Contributor identification - CIRED



Contributor identification |Contribution identification | |

|Name |Minea Skok |Session |5 |

|Company |Faculty of Electrical Engineering and Computing |Block |2 |

|Address |Unska 3, 10000 Zagreb, Croatia |Question n° |24 |

| | |Language used on the |English |

| | |floor | |

|Phone |+38516129693 |Accompanying visuals on |(Skok_HR_author_ALPHA5_BLOCK2_selected_pape|

| | |file ? |r.ppt) |

|e-mail |Minea.Skok@fer.hr | | |

Geographical Information System and Genetic Algorithm based planning tool for MV distribution networks

The aim of this presentation is to give some introductory remarks on use of genetic algorithms (GA), or to be more general evolutionary algorithms (EA), in distribution systems. Furthermore, as a didactic illustration of GA application in long-term large-scale urban distribution network planning, operational methodology named CADDiN used by several Croatian Distribution Utility Companies will shortly be proposed.

Genetic algorithms belong to the class of evolutionary algorithms which is an umbrella term used to describe all computer-based problem solving systems which model some known evolution mechanisms. The most popular evolutionary algorithms are:

• genetic algorithms,

• evolutionary programming,

• evolution strategies,

• genetic programming,

• and classifier systems.

Over the last years, interest in evolutionary algorithms has risen amongst researchers in all fields of power systems. This is because these approaches are very well suited to deal with all those kinds of problems that usually represent nightmares for researchers and developers:

• integer variables,

• non-convex functions,

• non-differentiable functions,

• badly-behaved functions,

• domains not connected,

• multiple local optima,

• multiple objectives,

• fuzzy data, etc.

Such complexity is what is required in order to build larger distribution system models with more adherences to reality. This leads to the NP complete problems and EA seem to be the only practical tool available to reach global optimization. Of course, if simplified (approximate, relaxed) models are used, then specialized knowledge-base or conventional operational research based tools will, by no means, outperform the evolutionary algorithms. Still, they remain the best method for reaching quickly high-quality solutions.

Over the last few years many papers have been published on use of evolutionary algorithms in power systems. For example surveys can be found in the following references:

1. V. Miranda, D. Srinivasan, L.M. Proenca, Evolutionary computation in power systems, Electrical Power & Energy Systems, Vol.20, No.2, 1998, pp. 89-98.

2. D. Srinivasan, F.S. Wen, C.S. Chang, A.C. Liew, A survey of applications of evolutionary computation to power systems, Proceedings of ISAP’96, Orlando, USA, 1996, pp. 35-43.

3. J.T. Alexander, An indexed bibliography of genetic algorithms in power engineering, Report 94-1-Power, Department of Information Technology and Production Economics, University of Vassa, Finland, February 1996.

4. M.A. Laughton, Genetic algorithms in power system planning and operation, IEE Coloquium on Artificial Intelligence in Power Systems, IEE Digest No. 075, London, UK, 1995, pp. 5/1 -5/3.

The core to this presentation are distribution system applications which are reproduced in Table1 based upon the results from aforementioned surveys.

Table 1 Survey on use of EA in distribution systems

|Area |Field |GA |ES |EP |GP |Hybrid |

|Expansion planning |distribution |X |X | | |GA+Fuzzy |

| |VAr planning, capacitor placement |X | |X | | |

|Distribution |Loss minimization, switching |X | | | | |

|operation | | | | | | |

| |Fault diagnosis |X | | | |GA+NN |

| |Service restoration |X | | | |GA+Exp.Sys, PGA |

| |Load management | |X | | | |

| |Load forecasting |X | |X | |GA+NN |

|Analysis |Harmonics |X | | | | |

It could be observed that the vast majority of the applications use genetic algorithms. However, the interest in the use of evolutionary strategies and programming is rising fast. The genetic programming and classifier systems are the newest techniques, but they show immense potential. Furthermore, the fruitful areas of further research are hybrid systems combining evolutionary computation tools with conventional as well as techniques such as Fuzzy Systems and Artificial Neural Networks.

