算力网络下的算力边缘服务器部署算法

算力边缘服务器部署问题是构建算力网络的基础性问题。在实践过程中，算力边缘服务器靠近算力资源并为其加入算力网络提供接入服务。然而，算力资源的整体结构往往由现实需求所决定，并时刻随需求的变化而变化。在算力边缘服务器资源有限的情况下，如何合理部署算力边缘服务器，使得其能够保障算力网络有效地建设已成为当前各界所关注的热点。首先，对算力边缘服务器部署问题进行分析，并将其转换为带约束的多目标优化问题。针对该问题，提出一种改进型遗传算法予以解决。该算法优点在于：寻找无重复可行解作为初始种群，为选择操作提供了更多挑选的余地；选择时，采用个体均衡选择策略，保证了迭代群体的多样化与分散化；交叉和变异时，分别采用不同种类的随机两点交叉与轮流随机单点变异的策略，从而保障了新生种群的多元性与多样性。实验从算力资源总量偏差率、负载平衡误差率、收敛率、期望最优解误差率四个方面验证，该算法适合应用于算力边缘服务器的部署。

Computing first edge server deployment algorithm for computing first network

The problem of computing first edge server deployment is a fundamental problem in computing first network. In the actual scenario, the computing first edge server is close to computing power resources and provides access services for them to join the computing first network. However, the structure of computing resources is often determined by the actual demand, and changes with the change of demand. Under the constraint of computing first edge server resources, how to reasonably deploy computing first edge servers to ensure the effective construction of computing networks has become a hot topic of concern for all sectors. Firstly, this paper analyzed the deployment problem of computing first edge servers and transformed it into a multiobjective optimization problem with constraints. It proposed adopted an improved genetic algorithm to address this issue. The advantages of this algorithm were as follows. it found non repetitive feasible solutions as the initial population provides more room for selection operations. When selecting, it adopted an individual balanced selection strategy to ensure the diversity and decentralization of the iterative population. When crossing and mutating, it adopted different types of random two point crossing and rotating random single point mutation strategies, thereby ensuring the diversity and diversity of the newborn population. The experiment verified by resources deviation rate, load error rate, convergence rate. And expectation solution error rate shows that the algorithm is very effective and reasonable.