A new multi-objective wind driven optimization algorithm is proposed to size a standalone photovoltaic system’s components to meet the load demand for a mobile network base station at a 1% loss of load probability or less with a minimum annual total life cost. To improve the sized model’s accuracy, a long short-term memory deep learning model is utilized to forecast the hourly performance of a photovoltaic module. The long-term memory model’s performance is compared with those obtained by a linear photovoltaic model and an artificial neural network model. The comparison is carried out based on the values of normalized root mean square error, normalized mean bias error, mean absolute percentage error, and the training and testing time. Accordingly, on the values obtained for these statistical errors, the long short-term memory model outperforms better than the linear model and the artificial neural network model based. In addition, a dynamic battery model is utilized to characterize the dynamic charging and discharging process. The findings show that the optimal number of the photovoltaic array and the capacity of the storage battery required to cover the load demand of a mobile network base station are 5.4 kWp and 2640 Ah/48 V, respectively. Besides, the annual total life cycle cost for the sized photovoltaic/battery configuration is 4028.33 AUD/year. The simulation time for the proposed method is 421.25 s. To generalize the sizing results for the mobile network base stations based on Sydney weather conditions, the photovoltaic array and storage battery ratios are calculated as 0.324 and 0.223, respectively. In addition, the cost of an energy unit generated by the optimized system is 0.254 AUD/kWh. Here, the results of the proposed method have been compared with those obtained by developed and recent benchmark published methods. The comparison outcomes show the effectiveness of the proposed method in terms of providing a high availability sized system at minimum cost within less simulation time than the other considered methods.

提出了一种新的多目标风力驱动优化算法，该算法可确定独立光伏系统组件的大小，以使移动网络基站的负载需求降低1％或更低的负载概率，并以最低的年度总寿命成本实现。为了提高大小模型的准确性，使用了长期短期记忆深度学习模型来预测光伏模块的每小时性能。将长期记忆模型的性能与通过线性光伏模型和人工神经网络模型获得的性能进行比较。根据归一化均方根误差，归一化平均偏差误差，平均绝对百分比误差以及训练和测试时间的值进行比较。因此，在针对这些统计误差获得的值上，长短期记忆模型的性能优于线性模型和基于人工神经网络的模型。另外，利用动态电池模型来表征动态充电和放电过程。研究结果表明，满足移动网络基站负载需求所需的最佳光伏阵列数量和蓄电池容量分别为5.4 kWp和2640 Ah / 48V。此外，大型光伏/电池配置的年度总生命周期成本为4028.33 AUD /年。该方法的仿真时间为421.25 s。为了概括基于悉尼天气状况的移动网络基站的规模计算结果，光伏阵列和蓄电池的比率分别计算为0.324和0.223。此外，经过优化的系统产生的能源单位成本为0.254澳元/千瓦时。在这里，已将所提出的方法的结果与通过已开发的和最新的基准发布方法所获得的结果进行了比较。比较结果表明，与其他考虑的方法相比，该方法在以最小的成本提供最小的成本下提供高可用性大小的系统方面是有效的。