基于多目标遗传算法的GSM网络频率优化

摘要

Frequency optimization is an important way for effectively cancelling interference and ensuring quality of services in GSM networks. The current frequency optimization schemes usually take single-object optimizing algorithms that adopt minimizing of the linear combination of same frequency interference and adjacent frequency interference as the optimal targets. However, the linear combination of these two types of interference could not reflect their different impor-tance reasonably, and the position and azimuth relationships among sectors having same frequency or adjacent frequency should also be considered in frequency assignment. This paper proposes a multi-object genetic algorithm based frequency optimization algorithm, which takes the assignment matrix to model individuals in each generations, and utilizes the same frequency interference, adjacent frequency interference and overlapping degree as three optimal sub-objects respectively. To avoid trapping into local optimal solutions, the crossover operator adopts the geometrical distance between individuals to determine the pairing, and the mutation operator dynamically adjusts mutation probability by evolutionary generations. To maintain the uniform distribution and diversity of the populations, niche based shared fitness and the dynamic adaptive grid are used. Finally, analytic hierarchy process is introduced to select the preference solutions in the solution space. Applying of the above-mentioned scheme to the practical GSM networks demonstrate that MOGA can provide proper preference so-lutions quickly. It is also proved that MOGA is a feasible solution to frequency optimization in practical GSM networks, and has good application prospects.%频率优化是 GSM 网络优化的重要内容，可以有效地降低干扰，保障网络服务质量。传统的频率优化采用单目标算法，将同频干扰、邻频干扰作为干扰的衡量指标，用目标组合函数将同邻频干扰组合成单个目标。然而，简单的线性组合难以反映同邻频干扰不同的重要性，且实际网络中，还需要考虑具有同频或邻频的小区的地理位置关系、方向角关系等。因此本文将多目标遗传算法MOGA应用于GSM网络的频率优化，采用二维分配矩阵的基因编码方式，并将对打度纳入到频率优化的目标中，以极小化同频干扰、邻频干扰和对打度作为三个优化子目标。通过在交叉算子中以个体的几何距离决定交叉地配对，在变异算子中动态地调整变异概率等方式，增大差异后代产生得概率，防止算法陷入局部极大极小解。使用共享适应度值的小生境技术及自适应网格法维持种群在解空间的均匀分布及解集多样性。并采用层次分析法，从Pareto 最优解集中，选择偏好最优解。运用大规模现网数据实验结果证明，该算法能在较快时间内，得到符合偏好、满足多个优化目标的频率优化方案。上述工作说明，在实际网络频率优化工作中，基于多目标遗传算法的频率优化算法能够合理地反映多种频率优化目标，快速得到有效解，具有很好的应用价值。