基于Stacking集成算法的抛石护岸水毁破坏预测研究

抛石护岸在顶冲等极端情况下易发生水毁破坏,给人民的生命财产带来威胁。通过水槽试验获取496组样本数据,利用互信息(MI)筛选出6个关键特征属性,并采用支持向量机(SVR)、广义回归神经网络(GRNN)和随机森林(RF)等机器学习算法构建多个预测模型。然后,将这些模型作为基学习器,结合BP神经网络(BPNN)作为元学习器,采用Stacking集成学习方法构建抛石护岸破坏程度预测模型。最后,通过决定系数(R²)、均方根误差(R_RMSE)及平均绝对误差(M_MAE)等评价指标对模型性能进行评估。结果表明,Stacking模型在抛石护岸破坏高度、长度、范围上的平均R²为0.98、R_RMSE为0.02、M_MAE为0.03,相较于单一模型(SVR、GRNN、RF),Stacking模型的R_RMSE、M_MAE皆为最小,R²最高。在抛石护岸水毁破坏程度的预测中,融合的Stacking模型展现出更高的准确性与稳...     

IMERG降水卫星在贵州省三岔河流域降水预报误差评估中的应用

以贵州省三岔河流域为研究区域，利用2018年1月1日～2020年12月31日期间的76个气象站观测数据评估全球降水观测计划多卫星降水产品IMERG(Integrated Multi-satellitE Retrievals for GPM)的雨量站订正产品IMERG Final Run (IMERG＿FR)和准实时产品IMERG Early Run (IMERG＿ER)在岩溶地貌的小流域的降水观测准确度，通过设定不同的降水阈值评估IMERG两种降水产品IMERG＿FR和IMERG＿ER对不同降水量区间的探测能力，评估指标包括相关系数(C_CC)、相对偏差(R_RB)、均方根误差(R_RMSE)、探测率(P_POD)、临界成功指数(C_CSI)和误报率(F_FAR)。结果表明，在整体上，IMERG＿FR和IMERG＿ER在该小型流域有一定降水探测能力，对降水强度的估算精度还存在提升空间。     

基于1D CNN-SLSTM模型的太湖蓝藻密度预测方法

太湖中的水华爆发对太湖及其沿岸居民造成巨大影响,因此保证提前预测蓝藻密度非常重要。为精确预测太湖蓝藻密度,在长短时记忆网络（LSTM）模型的基础上,加入一维卷积模型,筛选并优化激活函数,提出一种基于1D CNN-SLSTM的预测模型预测蓝藻密度。试验结果表明,1D CNN-SLSTM模型的R_RMSE、M_MAPE、M_MAE值分别比单独使用LSTM模型降低30.38%、1.85%、16.89%,R值和N_NSE值则提升了0.08、0.17,验证了使用扩展型指数线性单元激活函数（Selu）的LSTM神经网络（1D CNN-SLSTM）预测效果最好。     

基于EEMD-Xgboost组合模型的渭北流域月径流序列模拟研究

为提升月径流序列的模拟精度,利用集合经验模式分解（EEMD）析出原始流量数据的模态分量,将极限梯度下降（Xgboost）作为预测函数,构建了基于EEMD-Xgboost的月径流预测模型,并应用EEMD-Xgboost模型训练了黄土坮塬区漆水站1951～1996年月径流序列变化规律,预测了1997～2020年的月径流量。结果表明,与单一Xgboost模型相比,EEMD-Xgboost模型的Nash-Sutcliffe效率(N_NSE)提升了20.27%、均方根误差(R_RMSE)减小了93.23%;且EEMD-Xgboost模型优于EEMD-ELM、EEMD-RF模型(N_NSE分别提高2.30%、3.49%;R_RMSE减小2.64%、11.75%)。EEMD-Xgboost混合模型集合了数据自适应分析与非线性映射的优点,改善了传统单一模型的预测能力。     

基于贝叶斯优化—随机森林回归的燃煤锅炉NOx预测模型

根据某超超临界1 050 MW燃煤机组实际运行数据,采用随机森林(RF)算法建立燃煤锅炉炉膛出口烟气中NO_x质量浓度预测模型,并利用贝叶斯优化(BO)进行超参数寻优,将BO-RF模型与网格搜索优化的RF模型(GSO-RF)进行对比。为了更好地评价预测模型,以平均绝对百分比误差δ_MAPE和决定系数R²作为评价指标,将所建立的BO-RF模型与目前常见的基于贝叶斯优化的BP神经网络(BO-BPNN)模型、最小二乘支持向量机(BO-LSSVM)模型进行比较。结果表明：BO-RF模型比GSO-RF模型的预测精度更高,且BO-RF模型的δ_MAPE为1.478%,R²为0.916 2,均优于BO-BPNN模型和BO-LSSVM模型的预测结果,证明BO-RF模型具有更高的预测精度和更优的泛化性能。     

