In order to solve the problem of slow convergence speed of arithmetic optimization algorithm and difficulty in finding local optimal value, an improved arithmetic optimization algorithm IAOA with updating strategy of optimization candidate solution is proposed. First, the update strategy of each dimension value of each candidate solution in the arithmetic optimization algorithm is changed to multiplication and division or addition and subtraction operations. Secondly, the position update strategy of the algorithm candidate solution is optimized to increase the global exploration capability and candidate solution diversity, so that the algorithm can find a better value. The simulation experiment of 21benchmark functions shows that IAOA algorithm achieves the optimal value in 11 benchmark functions, and has faster convergence speed in most benchmark functions, compared with the Tartar sea squirt group algorithm, grey wolf optimization algorithm, Archimedes optimization algorithm and arithmetic optimization algorithm, in order to provide new ideas for solving arithmetic optimization algorithm.

In the field of machine learning, Extreme Learning Machine (ELM) has been widely used in classification and regression tasks. However, like many traditional machine learning algorithms, the classification results of ELM are often not good enough when facing imbalanced data. For this reason, we proposed an extreme learning machine algorithm with output weight adjustment called OWA-ELM, which can make the decision boundary of ELM move to majority classes, and improve the classification performance of imbalanced data. Specifically, in ELM, we add a reasonable increment $\Delta $ to the connection weights between hidden layer neurons and minority output neurons, so that the output value of minority output neuron increases. Finally, the classification accuracy of the minority class samples can be improved without significantly affecting the classification of the majority class samples. The performance of OWA-ELM was compared with ELM, WELM, CS-ELM and CCR-ELM. In the experiments on 22 data sets, the OWA-ELM algorithm has achieved 9 times optimal and 4 times suboptimal results on G-mean. In F-measure, 13 times optimal and one suboptimal results were obtained. Therefore, the OWA-ELM algorithm is effective to deal with imbalanced data classification.

针对飞鼠搜索算法(SSA)求解精度不高、容易早熟等缺点,提出一种采用反向学习和差分进化改进的飞鼠搜索算法(ODESSA).首先对最初种群使用反向学习生成其反向种群,增加初始种群的多样性.其次,在SSA中利用差分进化算法中的变异、交叉、选择机制扩大飞鼠搜索范围,以此增强算法的全局勘探能力.最后再次利用反向学习生成所有搜索个体的反向解,进一步增强算法的全局勘探能力和算法的求解精度,并将ODESSA与其他3种智能算法在12个基准函数上进行性能比较.结果表明,ODESSA有着最佳的寻优结果,收敛速度也快于SSA.

短时的降雨和温度等预报一直是天气预报中的重要问题.为了准确和及时预测局部区域的降雨及温度,提出了一种基于Attention和LSTM组合模型(ALSTM)的关联多值预测算法.该算法利用天气时间序列中的前期数据,对下一小时的降雨量和温度进行关联预测,以此实现对天气要素的多值预测.该算法首先对输入数据进行归一化处理;然后利用数据对ALSTM模型进行训练;最后将训练好的模型用于多值预测.将ALSTM模型与LSTM、BP以及基于LSTM的深度循环神经网络(DRNN)的预测结果进行了比较.实验结果表明,ALSTM模型的温度和降雨预测精度优于比较的其他模型,其平均预测精度在97％以上.

针对影响芒果产量的相关气象要素繁多,它们与产量之间的关联关系复杂、难以用数学函数准确地描述的问题,提出一种基于自注意力机制具有长短期记忆功能的双向门控循环单元和卷积神经网络组合(Self-attention C-BiGRU)模型.首先,利用CNN卷积层(1D CNN)提取局部特征;其次将Self-attention机制用于进一步提取依赖特征,然后双向门控循环单元(BiGRU)会充分考虑年份之间的关联性,学习长期依赖特征;最后,利用广西某地3个气象站所收集到的24个芒果生产周期年份(从前一年第22旬到当年第21旬)每旬9个气象要素及芒果产量数据进行分析建模,建立了芒果产量预测Self-attention C-BiGRU模型.实验结果表明,Self-attention C-BiGRU模型预测的产量与实际产量的均方根误差为10.67,比支持向量回归(SVR)、误差后向传播神经网络(BPNN)、门控循环单元(GRU)、基于注意力机制的双向门控循环单元(BiGRU-Attention)、门控循环单元和卷积神经网络组合模型(GRU-CNN)、双向门控循环单元和卷积神经网络组合模型(C-BiGRU)分别平均降低了37.7％、42.1％、17.6％、4.1％、5.3％和5.9％.Self-attention C-BiGRU模型具有较高的预测准确性,对提升广西芒果产业发展、推进农业信息化有重要意义.

