模型和数据联合驱动的ARIMA-IDSSA-LSSVM建筑安全事故预测  

ARIMA-IDSSA-LSSVM construction safety accident prediction method driven by model and data

作　　者：曹红梅[1] 陈元 CAO Hongmei;CHEN Yuan(Taiyuan City Vocational and Technical College,Taiyuan 030027,China;Department of Civil Engineering,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]太原城市职业技术学院,山西太原030027 [2]浙江大学土木工程系,浙江杭州310058

出　　处：《自然灾害学报》2025年第2期129-139,共11页Journal of Natural Disasters

基　　金：2023年度山西省基础研究计划项目(202303021222286)。

摘　　要：针对传统单一模型在解决建筑安全事故预测问题存在精度低等问题,考虑模型和数据联合驱动方式,提出一种结合差分自回归移动平均(autoregressive integrated moving average,ARIMA)模型和改进的自适应樽海鞘优化最小二乘支持向量机(improved adaptive salp swarm algorithm optimized least squares support vector machine,IDSSA-LSSVM)的组合预测模型。首先利用ARIMA模型获得时序数据中线性部分,利用IDSSA-LSSVM模型分析ARIMA模型获得的残差,获得时序数据中非线性部分;然后通过线性部分和非线性部分相加获得最终组合预测值;最后通过2010—2020年房屋市政工程生产安全事故数据对所提算法进行验证。结果表明,所提预测模型在E_(rmse)上较其他算法分别下降73.73%、77.21%、46.09%、46.80%、78.19%,在E_(mae)上较其他算法分别下降74.20%、77.44%、48.15%、48.85%、77.50%,在E_(mape)上较其他算法分别下降84.95%、87.77%、75.97%、88.49%、80.27%。在不同规模的数据集下,文中算法在E_(rmse)指标下均最优。同时能够通过预测未来阶段事故,提供辅助决策。表明ARIMA-SSA-LSSVM组合模型能够充分挖掘建筑安全事故数据的隐藏信息,在准确性、泛化性和应用性3个角度均表现不错,优势明显。Aiming at the low accuracy of the traditional single model in solving the construction safety accident prediction problem,considering the model and data joint driving method,a combined prediction model combining autoregressive integrated moving average(ARIMA)model and improved adaptive salp swarm algorithm optimized least squares support vector machine(IDSSA-LSSVM)was proposed.Firstly,ARIMA model is used to obtain the linear part of the time series data.IDSSA-LSSVM model was used to analyze the residual of ARIMA model and obtain the nonlinear part of time series data.Secondly,the predicted value of the final combination is obtained by adding the linear part and the nonlinear part.Finally,the proposed algorithm is verified by the production safety accident data of housing municipal engineering from 2010 to 2020.The results show that,the proposed prediction model decreased by 73.73%,77.21%,46.09%,46.80%and 78.19%compared with other algorithms on E_(rmse),and decreased by 74.20%,77.44%,48.15%,48.85%and 77.50%compared with other algorithms on E_(mae),and decreased by 84.95%,87.77%,75.97%,88.49%and 80.27%compared with other algorithms on E_(mape).The results show that the combined model of ARMI-SSA-LSSVM can fully mine the hidden information of construction safety accident data,and the prediction accuracy is higher.Under different scale data sets,the algorithm in this paper is optimal under index E_(rmse).At the same time,it can provide auxiliary decision-making by predicting future stage accidents.The results show that the combined model of ARMI-SSA-LSSVM can fully excavate the hidden information of construction safety accident data,and has good performance in accuracy,generalization and application,with obvious advantages.

关 键 词：建筑安全 事故预测 联合驱动 差分自回归移动平均模型 支持向量机 

分 类 号：X92[环境科学与工程—安全科学] TP181[自动化与计算机技术—控制理论与控制工程]