基于FCM-WOA-LSTM的大坝变形预测模型及其应用

随着大坝变形监测资料的持续积累和变形测点数量的不断增多,预测分析全部变形测点往往需耗费大量的时间,容易造成反馈不及时的问题。对此,引入模糊C-均值聚类算法(FCM),根据大坝变形规律的相似程度进行分区,将鲸鱼优化算法(WOA)用于长短期记忆神经网络(LSTM)模型的参数优化,建立基于FCM-WOA-LSTM的大坝变形预测模型,以某混凝土双曲拱坝的实测变形资料作为样本数据进行预测分析,并与LSTM模型和SVM模型的预测结果进行对比。结果表明,FCM-WOA-LSTM模型预测结果的平均绝对误差M_MAE、均方误差M_MSE、均方根误差R_RMSE均为3种模型中最小,且拟合段的3个评价指标值和预测段的3个评价指标值均接近,FCM-WOA-LSTM模型具有更高的预测精度和更好的适用性。     

基于SWAT模型的呼玛河流域径流对气候变化的水文响应

呼玛河流域降水及冰雪融水的补给对于黑龙江干流径流量变化影响巨大,而气候变化是影响呼玛河降水及冰雪融水的一个重要因子。因此,通过收集呼玛河流域呼玛、新林、塔河、呼中气象站及呼玛桥水文站的数据,结合DEM数据、土地利用数据及土壤数据,构建适用于呼玛河流域的SWAT径流模型并进行月径流模拟,同时通过天气发生器设定温度与降水梯度,研究呼玛河流域径流对气候变化的响应。结果表明,SWAT模型在呼玛河流域具有良好的适应性,率定期及验证期的决定系数R²、纳什系数N_NSE均达到了模型的评价标准;呼玛河流域径流对降水变化的敏感性远高于气温,降水是该流域径流变化的主要控制因子。     

基于高斯回归的百万二次再热机组NOx排放预测分析及应用

研究百万二次再热机组的NO_x排放量预测模型。采用高斯回归对锅炉燃烧过程进行建模,采用Spearman相关性系数对建模历史数据进行相关性分析,从49项测点数据中选取相关性较高的28项测点数据作为自变量,NO_x生成量为标签变量。结果显示：测试数据显示搭建回归模型的R²为0.913 271,相对误差在10%范围内;现场部署该模型并实时计算,结果的准确性较高,为运行调整提供参考意见。     

流域未来径流和蓄水量预测不确定性的量化评估

水文模拟精度优劣与参与水文过程的多源不确定性相关,且其累积效应将会造成预测不确定性进一步扩大。因此,以黄河源区为例,利用极大似然不确定性估计法(GLUE)识别新安江模型有效参数组,通过耦合CMIP5下4种气候模式、3种气候变化情景、8种有效参数组,揭示了各源不确定性对流域径流量和蓄水量预测的影响,并采用方差分析方法(ANOVA)量化分离了其对月尺度流域径流量和蓄水量预测不确定性的相对贡献。结果表明,利用统计降尺度模型(SDSM)获得的降水量、最低和最高气温数据可较好的应用于研究区,其相关系数R²均大于0.70,且均方根误差R_RMSE小于30%;参数和GCMs不确定性对流域径流量的影响占主导地位,前者对流域蓄水量的相对贡献高达0.98,且多源不确定性之间交互作用对汛前和汛后的贡献大于非汛期。研究结果对于流域防洪减灾、降低水文模拟过程认知不确定性具有重要意义。     

中国不同气候区日总太阳辐射计算模型适用性分析及通用计算模型优化

利用中国5个气候区96个气象台站的1981—2010年的日值气象数据,对比分析12个基于日照百分率和12个基于温度的日总太阳辐射计算模型在不同气候区的适用性。采用判定系数（R²）、均方根误差（RMSE）、平均绝对误差（MABE）、平均误差（MBE）和全局性能系数（GPI）5个评价指标,确定各气候区最适宜的模型形式。以该模型为基准,建立适用于中国不同气候区的基于日照百分率和基于温度的日总太阳辐射通用计算模型。结果表明,三次方形式的基于日照百分率和基于日较差-平均温度的模型在各气候区计算精度均最高;以该模型为基础,建立适用于中国不同气候区的基于日照百分率和基于温度的日总太阳辐射通用计算模型,其平均R²分别为0.91和0.68。     

