基于GRU和XGBoost的矿压显现规律预测

采用光纤传感器监测的光纤频移值对矿压显现规律进行表征的过程中,传感器采集的数据存在缺失现象,无法准确预测矿压显现规律。针对该问题,以千秋煤矿为工程背景,在假设光纤下半部分数据丢失的前提下,引入GRU(门控循环单元)和LSTM(长短期记忆网络)2种预测模型,对缺失的光纤频移值进行对比预测,得出GRU模型的收敛速度优于LSTM模型的收敛速度,说明基于GRU模型的缺失值处理方法较优。将原始完整的光纤频移值转换为可表征矿压显现位置的光纤平均频移变化度,引入XGBoost(极端梯度提升)模型和BP神经网络模型进行对比预测,XGBoost模型能准确预测出测试集中所有出现“尖峰”的位置,而BP神经网络模型只预测出2处“尖峰”位置,说明XGBoost模型的预测效果优于BP神经网络模型的预测效果。将预测出的光纤频移缺失值替换至缺失位置,形成“完整”光纤频移值数据,将该数据转换为光纤平均频移变化度后,采用XGBoost模型进行预测。验证结果表明:LSTM模型及GRU模型均可准确预测出光纤下半部分的数据,且GRU模型准确性较LSTM模型准确性高;使用XGBoost可准确预测出测试集中出现的周期来压;通过GRU模型预测出的缺失数据经整合至缺失位置后,使用XGBoost模型仍可进行有效的矿压预测。     

基于联合特征和XGBoost的活动语义识别方法

针对以往活动语义识别研究单纯提取时间维度上的序列特征以及周期特征、缺乏对空间信息的深度挖掘等问题,提出一种基于联合特征和极限梯度提升（XGBoost）的活动语义识别方法。首先,挖掘时间信息中的活动周期性特征和空间信息中的经纬度特征;然后,使用经纬度信息通过具有噪声的基于密度的聚类（DBSCAN）算法提取空间区域热度特征,将这些特征组成特征向量来刻画用户活动语义;最后,采用集成学习方法中的XGBoost算法建立活动语义识别模型。在FourSquare的两个公共签到数据集上,基于联合特征的模型比基于时间特征的模型在识别准确率上提高了28个百分点,与上下文感知混合（CAH）方法和时空活动偏好（STAP）方法对比,所提方法的识别准确率分别提高了30个百分点和5个百分点。实验结果表明所提方法与对比方法相比在活动语义识别问题上更加准确有效。     

基于鱼群运动特征和XGBoost的异常水质监测

使用鱼群运动行为特征与XGBoost分类器相结合的方法实现水质监测。根据水质变化对鱼群运动行为的影响,引入从鱼群的运动行为中提取出来的一些具有代表性的特征参数,如中心距指数、离散度、分布面积 3个特征参数;其次分别提取正常和异常水质中鱼群游动距离、速度、加速度、曲率、邻近特征5个特征参数,建立水质异常评价因子数据库;最后将不同评价因子输入到XGBoost分类器中进行分类识别。实验结果表明:中心距指数、离散度、分布面积 3个特征参数能够高效、准确地反映水质状况。     

Analyzing process parameters for industrial grinding circuit based on machine learning method

The grinding and classification processes are systematic engineering that must comprehensively consider the influence of several factors to ensure good grinding fineness. Based on the machine learning method, this study analyzed the full process parameters (i.e., ball mill power, fresh ore feed rate, hydrocyclone feed pump power, hydrocyclone pressure, mill feed water flow rate, dilution water flow rate, and sump level) for industrial grinding circuit. The collected real data (42,101 records) were employed to train and test the extreme gradient-boosting (XGBoost) regression model. The XGBoost model’s prediction ability and accuracy were evaluated and analyzed. The validated model was employed to evaluate the relative importance and influence mechanisms of process parameters. It was found that hydrocyclone feed pump power, dilution water flow rate, hydrocyclone pressure, and mill feed water flow rate significantly affected the grinding fineness, which were consistent with the actual operation of grinding circuit.     

Air pollution forecasting based on attention-based LSTM neural network and ensemble learning

With air pollution having become a global concern, scientists are committed to working on its amelioration. In the field of air pollution prediction, there have been good results in experimental research so far, but few studies have integrated weather forecast information and the properties of air pollution drift. In this work, we propose a novel wind-sensitive attention mechanism with a long short-term memory (LSTM) neural network model to predict the air pollution - PM2.5 concentrations by considering the influence of wind direction and speed on the changes of spatial–temporal PM2.5 concentrations in neighbouring areas. Preliminary predictions for PM2.5 are then made by an LSTM neural network regarding neighbouring pollution; these predictions are “paid attention to” and we finally apply an ensemble learning method based on e X treme G radient B oosting (XGBoost) to combine the preliminary predictions with weather forecasting to make second phase predictions of PM2.5. The experiment is conducted using PM2.5 data and weather forecast data. Our results illustrate that the proposed method is superior to other methods in predicting PM2.5 concentrations, including multi-layer perceptron, support vector regression, LSTM neural network, and extreme gradient boosting algorithm.     