CADDiN method is based on several software tools each aimed at different part of medium voltage (MV) distribution network planning (Figure 1):

• extensions to commercial Geographical Information System (GIS) and CAD software

o analyzing the existing distribution system

o preparing necessary data (collecting, converting, calculating)

o evaluating different expansion alternatives

o interpreting and analyzing planning results (strategy).

• optimization module

o distribution network planning in urban areas

▪ open loop

▪ link layout (connective and clasp layout)

o distribution network planning in rural areas

• spatial load forecasting module.

[pic]

Fig.1. CADDiN architecture

The most challenging (due to its complexity) is the evolutionary optimization module for link distribution network planning with multiple HV/MV substations and switching stations and therefore it will be briefly described in what follows.

Link distribution networks are the common practice in Croatian MV Distribution Networks. Namely, such very structured design is relatively easy to layout, engineer, build and operate, and what is becoming of a great importance feeder and transformer contingency back-up are espoused.

With respect to the observed network expansion alternative evolutionary algorithm has the following functions:

• optimal feeder routing,

• switching stations and transformer substations sitting and sizing,

• service areas of HV/MV substations,

• contingency switching (tie-lines).

The objectives are:

• minimal capital investment costs which include new substations and transformers costs, costs of new feeder sections and costs of adding new feeders to supply and switching substations),

• power and energy losses costs,

• energy not supplied costs,

• and maintenance costs.

Subject to the following constraints:

• voltage drop,

• loading limits,

• contingency margin rules,

• network layout,

• the total load with each link

• the total number of MV/LV substations per link.

The problem formulated in the foregoing is an optimization problem, but it is not a conventional convex programming type of problem. Because of the nature of the objective function — a complex, non-linear, non-analytical function — and because of the nature of the solution space — a graph and link layout — the optimization problem is not susceptible to mathematical programming algorithms. Our experience has led, upon successive iterations, to design specific evolutionary-based algorithms to search for the optimal solution.

Driven by difficulties in handling topology (connectivity and radiality) constraints, inherent to some previously proposed EA based methods for MV distribution network planning, in CADDiN both genotype and evolutionary operators are able to process meaningful topological information – network layout (link, open-loop, etc). Different link distribution networks are defined by the order of MV/LV substations (Fig.2), meaning that GA approaches the problem as a sequencing type. Before evaluating the fitness of some chromosome two-step decoding procedure converts these orders into real link distribution network (Fig.2). In the first step, for each MV/LV substation in turn the “closest” (in terms of costs determined in GIS) feeder ending is determined and then in the second step, when all MV/LV substation are connected to their corresponding feeders, the “closest” feeder endings are found with regard of predefined permissible supply/switching substations pairs.

[pic]

Fig.2. Chromosome coding and decoding

Comparing different sequencing type crossover and mutation operators FRX and CX crossover and OBM mutation operator were selected.

To get more distinct flavor and understand CADDiN’s potential if adopted in search for strategies based on robust decisions in the uncertainty and multiple criteria framework interested readers are referred to the paper.

-----------------------

Distribution system expansion planning – feeder routing, transformer substations and switching stations sitting and sizing, service areas, contingency switching (tie lines), etc.

Load forecasting

Analizing the existing distirubtion system –

distribution of load, transfer capability of existing cable system and overhead lines, capacity limitations and supply areas of substations, etc.

Evaluating different expansion alternatives.

Interpreting and analyzing the expansion planning results (strategy).

Preparing necessary data -

collecting, converting and calculating.

Geographical Information System Extensions

12

coding

The result of the second step of the decoding procedure

15

14

13

16

10

8

9

6

7

2

0

11

5

4

3

1

The result of the first step of the decoding procedure

15

14

13

16

10

12

8

9

6

7

2

0

11

5

4

3

1

supply substation

load point

decoding

chromosome 14 12 15 8 9 13

15

14

13

16

10

8

9

7

2

0

11

5

4

3

1

12

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download