Performance analysis of support vector machine,Gaussian Process Regression,sequential quadratic programming algorithms in modeling hydrogen-rich syngas production from catalyzed co-gasification of biomass wastes from oil palm

The quest to attain net-zero emissions has increased the drive for more renewable energy potential from the co-gasification of biomass. The co-gasification of coconut shell and oil palm wastes is a potential technological route to produce hydrogen-rich syngas. However, the complexity of the gaseous-phase reaction often results in process uncertainties which could lead to energy and material wastage. Taking advantage of the data generated from the process, this study explores the performance of twelve machine learning algorithms built on the support vector machine (SVM), the Gaussian process regression (GPR), and the non-linear response quadratic model (NLRQM) using Sequential quadratic programming, and the Levenberg-Marquardt algorithms. The co-gasification of coconut shell and oil palm wastes blend catalyzed by Portland cement, dolomite, and limestone resulted in the maximum syngas production of 42 mol.%, 38 mol.%, 45 mol.%, respectively. The co-gasification process was modeled using SVM regression incorporated with linear, quadratic, and cubic kernel functions, GPR incorporated with rotational, squared, Matern 5/2, and exponential kernel functions, and non-linear response quadratic model (NLRQM) using sequential quadratic programming (SQP), and Levenberg-Marquardt (LM). The performance analysis of the models revealed that the SVM incorporated with linear kernel had the least performance with R² in the range of 0.3–0.7. Whereas the best performance in terms of prediction of the syngas composition was obtained using the NLRQM algorithm with an inbuilt SQP and LM algorithms. The observed syngas composition was consistent with predicted values with R² > 0.97 for the three catalyzed co-gasification processes. The low RMSE (<1) and MAE (<1) obtained from the models are indications of the robustness of the accurate prediction of the NLRQM-LM and NLRQM-SQP algorithms. The sensitivity analysis revealed that the co-gasification temperature, catalysts loading, and the blending amount play a significant role in the predicted syngas composition. However, the co-gasification temperature had the highest influence as indicated by the level of importance values.     

Application of machine learning algorithms for predicting the engine characteristics of a wheat germ oil–Hydrogen fuelled dual fuel engine

In this research work, performance and emission parameters of wheat germ oil (WGO) -hydrogen dual fuel was investigated experimentally and these parameters were predicted using different machine learning algorithms. Initially, hydrogen injection with 5%, 10% and 15% energy share were used as the dual fuel strategy with WGO. For WGO +15% hydrogen energy share the NO emission is 1089 ppm, which is nearly 33% higher than WGO at full load. As hydrogen has higher flame speed and calorific value and wider flammability limit which increases the combustion temperature. Thus, the reaction between nitrogen and oxygen increases thereby forming more NO. Smoke emission for WGO +15% hydrogen energy share is 66%, which is 15% lower compared to WGO, since the heat released in the pre-mixed phase of combustion is increased to a maximum with higher hydrogen energy share compared to WGO. Different applications including internal combustion engines have used machine learning approaches for predictions and classifications. In the second phase various machine learning techniques namely Decision Tree (DT), Random Forest (RF), Multiple Linear Regression (MLR), and Support Vector Machines (SVM)) were used to predict the emission characteristics of the engine operating in dual fuel mode. The machine learning models were trained and tested using the experimental data. The most effective model was identified using performance metrics like R-Squared (R²) value, Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). The result shows that the prediction by MLR model was closest to the experimental results.     

基于组合预测方法的风电场短期风速预测

基于持续法、人工神经网络法(ANN)和支持向量机(SVM)3种不同预测模型对内蒙古某风电场短期风速进行了预测研究,比较了不同单一预测模型的预测精度,并进行了4种不同预测模型的组合预测。计算结果表明,单一预测模型中支持向量机方法精度最高,而组合预测中3种方法组合的预测精度最高,并且组合预测精度均高于单一预测方法的精度。同时发现,当单一模型预测误差之间存在较强的负相关关系时,组合预测精度提高明显;而当单一模型预测误差之间存在较强的正相关关系时,则组合预测精度改进有限。     

