日益严重的空气污染, 严重影响日常生产生活. 因此, 亟 需对空气质量进行预测. 为了实现高效、科学的预测, 需准确地选择出 空气质量数据集中的关键影响因素, 故提出了基于烟花进化人工鱼群算 法和多重分形的属性选择方法, 并应用于空气质量预测中. 首先, 采用 混沌初始化方式生成初始种群, 对人工鱼群算法进行离散化改进, 并引 入烟花进化机制, 提出烟花进化人工鱼群算法(FEAFSA), 提高算法的搜 索效率; 其次, 融合FEAFSA 和多重分形维数 (MFD), 对空气质量数据 集进行属性选择, 约简冗余、不相关属性, 保留空气质量关键属性; 最 后, 在8 个UCI 数据集上的实验结果表明, 相较于其他属性选择方法, 其能 有效剔除冗余因素, 性能更优, 同时表明其有效性、稳定性和显著性. 在进 行性能测试之后, 将其应用于北京、上海和广州地区的空气质量等级和指数 预测中, 取得了良好的预测效果.

The increasingly air pollution seriously affects daily production and life. Therefore, it is urgent to predict air quality. In order to achieve efficient and scientific prediction, the key influencing factors of air quality should be selected accurately. So an attribute selection method based on fireworks evolution artificial fish swarm algorithm and multi-fractal dimension is proposed, and it is applied to air quality prediction. First, fireworks evolution artificial fish swarm algorithm (FEAFSA) is proposed by designing an initial population using chaos method, improving the moving way of artificial fish swarm algorithm, and introducing fireworks evolution mechanism. Second, attribute selection of air quality dataset is achieved by combining FEAFSA and multi-fractal dimension (MFD), which reduces the redundant attributes, and retains the key attributes of air quality dataset. Finally, experimental results on 8 benchmark UCI datasets demonstrate that the proposed attribute selection method can attain better results than other approaches, it can effectively eliminate redundant attributes, and it has good effectiveness, stability and significance. After its performance verification, it is applied to the air quality levels and index prediction of Bejing, Shanghai and Guangzhou areas, and it achieves good predictive results.