钢的疲劳断裂是工业中最常发生的灾难之一。为有效预测钢疲劳,提出了一种基于多层置信规则库(Multilayer Belief Rule Base, MBRB)预测模型。首先,采用主成分分析对关键特征进行筛选。其次,利用具有可解释性约束的投影协方差矩阵自适应进化策略(P-CMA-ES)对模型参数进行优化,以提高模型的精度。最后,以美国国家材料科学研究所(NIMS) MatNavi的钢疲劳数据集为例进行了钢疲劳预测,验证了该模型的有效性,同时多层BRB解决了传统BRB组合规则爆炸的问题。与其他方法相比,该模型具有更高的精度与透明的推理过程。Fatigue fracture of steel is one of the most common disasters in industry. In order to effectively predict steel fatigue, a prediction model based on Multilayer Belief Rule Base (MBRB) was proposed. Firstly, principal component analysis was used to screen the key features. Secondly, the projection covariance matrix adaptive evolution strategy with interpretability constraints (P-CMA-ES) was used to optimize the parameters of the model to improve the accuracy of the model. Finally, the steel fatigue dataset of MatNavi of the National Institute of Materials Science (NIMS) of United States is used as an example to predict the effectiveness of the model, and the multi-layer BRB solves the problem of the explosion of traditional BRB combination rules. Compared with other methods, the model has higher accuracy and transparent inference process.
随着移动应用的快速发展,代码异味问题日益凸显,严重影响了软件的质量和性能。本文提出了一种基于机器学习的缓慢循环异味检测方法,旨在提高Android应用中代码异味的检测效率和准确性。研究首先构建了一个包含7000个样本的数据集,然后采用决策树(C4.5)、朴素贝叶斯(NB)、逻辑回归(LR)、随机森林(RF)和基于规则的归纳算法(JRip)五种机器学习算法进行缓慢循环异味的检测。实验结果表明,随机森林算法在查准率、查全率和F1值上均表现优异,其次是JRip算法。本研究的方法和结果为Android应用开发中代码异味的自动检测提供了有效的技术支持。With the rapid development of mobile applications, the issue of code smells has become increasingly prominent, severely affecting the quality and performance of software. This paper proposes a machine learning-based method for detecting slow loop smells, aiming to improve the efficiency and accuracy of detecting code smells in Android applications. The study first constructed a dataset containing 7000 samples, and then used five machine learning algorithms, including Decision Tree (C4.5), Naive Bayes (NB), Logistic Regression (LR), Random Forest (RF), and Rule-based Inductive Algorithm (JRip), to detect slow loop smells. The experimental results show that the Random Forest algorithm performed excellently in terms of precision, recall, and F1 score, followed by the JRip algorithm. The methods and results of this study provide effective technical support for the automatic detection of code smells in the development of Android applications.
混凝土抗压强度的高低直接影响着建筑物的安全和稳定性,传统的混凝土抗压预测方式周期长且成本高。针对解决混凝土抗压强度分析所面临的材料成分、分析过程复杂等问题,提出了一种基于证据推理(Evidential Reasoning, ER)和置信规则库(Belief Rule Base)的混凝土抗压强度预测方法。该方法首先利用随机森林(RF)算法得出部分指标的重要度,利用证据推理算法赋权融合各项指标。其次利用置信规则库专家系统将混凝土抗压指标中定性知识与定量的数据相结合,建立置信规则库预测模型。然后采用投影协方差矩阵自适应进化策略算法(P-CMA-ES)优化模型的参数。最后通过UCI数据库混凝土抗压强度数据集,对提出的方法进行了验证。实验结果表明,该方法保证了模型推理的透明,本文提出的预测方法具有较高的精度且具有一定的可解释性。The compressive strength of concrete directly affects the safety and stability of buildings, and the traditional concrete compressive strength prediction method has a long period and high cost. In order to solve the problems of material composition and complex analysis process faced by concrete compressive strength analysis, a concrete compressive strength prediction method based on Evidential Reasoning (ER) and Belief Rule Base was proposed. Firstly, the random forest (RF) algorithm is used to obtain the importance of some indicators, and the evidence inference algorithm is used to empower and fuse various indicators. Secondly, the confidence rule base expert system is used to combine the qualitative knowledge and quantitative data in the concrete compressive index, and the confidence rule base prediction model is established. Then, the projection covariance matrix adaptive evolution strategy algorithm (P-CMA-ES) was used to optimize the parameters of the model. Finally, the proposed method is verified by the concrete compressive strength dataset of the UCI database. Experimental results show that the proposed method ensures the transparency of model reasoning, and the prediction method proposed in this paper has high accuracy and interpretability.
随着移动应用的快速发展,代码异味问题日益凸显,严重影响了软件的质量和性能。本文提出了一种基于深度学习的过度耦合的消息链异味检测方法,旨在提高代码异味的检测效率和准确性。为了自动获取深度学习模型所需的大量标签数据,提出一种基于静态程序分析的正负样本自动生成方法,并实现自动化工具ASSD。然后,使用程序文本信息作为特征集训练三种深度学习模型,实现异味检测。实验结果表明,使用深度学习模型可以检测过度耦合的消息链异味。卷积神经网络模型在查准率、查全率和F1值上均表现优异,其次是循环神经网络模型。本研究的方法和结果为Android应用开发中代码异味的自动检测提供了有效的技术支持。With the rapid development of mobile applications, the problem of code odor has become increasingly prominent, seriously affecting the quality and performance of software. This article proposes a deep learning based over coupled message chain odor detection method aimed at improving the efficiency and accuracy of code odor detection. In order to automatically obtain the large amount of labeled data required for deep learning models, a method for generating positive and negative samples based on static program analysis is proposed, and the automation tool ASSD is implemented. Then, three deep learning models were trained using program text information as the feature set to achieve odor detection. The experimental results indicate that using deep learning models can detect the odor of over coupled message chains. Convolutional neural network models perform well in precision, recall, and F1 score, followed by recurrent neural network models. The methods and results of this study provide effective technical support for automatic detection of code smells in Android application development.
为解决中央处理器(Central Processing Unit, CPU)性能分析所面临的分析指标复杂、分析过程不具有可解释性、分析结果不可追溯的问题,提出了一种融合ER(Evidence Reasoning)和分层BRB(Belief Rule Base)的CPU性能分析模型.首先,利用ER...
详细信息
为解决中央处理器(Central Processing Unit, CPU)性能分析所面临的分析指标复杂、分析过程不具有可解释性、分析结果不可追溯的问题,提出了一种融合ER(Evidence Reasoning)和分层BRB(Belief Rule Base)的CPU性能分析模型.首先,利用ER算法从不同层面对处理器影响因素进行指标评估,其次,通过分层BRB实现对CPU性能的综合分析,最后,采用鲸鱼优化算法(Whale Optimization Algorithm, WOA)对模型参数优化.通过UCI数据库(University of California Irvine, UCI)计算机硬件数据集验证了模型的有效性.整个分析模型建立在ER算法上,保证了模型推理的可解释性,而分层BRB方法解决了传统BRB的组合规则爆炸问题,同时结合优化算法有效的提高模型的准确度.
暂无评论