基于PCA-GPR的锂离子电池剩余使用寿命预测

从充电过程中的电压-容量曲线中提取出一个与电池寿命高度相关健康因子（HI）。然后利用主成分分析（PCA）对影响电池寿命的多维因素进行分析和降维,结合高斯过程回归（GPR）机器学习方法提出一个基于PCA-GPR的锂离子电池剩余使用寿命预测模型。最后进行锂离子电池剩余使用寿命预测并与PCA-BP神经网络、PCA-支持向量机（SVM）模型进行比较。结果表明,利用该文提出的HI及预测模型可有效提高锂离子电池剩余使用寿命预测精度,其中通过贝叶斯优化器优化后的PCA-GPR模型的预测效果最佳。     

Comparison of artificial intelligence and empirical models for estimation of daily diffuse solar radiation in North China Plain

Accurate diffuse solar radiation (H_d) data is highly crucial for the development and utilization of solar energy technologies. However, due to expensive cost and technology requirements, measurements of H_d are not available in many regions of North China Plain (NCP), where the diffuse and direct solar radiation are affected by severe particulate pollution. Thus, development of models for precisely estimating H_d is indeed essential in NCP. On this account, the present studies proposed four artificial intelligence models, including the extreme learning machine (ELM), backpropagation neural networks optimized by genetic algorithm (GANN), random forests (RF), and generalized regression neural networks (GRNN), for estimating daily H_d at two meteorological stations of NCP. Daily global solar radiation and sunshine duration along with the estimated extraterrestrial radiation and maximum possible sunshine duration were selected as model inputs to train the models. Meanwhile, the proposed AI models were compared with the empirical Iqbal model to test their performance using measured H_d data. The results indicated that the ELM, GANN, RF, and GRNN models all performed much better than the empirical Iqbal model for estimating daily H_d. All the models underestimated H_d for both stations, with average relative error ranging from ?5.8% to ?5.4% for AI models and 19.1% for Iqbal model in Beijing, ?5.9% to ?4.3% and ?26.9% in Zhengzhou, respectively. Generally, GANN model had the best accuracy, and ELM ranked next, followed by RF and GRNN models. The ELM model had a slightly poorer performance but the highest computation speed, and both the GANN and ELM models could be highly recommended to estimate daily H_d in NCP of China.     

SVM-RBFNN组合模型在某混凝土双曲拱坝变形监测中的应用

大坝变形监测数据由于受水位、温度、时效等因素的影响,是一个非线性、非平稳的时间序列。考虑到支持向量机模型(SVM)对小样本、非线性问题有很好的预测效果,采用SVM模型刻画样本数据的趋势性信息;同时运用径向基函数神经网络(RBFNN)对残差序列进行分析和预测,并将时间序列的趋势项和误差项线性叠加,组成SVM-RBFNN组合预测模型。实例应用表明,SVM-RBFNN组合模型精度高于SVM模型,具有一定的实用性。     

Li2CO3-Na2CO3-K2CO3相图的理论预测

熔盐是一种重要的聚光太阳能系统传热流体,多元熔盐混合是满足实际需求的有效方法,而熔盐混合物的相图是其筛选的主要因素之一.该文基于热力学原理及吉布斯自由能对Li2CO3-Na2CO3-K2CO3的相图进行预测与计算,结果显示:该熔盐混合物的共晶点为421.2℃,Li2CO3、Na2CO3和K2CO3的质量分数分别为25....     

淮河上游区域GPM IMERG卫星降水数据应用评价

为系统评估GPM IMERG V05B卫星降水产品在淮河息县以上流域的应用,以流域内9个地面站实测雨量数据作为参考,应用统计指标、分类指标及极端降水指标,对2015～2017年GPM IMERG产品分别评价分析了日、月、年尺度上统计指标与分类指标,评估了不同雨强、极端降水探测性能。结果表明,GPM IMERG卫星降水产品与地面监测数据在日、月尺度的降水监测值表现出较好的相关性,总体呈高估现象,日尺度探测率高于0.8,但错报率、汉森和凯普斯得分不理想,月尺度上在8、9月份存在低估,年雨量误差10.0%～23.8%;GPM IMERG对小雨综合探测性能强,但错报率较高,对中雨探测精度较高;极端降水事件探测能力表现不稳定。     

基于改进的湿润地区站点与卫星降雨数据融合的洪水预报精度分析

目前卫星降雨数据是水文研究中重要的数据来源,但其精度尚未达到洪水预报的要求。将TRMM3B42RT、TRMM 3B42、GPM IMERG Early及GPM IMERG Late分别与站点实测降雨数据融合,并利用HEC-HMS模型进行洪水模拟,从而改进卫星降雨洪水预报精度。结果表明,卫星降雨数据在经过与实测降雨数据的融合处理后,误差明显降低,相关系数由0.7提高到0.9以上,进而提高了模拟洪水的精度,在洪水预报方面有一定的可行性。