Wind farm layout optimization using genetic algorithm with different hub height wind turbines

Outline

• Abstract
• Highlights
• Keywords
• 1. Introduction
• 2. Methodology
• 2.1. Wake Model
• 2.2. Genetic Algorithm
• 3. Pretest and Ga Parameters Validation
• 4. Case Studies with Results and Discussion
• 4.1. First Case Study: Constant Wind Speed and Direction
• 4.2. Second Case Study: Constant Wind Speed and Various Wind Directions
• 4.3. Third Case Study: Various Wind Speeds and Directions
• 4.3.1. Third Case Study with Simplified Cost Model
• 4.3.2. Third Case Study with Comprehensive Cost Model
• 4.4. Large Wind Farm with Commercial Wind Turbines
• 5. Conclusion
• Acknowledgements
• References

Conclusions

In this paper, the authors first investigate the effects of using different hub height wind turbines in a small onshore wind farm with nested GA. The GA parameters are first validated through pretest and comparison with previous research results. Three case studies are conducted by comparing wind farms using different hub height wind turbines and using same hub height wind turbines. The results demonstrate that the power output of wind farm with different hub height wind turbines will be better even when the total numbers of wind turbines are same. Different cost models are also taken into account in the analysis, and results indicate that different hub height wind turbines can also improve cost per unit power generated by the wind farm. At last, a large wind farm with commercial wind turbines is analyzed to further test the benefits of using different hub height wind turbines in realistic conditions, where the results demonstrate the conclusion obtained from first three case studies. The method presented in this paper is not only limited to two hub height wind turbines. Real coded GA can be employed with proposed nested GA by using the codes to represent different hub heights choices in GA2. For example, if there are five hub height wind turbines as candidates, in GA2, we can use ‘‘0–4’’ to represent the corresponding hub height with same possibility for GA to choose each hub height. The proposed nested GA can be also used to solve problems involving both wind turbine types and hub height selection. For example, if GA1 obtains the number of wind turbines placed in the wind farm as 20, the string length in GA2 should be 40, where the first 20 bits represent the turbine types and the rest 20 bits represent turbine hub heights. However, real coded method may have some drawbacks when the hub height choice increases to three or more in one position since the possibility of choosing a wind turbine to locate there will be decreased. For example, if there are five potential hub heights, the possibility that the position has no wind turbine will be 0.2 while the position can be placed a wind turbine will be 0.8. This may also increase the computational time. The same problem will not exist in the nested GA method since position selection is handled by GA1 with uniform possibility. In addition, real codes can represent coordinates of wind turbines’ positions and hub heights in one chromosomal string in parallel multi-objective GA as used in Section 4.4.

Meanwhile, the method can be also applied into other types of wind farms as long as the wake model stays the same. It is noted that the application of different hub height wind turbines in a wind farm actually takes advantages of decreasing wake effects, thus, this application will not be very helpful if the difference of two or more hub heights is too small. Because of nested GA may require more computational efforts, the number of individuals and generations may not be set too high, and the size of selected wind farm is small due to the same reason.

Parallel computing method in MATLAB will be used later to increase our computing capacity in the future research, so that more individuals and generations with large wind farm can be tested using proposed nested GA method. In the future research, we will compare the computational efforts needed of using nested GA and real coded GA. Meanwhile, although the analytical wake model used in this paper is commonly used in other similar research works, it is actually a very simplified model that may not be suitable for complex terrain conditions. And the surface roughness length in this paper is also an assumed value. Different wake model with real surface roughness length values generated by geographic information system should be tested in the future research to make the proposed method more applicable. Finally, we need to pay attention to how much the maintenance cost will increase due to increased partial load after introducing different hub height wind turbines.