposed in this paper is to design such link distribution network that satisfies the growing and changing load demand during the long term planning period with minimal service costs subject to voltage drop, cable and transformer capacity limits, limited number of feeders outgoing at HV/MV substations and limited number of load points per link. The method is based on biologically inspired genetic algorithm (GA). Basic GA procedures adapted to the given problem that enable simultaneous routing of all links between HV/MV substations are presented. Six versions of crossover operators are compared. The results reveal that that the method is able to produce results of a kind not easily obtained before namely in terms of an amount of information about the solutions and the solution space.