基于行为识别的课堂深度学习成绩预测模型研究

课堂学习行为智能识别和大数据分析为课堂高阶认知评价带来契机。利用数据技术挖掘学生课堂学习行为对深度学习成绩的影响关系,解决识别与评价课堂高阶认知不足的现实问题。以英语和化学两门学科为例,以课堂学习行为为自变量,以测试试卷深度学习成绩为因变量,通过高斯消元、回代总样本求均方差最优解等算法,分析出课堂“听讲”“阅读”等8种课堂行为对深度学习成绩的权重影响,构建课堂学习行为与深度学习成绩间的预测模型,依据模型可识别和评价学生的高阶认知是否发生,为常态课堂高阶认知规模化评价提供科学依据和技术支撑。

一种股票市场的深度学习复合预测模型

深度学习能够从大量原始数据中提取高级抽象特征而不依赖于先验知识,对于金融市场预测具有潜在的吸引力。基于"分解—重构—综合"的思想,提出了一种全新的深度学习预测方法论,并在此基础上构建了一种股票市场单步向前的深度学习复合预测模型——CEEMD-LSTM。在此模型中,序列平稳化分解模块的CEEMD能将时间序列中不同尺度的波动或趋势逐级分解出来,产生一系列不同特征尺度的本征模态函数(Intrinsic Mode Function,IMF);采用深度学习中适合处理时间序列的长短期记忆网络(Long-Short Term Memory,LSTM)分别对每个IMF与趋势项提取高级、深度特征,并预测下一交易日收盘价的收益率;最后,综合各个IMF分量以及趋势项的预测值,得到最终的预测值。基于3类不同发达程度股票市场的股票指数的实证结果表明,此模型在预测的两个维度即预测误差与预测命中率上均要优于其他参照模型。

基于燃烧模拟和深度学习的钢包烘烤优化

为了解决钢包烘烤的粗放式控制问题,提升钢包烘烤的燃烧效果,减少污染物排放量,提出了一种耦合燃烧模拟和深度学习技术进行数据驱动建模的钢包烘烤优化方法,同时采用了修改后的加权灰色气体模型确保CFD燃烧模拟模型计算在富氧条件下保持可靠.计算结果用于建立长短时记忆(LSTM)深度学习预测模型,预测氮氧化物(NO_(x))排放量、燃烧效率和升温速率,平均相对误差分别为0.278%,0.244%和0.189%.最后对计算结果进行多目标优化,可以在一定范围内减少氮氧化物排放量,提高燃烧效率和升温速率.结果表明:氮氧化物排放量、燃烧效率和升温速率的优化量分别为242.1 mg/m^(3),7.718%和3.631 K/h,达到了优化目的.

基于深度学习的全球平均表面温度年际信号时间序列的预测

利用集合经验模态分解(Ensemble Empirical Mode Decomposition,EEMD)有效地分解了全球平均表面温度(Global Mean Surface Temperature,GMST)时间序列,得到其不同尺度的、不同特征的子序列(Intrinsic Mode Function,IMF)。在此基础上,利用在预测长期、复杂、非线性变化的时间序列上具有显著优势的滑动自回归机器学习(Autoregressive Integrated Moving Average,ARIMA)模型和长短期记忆网络(Long Short-Term Memory,LSTM)模型开展GMST年际信号预测研究。结果表明:深度学习模型LSTM能很好地拟合并预测了长程相关性强的子序列(第2~6个IMF),而代表GMST年际尺度变化的IMF1则在一定程度上受到太平洋大西洋多重气候信号的影响和调制,因此进一步将3个气候指数作为预报前兆因子加入预测模型来更准确地预测IMF1的时间演变。通过利用多套GMST数据的对比,最终选定了考虑实时ENSO信息的LSTM(ENSO)模型来提前预测年际GMST信号,并预测2020年将有较大概率会成为史上最热的年份之一。

融合相似性原理的涡轮叶型流场预测方法研究

计算流体力学(CFD)方法是涡轮叶片等设计阶段性能评估的重要手段.然而,基于CFD的数值仿真方法通常比较耗时,难以满足涡轮叶型设计阶段快速迭代的需求.为实现快速性能评估并克服纯数据驱动预测模型泛化能力不足的问题,受到物理增强的机器学习思路的启发,将相似性原理与深度学习模型相结合,提出了一种泛化能力强的涡轮叶型流场预测新方法.以涡轮叶片表面等熵马赫数分布预测为例,提出采用相似性原理对叶型几何变量和气动参数进行归一化,进而在归一化参数空间构建训练样本集与深度学习预测模型,由此建立统一的流场预测模型,对几何尺寸、边界条件差异较大的叶型气动性能进行评估.在完成模型训练后,对归一化条件下不同工况/不同形状叶型的流场、真实环境下不同工况/不同尺寸叶型的流场以及GE-E3低压涡轮不同截面叶型的流场进行预测,结果表明预测结果的分布曲线与CFD评估结果吻合良好,平均相对误差在1.0%左右,由此验证了所提出的融合相似性原理的流场预测模型的精度与泛化能力.