新冠肺炎疫情影响下基于多源数据驱动的电力系统负荷预测方法

新冠肺炎疫情的爆发及快速传播给中国乃至全球造成了深远的影响,对社会带来全方面影响的同时也改变了大众的能源消费习惯,催生了新的电力需求侧的变革与契机。疫情对社会各方面的影响可以通过病例数、在线办公人数、电力市场数据、外卖快递数据、社交距离数据、城市亮度数据与二氧化氮等数据得到更为直接的展现,利用大数据相关系数算法,结合历史数据与天气数据,表征疫情下各个社会属性量与电力系统负荷之间的强弱联系。提出了疫情下基于极端梯度增强算法(extreme gradient boosting,XGBoost)分类器的电力系统负荷预测模型,利用新冠肺炎病例数据将地区划分为重度疫情期、轻度疫情期与疫情恢复期,探讨不同时间段新冠肺炎疫情影响下,各个社会属性量数据与电力系统负荷之间的关系,提出了多源数据驱动下的电力系统负荷预测模型,相关试验结果验证了提出方法的有效性与科学性。     

改进的XGBoost模型在股票预测中的应用

随着时代的不断进步，人民生活水平日益提高。在解决温饱问题之余，有了可供投资的余财。越来越多的人将目光转向股市投资，为股市发展提供了资金条件。然而在纷繁复杂的股票市场，如何寻找最优股成为亟待解决的问题。这不仅是投资者单方面的困惑，也是股票预测领域中学者们所关心的重点。通过网格搜索算法对XGBoost模型进行参数优化构建GS-XGBoost的金融预测模型，并将该模型运用于股票短期预测中。分别以中国平安、中国建筑、中国中车、科大讯飞和三一重工2005年4月至2018年12月28日的每日收盘价作为实验数据。通过实验对比，相较于XGBoost原模型、GBDT模型以及SVM模型，GS-XGBoost模型在MSE、RMSE与MAE三个评价指标上都表现出较好的预测结果。从而验证，GS-XGBoost金融预测模型在股票短期预测中具有更好的拟合性能。     

高海拔典型环境下电能表误差与环境影响因素研究

本文利用国网西藏典型环境试验基地监测平台的电能表基本误差试验数据,对高海拔环境下电能表误差与环境因素的关系进行了分析。首先通过建立XGBoost模型,利用环境数据,准确识别并预测了电能表超差。然后,通过建立多元线性回归模型,定量分析了环境因素对电能表误差的影响。最终得出环境因素中气温是影响电能表超差、误差数值的最重要因素,光照和湿度的影响也比较重要,风速、气压、雨雪的影响较小。     

基于XGBoost和改进灰狼优化算法的催化裂化汽油精制装置的辛烷值损失模型分析

为了降低催化裂化汽油精制装置的辛烷值损失,基于机器学习技术和改进灰狼优化算法建立了汽油辛烷值损失的预测和优化模型。首先通过Pearson相关系数法、最大互信息系数法(MIC)和基于随机森林的特征选择方法分别对影响汽油辛烷值的367个特征进行训练获得各特征的重要度评分,对3种方法的结果按权重法进行融合获得最终的特征重要度排序,根据特征重要度占比之和超过 95%的指标,选出25个特征作为建模主要变量;然后基于XGBoost算法建立汽油辛烷值损失预测模型,对比其他机器学习模型,验证了XGBoost在测试集上的预测性能最优,其均方根误差(RMSE)、平均绝对误差(MAE)和决定系数(R²)分别为1.3197、0.3581和0.9981;最后采用汽油辛烷值损失值与主要变量的映射函数作为目标函数,建立关于汽油辛烷值损失值最小的单目标优化模型,为了提高模型的求解速率和准确度,基于sigmoid函数的收敛因子调整策略和个体更新的差分变异策略,提出了一种改进的差分灰狼优化算法。结果表明,优化后的样本辛烷值损失值均减小到0.4左右,同时86.15%的样本辛烷值损失降幅在60%~80%之间,说明建立的优化模型和所提出的改进差分灰狼优化算法是合理的。通过数据挖掘技术建立的降低汽油辛烷值损失模型可以尽量减少汽油精制过程中的辛烷值损失,为石化企业和运营商提供决策分析。     

Xgboost application on bridge management systems for proactive damage estimation

Bridge inspection is one of the most fundamental tasks in bridge management practices. Because of limited professional manpower and budget constraints, providing prior information about possible damage can reduce inspection errors and time. The purpose of this study was to estimate the condition of bridges at a damage level, considering various influencing factors for seven different damage types by six different main structure types, using data from the Korean Bridge Management System. The extreme gradient boosting (XGBoost) method was used because it has the advantage of not assuming determinacy and independence, and it clearly can handle the numerous variables that affect damage to bridges. As a result, out of the 38 decision trees that were generated, 36 trees were derived with significant performance measures. The influence of the variables was calculated by the Shapley Additive Explanation (SHAP) value. Age, average daily truck traffic, vehicle weight limit, total length, and effective width were found to be the major factors that influenced damage to bridges. This study confirmed that more detailed structural factors were significant contributors to severe damage to complex structural designs and the use of multiple kinds of materials, such as the cross-sectional properties of girders for the concrete deck of bridges with steel girders compared to the properties of the decks for bridges made of a simple slab of reinforced concrete. The research findings emphasized the benefits of artificial intelligence in the analysis of the conditions of bridges and showed its potential for use in network-level decision making for preventive maintenance.