Stock market has received widespread attention from investors. It has always been a hot spot for investors and investment companies to grasp the change regularity of the stock market and predict its trend. Currently, there are many methods for stock price prediction. The prediction methods can be roughly divided into two categories: statistical methods and artificial intelligence methods. Statistical methods include logistic regression model, ARCH model, etc. Artificial intelligence methods include multi-layer perceptron, convolutional neural network, naive Bayes network, back propagation network, single-layer LSTM, support vector machine, recurrent neural network, etc. But these studies predict only one single value. In order to predict multiple values in one model, it need to design a model which can handle multiple inputs and produces multiple associated output values at the same time. For this purpose, it is proposed an associated deep recurrent neural network model with multiple inputs and multiple outputs based on long short-term memory network. The associated network model can predict the opening price, the lowest price and the highest price of a stock simultaneously. The associated network model was compared with LSTM network model and deep recurrent neural network model. The experiments show that the accuracy of the associated model is superior to the other two models in predicting multiple values at the same time, and its prediction accuracy is over 95%.

Weather forecast is very important to people's daily life. In order to improve the accuracy of weather forecast, in this paper, an improved Gaussian disturbance weighted salp swarm algorithm (GDWSSA) and a GDWSSA-optimized kernel extreme learning machine (GDWSSA-KELM) model is proposed for weather forecasting. GDWSSA is proposed in the model to improve the shortcomings of the original algorithm (SSA) that the convergence speed is too slow and it is easy to fall into the local optimal solution. Simulation results show that GDWSSA outperforms SSA, GA, PSO and GWO in the benchmark functions. Then, GDWSSA optimizes the parameters in the KELM model called GDWSSA-KELM. Finally, the model is used to build the weather prediction model, the results show that the temperature prediction accuracy is better than the traditional machine learning methods. Therefore, the proposed GDWSSA-KELM model prediction model is promising to serve as a powerful auxiliary tool for weather forecast with excellent performance in temperature prediction.

针对樽海鞘群算法(Salp Swarm Algorithm,SSA)在寻优过程中存在的收敛速度较慢、容易陷入局部最优的缺点,提出了一种改进的采用莱维飞行策略的条件化更新的樽海鞘群算法(Levy Flight-based Conditional Updating Salp Swarm Algo-rithm,LECUSSA),并将其运用于分类算法的特征子集选择过程.首先,利用莱维飞行策略的长短跳跃特点对领导者位置进行随机更新,以增强全局最优的搜索能力;其次,增加对追随者位置的更新条件,让追随者不再盲目地跟随,从而加快收敛速度.在23个优化基准函数上对LECUSSA算法与其他算法进行了性能比较实验;并把算法运用到支持向量机(Support Vector Ma-chine,SVM)算法的分类特征子集选择中,采用8个UCI数据集对特征选择后的分类结果进行了性能比较实验.实验结果表明,LECUSSA具有良好的全局最优搜索能力和较快的收敛速度,利用LECUSSA算法进行特征选择后,能够找到最佳分类准确率的特征子集.

针对决策粗糙集属性约简在引入代价后分类精度不高的问题,对其中代价敏感与分类精度的平衡进行了研究.将分类总代价和近似分类质量作为属性约简过程中的约束条件,结合模拟退火方法,提出了一种基于代价敏感和近似分类质量的决策粗糙集属性约简(ARACOQ)算法.利用UCI数据集对算法进行了模拟实验,实验结果验证了ARACOQ算法的有效性,该算法能够在可承受代价范围内找到一个分类精度最高的属性约简集.

对热带果树种植适宜度的研究不仅能够减少果农和企业的灾害损失,而且有利于广西区域今后热带水果产业的布局和发展.影响热带果树种植的相关环境因素繁多且复杂.针对该问题提出了一种基于深度置信网络分类模型,对热带果树种植适宜度进行评价的方法.该模型由多层受限玻尔兹曼机和一层反向传播神经网络所组成.它能够学习复杂环境因素之间的内在特征,从而提高分类准确率.利用来自广西50多个气象站点的4000多条气象数据和地理信息数据,对模型进行了测试.实验结果表明,DBN模型的平均分类准确率达到了95.15％,优于传统的支持向量机和反向传播神经网络模型.该研究可为热带果树种植适宜度的分析和评价提供技术参考.

热带果树种植适宜度分析对于发展热带果树生产、减少灾害影响具有重要意义.针对统计方法自适应能力低的问题,提出一种将改进的遗传算法(GA)和误差反向传播( BP)神经网络相结合对热带果树种植适宜度进行评价的方法(GA-BP神经网络方法).首先采用常用的自适应算法对GA的交叉概率和变异概率进行改进,再通过GA得到优化的BP神经网络的初始权值和阈值,在此基础上BP神经网络进行进一步的学习,得到满足误差要求的解.将GA-BP神经网络与传统BP神经网络在twonorm数据集上进行了比较测试,并使用实际气象数据进行了热带果树种植适宜度分析.实验结果表明, GA-BP神经网络的分类准确率较传统BP神经网络高4%左右,网络训练时间减少了3个轮次左右.该方法对热带果树种植适宜度的分析和评价具有应用推广价值.