基于深度学习模型的船舶碳排放时空预测研究

船舶碳排放的预测研究在挖掘碳排放时空分布特征、揭示排放时空演化规律、制订和调整脱碳政策等方面具有重要的理论与实践意义。然而目前,船舶航行状态辨识的准确性依旧制约着船舶碳排放测算的精度,而且,船舶碳排放预测研究多集中于预测其在时间维度的演化规律,尚未考虑兼顾时空双重维度的多维预测。因此,为解决船舶碳排放测算中航行状态辨识误判率高,以及传统碳排放预测方法难以兼顾空间相关性等问题,首先,本文以轨迹段为基本单元,在综合考虑航行状态持续时间、地理活动范围和航向变化等多维运动特征的基础上,提出了一种新的时空轨迹搜索算法,实现了航行状态的精准辨识;其次,本文引入了深度预测学习领域的ConvLSTM模型,构建了一种新的船舶碳排放时空预测框架,并提出一种基于GIS空间识别技术的高分辨率时空序列数据集构建方法。该框架不仅可预测船舶碳排放在时间维度上的演化规律,还可通过卷积运算,充分挖掘碳排放在空间上的局部依赖关系;最后,以我国渤海海域为例,基于船舶AIS数据,开展了船舶航行状态辨识和碳排放预测实验,结果表明:本文提出的航行状态辨识算法准确率达90%以上,有效解决了仅依据实时航速和主机负载划分航行状态的误判问题;而且,本文提出的船舶碳排放测算模型可在排放数据集上稳定收敛,并且能够准确预测排放的热点区域。本研究为我国船舶碳排放核算制度的完善,碳排放峰值、时间和区域的研判,以及碳中和路径的制订提供了重要的理论依据。

基于多种模型的流量预测研究

6G是下一代无线通信网络的发展方向,6G网络要求更高的峰值数据速率、移动性和更无处不在的智能连接。由于不同的场景和业务需要不同的网络切片(Network Slicing)承载,网络的性能指标各异,对各种资源的需求量也不尽相同,因此需要对网络流量在时间和空间上进行预测。对时空流量建模,并提出了3种以深度学习为基础的流量预测方法:基于深度学习的时空数据预测模型(DeepST)、残差网络(ResNet)模型和卷积长短期记忆网络模型(ConvLSTM),针对每种模型的应用场景和优劣予以研究陈述,在此基础上完成模型的有效性验证和比较;最后,仿真验证。仿真结果表明,基于ResNet的流量预测模型预测效果更佳。

电力系统中重合闸控制策略分析

重合闸控制策略在电力系统中具有重要作用,保证在短路或其他故障后能够迅速恢复供电。本文对目前使用的各种重合闸控制策略进行了深入分析。通过实际数据表明,采用适应性时间-电流特性的控制策略可以提高90%以上的成功重合闸率。此外,为确保更高的系统可靠性和稳定性,建议采用基于深度学习的预测模型对即将发生的短路进行预测和应对。

Anomaly detection of earthquake precursor data using long short-term memory networks

Earthquake precursor data have been used as an important basis for earthquake prediction.In this study,a recurrent neural network(RNN)architecture with long short-term memory(LSTM)units is utilized to develop a predictive model for normal data.Furthermore,the prediction errors from the predictive models are used to indicate normal or abnormal behavior.An additional advantage of using the LSTM networks is that the earthquake precursor data can be directly fed into the network without any elaborate preprocessing as required by other approaches.Furthermore,no prior information on abnormal data is needed by these networks as they are trained only using normal data.Experiments using three groups of real data were conducted to compare the anomaly detection results of the proposed method with those of manual recognition.The comparison results indicated that the proposed LSTM network achieves promising results and is viable for detecting anomalies in earthquake precursor data.

