
基于虚拟多任务二元粒子群算法和分形维数的雾霾天气预测方法
Virtual Multitasking Binary Particle Swarm Optimization and Fractal Dimension for Haze Forecast
雾霾天气已对人类日常生活产生严重影响, 有效预测雾霾天气, 帮助城市 居民规划出行安排具有十分重要的现实意义. 因雾霾天气影响因素众多, 冗余因素的 存在一方面浪费计算机存储空间, 另一方面干扰预测结果准确性. 文章首先充分挖掘二 元粒子群算法(binary particle swarm optimization, BPSO) 的``隐并行性'', 构造虚拟多任务环境, 主任务和辅助任务中粒子分别执行不同的位置更新策略, 且相互传递有效信息, 从而保持种群动态多样性, 提出虚拟多任务二元粒子群算法(virtual multitasking binary particle swarm optimization, VMBPSO), 然后结合分形维数(fractal dimension, FD) 剔除雾霾天气中的噪声属性, 得出雾霾天气关键影响因素, 最后采用SVM 算法利用前一天雾霾天气关键影响因素预测后一天是否有雾霾. 仿真实验通过对即将举办亚运会的杭州和湖州两大城市进行分析预测, 结果表明文章算法具有较高的预测准确率, 稳定性和可靠性较高.
Haze has already brought great impact on human daily life. Effectively forecasting the haze weather can help urban residents plan their travel arrangements. Haze weather involves multiple factors. Redundant factors can not only waste computer storage space, but also interference prediction accuracy. In this paper, we first fully explore the implicit parallel of binary particle swarm optimization (BPSO), and construct virtual multitasking environment, the main task and auxiliary task perform different position update strategy, also useful information transfer between them to keep the dynamic diversity of the main subpopulation. On this basis, a novel virtual multitasking binary particle swarm optimization (VMBPSO) is proposed. Then VMBPSO combined with fractal dimension (FD) is used to eliminate noise haze properties and obtain key haze influence elements. After that SVM algorithm is involved to use the previous day's haze weather key factors to predict whether there is haze the day after. Experimental results reveal that our proposed method has higher prediction accuracy, higher stability and reliability.
/
〈 |
|
〉 |