This paper studies the test-cost sensitive reduction on the positive region of decision theoretic rough sets (DTRS). Based on the traditional positive region-based reduction, this paper takes test cost into account for the pur-pose of discovering the positive region-based reduction with the minimum total test cost, and adopts the simulated annealing algorithm combined with traditional positive region-based reduction of DTRS algorithm to search for the result of the positive region-based reduction with the minimum test cost. Furthermore, this paper proposes a test-cost sensitive positive region-based reduction algorithm for DTRS (TCSPR), and analyzes the time complexity of the algorithm. The experimental results confirm the effectiveness of TCSPR that can find a positive region reduction with fewer attributes and less test cost in polynomial time, and whose solution is generally the optimum solution or second-best solution of the optimization objective, viz. the positive region-based reduction with the minimum total test cost, and whose classification ability is hardly reduced in part of the data sets.

人工蜂群算法具有较强的探索能力,但是开采能力差、搜索精度低、后期收敛速度慢。针对以上问题,本文提出一种基于混沌机制的人工蜂群算法,在搜索方程中引入历史平均最优解,避免探索和开采能力的失衡;迭代后期,若种群陷入局部极值,采用混沌序列对种群进行变异,以增强算法的开采能力和求解的质量,保持种群的多样性。经过函数测试结果表明,改进后的算法在求解速度和精度上均优于基本ABC算法和其他改进算法。

While basic glowworm swarm optimisation (GSO)is applied to solving the problem of global optimisation of functions,it has the problems of slow convergence in later period and being prone to falling into local optimum.Therefore we put forward a hybrid mutation-based algorithm of glowworm swarm optimisation.The algorithm uses chaotic mutation and boundary mutation to improve the diversity of the population,and prevents the algorithm from falling into local optimum;moreover this enables the algorithm to achieve a more accurate solution.Six standard test functions are applied to the test,results show that the improved glowworm swarm optimisation is better than the basic GSO in terms of the optimisation speed and precision and the convergence rate.

针对决策中普遍存在的代价问题,在模糊理论和决策粗糙集的基础上,对其代价敏感属性约简方法进行了研究.在模糊决策粗糙集属性约简中引入了包含误分类代价和测试代价的总代价.因此约简的目标不再只是考虑正域的大小,而是寻找使得总代价最小的最优属性子集.提出了一种模糊决策粗糙集代价敏感属性约简(COSAR)算法,该算法采用启发式方法搜索最优属性子集.给出了算法的步骤,并将该算法与已有的模糊粗决策粗糙集属性快速约简(QuickReduct)算法进行了性能对比.实验结果表明,COSAR算法比QuickReduct算法具有更强的属性约简能力、更低的分类总代价、更短的运行时间,且随着测试样本的增加,分类总代价差值也越来越大.

行人检测在智能汽车、监控系统和高级机器人等领域有广泛的应用.针对低分辨率和需要实时处理的行人检测应用场景,提出了采用一致化LBP直方图特征结合Ad-aBoost分类器的高效行人检测方法.在AdaBoost的训练过程中,采用了CART(Classification And Regression Tree)作为弱分类器,并结合基于Gini不纯度的剪枝方法,有效地提高了训练速度和分类器的性能.针对Caltech行人检测数据集的实验结果表明,基于LBP特征和CART弱分类器的AdaBoost分类行人检测方法具有较好的性能.

大学生的心理健康问题受到了各高校和相关研究人员的普遍关注.采用C4.5算法对大学生的心理健康问题进行分类分析.通过收集学生基础数据和SCL-90心理健康测评信息并提取相关属性,以C4.5算法构造大学生心理健康问题分类模型,发现影响学生心理健康的主要因素和它们之间的关系,并生成分类规则.利用测试集的数据对分类模型进行测试,测试结果表明该模型有较好的准确率.该模型一定程度上可为高校开展心理健康教育的规划、决策提供参考.

传统的三支决策粗糙集模型需要设置合适的阈值,需要运用该模型的人员具备相关的专业知识和经验,这阻碍了该模型在实际中的应用。针对此不足,本文提出用人工鱼群算法来自动生成阈值,而不需要先验知识。以样本的条件概率作为解空间,以决策风险最小化为目标,利用人工鱼群算法,能有效地从数据中学习到三支决策粗糙集模型所需要的阈值,使得风险损失最小。在部分UCI数据集上的实验表明,该算法在运行时间上和利用学习到的阈值构建的分类器的分类性能都明显优于自适应算法。

Fuzzy C-means clustering algorithm uses a random manner to select the cluster centers at the beginning,which makes fuzzy C-means clustering algorithm extremely sensitive to the selected initial cluster centers,and it is more easily to get the optimal solution in the local area,but the effect is not very well in the global scope. Ant colony clustering algorithm arbitrarily sets the probability of ants picking up or down the data object according to the priori knowledge,lack of rigorous mathematical basis. Focusing on the shortage of FCM algorithm and ant colony clustering algorithm,in this paper,apply the granular computing to the ant colony clustering algorithm,and combined the improved ant colony clustering algorithm and fuzzy C-means clustering algorithm,propose an improved fuzzy C-means clustering algorithm. Verified by experiments,the improved algorithm is better than the original algorithm on clustering effect.