复杂地层中盾构掘进速度的调控分析——以新建铁路横琴至珠海机场段HJZQ-2标隧道工程为例

为解决复杂软弱地层中土压平衡盾构掘进速度难以用理论方法预测调控的问题,基于皮尔森相关系数分析了掘进参数与掘进速度的相关性,并提出盾构掘进速度的深度学习预测模型。模型运用粒子群优化算法对BP神经网络的权值与偏置值进行优化,以克服基于梯度下降算法的传统BP神经网络易陷入局部最小值和预测误差大等缺点,预测模型将地质参数与掘进参数作为输入值,盾构掘进速度作为输出值(预测值)。以新建铁路横琴至珠海机场段HJZQ-2标隧道工程为依托,基于贯入度与掘进速度的相关性最高且呈正相关的分析结果,采用监测数据对模型进行训练,利用训练后的深度学习模型对掘进速度进行预测分析。结果显示,具有2层隐藏层的深度学习PSO-BP模型的预测误差基本控制在±4 mm/min(误差在10%以内),满足实际工程要求,从而验证了模型的有效性与适用性。

Short-time prediction for traffic flow based on wavelet de-noising and LSTM model

Aiming at the problem that some existing traffic flow prediction models are only for a single road segment and the model input data are not pre-processed,a heuristic threshold algorithm is used to de-noise the original traffic flow data after wavelet decomposition.The correlation coefficients of road traffic flow data are calculated and the data compression matrix of road traffic flow is constructed.Data de-noising minimizes the interference of data to the model,while the correlation analysis of road network data realizes the prediction at the road network level.Utilizing the advantages of long short term memory(LSTM)network in time series data processing,the compression matrix is input into the constructed LSTM model for short-term traffic flow prediction.The LSTM-1 and LSTM-2 models were respectively trained by de-noising processed data and original data.Through simulation experiments,different prediction times were set,and the prediction results of the prediction model proposed in this paper were compared with those of other methods.It is found that the accuracy of the LSTM-2 model proposed in this paper increases by 10.278%on average compared with other prediction methods,and the prediction accuracy reaches 95.58%,which proves that the short-term traffic flow prediction method proposed in this paper is efficient.

Probabilistic interval prediction of metro-to-bus transfer passenger flow in the trip chain

To accurately analyze the fluctuation range of time-varying differences in metro-to-bus transfer passenger flows,the application of a probabilistic interval prediction model is proposed to predict transfer passenger flows.First,bus and metro data are processed and matched by association to construct the basis for public transport trip chain extraction.Second,a reasonable matching threshold method to discriminate the transfer relationship is used to extract the public transport trip chain,and the basic characteristics of the trip based on the trip chain are analyzed to obtain the metro-to-bus transfer passenger flow.Third,to address the problem of low accuracy of point prediction,the DeepAR model is proposed to conduct interval prediction,where the input is the interchange passenger flow,the output is the predicted median and interval of passenger flow,and the prediction scenarios are weekday,non-workday,and weekday morning and evening peaks.Fourth,to reduce the prediction error,a combined particle swarm optimization(PSO)-DeepAR model is constructed using the PSO to optimize the DeepAR model.Finally,data from the Beijing Xizhimen subway station are used for validation,and results show that the PSO-DeepAR model has high prediction accuracy,with a 90%confidence interval coverage of up to 93.6%.

Terminal Traffic Flow Prediction Method Under Convective Weather Using Deep Learning Approaches

In order to improve the accuracy and stability of terminal traffic flow prediction in convective weather,a multi-input deep learning(MICL)model is proposed.On the basis of previous studies,this paper expands the set of weather characteristics affecting the traffic flow in the terminal area,including weather forecast data and Meteorological Report of Aerodrome Conditions(METAR)data.The terminal airspace is divided into smaller areas based on function and the weather severity index(WSI)characteristics extracted from weather forecast data are established to better quantify the impact of weather.MICL model preserves the advantages of the convolution neural network(CNN)and the long short-term memory(LSTM)model,and adopts two channels to input WSI and METAR information,respectively,which can fully reflect the temporal and spatial distribution characteristics of weather in the terminal area.Multi-scene experiments are designed based on the real historical data of Guangzhou Terminal Area operating in typical convective weather.The results show that the MICL model has excellent performance in mean squared error(MSE),root MSE(RMSE),mean absolute error(MAE)and other performance indicators compared with the existing machine learning models or deep learning models,such as Knearest neighbor(KNN),support vector regression(SVR),CNN and LSTM.In the forecast period ranging from 30 min to 6 h,the MICL model has the best prediction accuracy and stability.