Suitable error evaluation criteria selection in the wind energy assessment via the K-means clustering algorithm

In this paper, wind energy potential of four locations in Xinjiang region is assessed. The Weibull distribution as well as the Logistic and the Lognormal distributions are applied to describe the distributions of the wind speed at different heights. In determining the parameters in the Weibull distribution, four intelligent parameter optimization approaches including the differential evolutionary, the particle swarm optimization, and two other approaches derived from these two algorithms and combined advantages of these two approaches are employed. Then the optimal distribution is chosen through the Chi-square error (CSE), the Kolmogorov–Smirnov test error (KSE), and the root mean square error (RMSE) criteria. However, it is found that the variation range of some criteria is quite large, thus these criteria are analyzed and evaluated both from the anomalous values and by the K-means clustering method. Anomaly observation results have shown that the CSE is the first one should be considered to be eliminated from the consequent optimal distribution function selection. This idea is further confirmed by the K-means clustering algorithm, by which the CSE is clustered into a different group with KSE and RMSE. Therefore, only the reserved two error evaluation criteria are utilized to evaluate the wind power potential.     

Optimizing mixture properties of biodiesel production using genetic algorithm-based evolutionary support vector machine

Nowadays, biodiesel is used as one of the alternative renewable energy due to the increasing energy demand. However, optimum production of biodiesel still requires a huge number of expensive and time-consuming laboratory tests. To address the problem, this research develops a novel Genetic Algorithm-based Evolutionary Support Vector Machine (GA-ESIM). The GA-ESIM is an Artificial Intelligence (AI)-based tool that combines K-means Chaotic Genetic Algorithm (KCGA) and Evolutionary Support Vector Machine Inference Model (ESIM). The ESIM is utilized as a supervised learning technique to establish a highly accurate prediction model between the input--output of biodiesel mixture properties; and the KCGA is used to perform the simulation to obtain the optimum mixture properties based on the prediction model. A real biodiesel experimental data is provided to validate the GA-ESIM performance. Our simulation results demonstrate that the GA-ESIM establishes a prediction model with better accuracy than other AI-based tool and thus obtains the mixture properties with the biodiesel yield of 99.9%, higher than the best experimental data record, 97.4%.     

基于人工蜂群算法与BP神经网络的水质评价模型

针对BP网络水质评价模型的不足,引入人工蜂群(ABC)算法,将求解BP神经网络各层权值、阀值的过程转化为蜜蜂寻找最佳蜜源的过程,提出了一种新的结合人工蜂群算法的BP网络水质评价方法(ABC-BP)。并以2000—2006年渭河监测断面的10组实测数据作为测试样本对其水质进行了评价,实验结果表明该方法得到的水质评价结果准确,并具有很强的稳定性和鲁棒性。     

铁路危险货物办理站点优化配置方法研究

铁路危险货物运输办理站点的优化配置,涉及到需求分布、设置地点、区域规划、环保要求、业务功能以及与其他运输方式的分工、衔接等,是一个复杂的系统工程。分析了已有铁路危险货物办理站点整合优化模型的不足,首先从铁路部门角度综合考虑现有铁路危险货物办理站点的各方面条件,进行多准则决策评价,从运输需求者角度考虑运输方式多属性决策,进行铁路危险货物办理站点的综合效用评价,并基于投资成本、环境风险、运输效用和区域公平性等方面构建了铁路危险货物办理站点优化配置模型,为了有效解决多站点、多需求、多区域、多目标铁路危险货物办理站点优化配置模型因素复杂、规模大、层次多的问题,提出了运用大系统分解协调的多目标遗传算法求解模型的思路。     

改进的BP算法在黄河下游枯季径流预测中的应用

本在黄河下游地区应用多层前向人工神经网络理论,通过改进BP算法,建立下游枯季径流预测的BP神经网络模型,使用花园口-利津水站26年的完整序列测流资料训练和检验网络并用于预测,结果表明,通过本次研究建立的BP网络模型是合理的,可靠的,它较好地反映了黄河下游的枯季径流规律,可为今后黄河流域水资源的统一调度,管理,尤其是预防黄河下游再次出现断流提供科学依据。     

煤炭产业国际竞争力的遗传—投影寻踪评价

根据煤炭产业发展的实际状况,提出了煤炭产业竞争力评价的指标体系。采用探索性数据分析—投影寻踪法,结合遗传算法,建立了遗传—投影寻踪综合评价模型。以全球11个主要产煤国为对象进行了实证研究,验证了该方法的科学性和实践的可行性,提出提升我国煤炭产业竞争力的对策建议。     

遗传投影寻踪回归在城市环境质量评价中的应用

引入投影降维的思想,将遗传投影寻踪与回归分析技术运用到城市环境质量评价中。将此技术与神经网络方法进行实例比较,投影寻踪回归方法不但可以合理地作出环境质量的综合评价,而且消除了神经网络方法中类别判断不够精确的影响。     

基于L-M神经网络的道路交通噪声预测研究

神经网络具有很强的预测功能.根据石家庄公路交通噪声的实测数据,利用L-M优化算法的多层神经网络预测模型进行道路交通噪声的预测,经检验,计算值与实测值接近,预测精度令人满意.     

响应面法和神经网络优化Acinetobacter sp. DNS32发酵基质

为了提高阿特拉津降解菌Acinetobacter sp.DNS32的产量,分别采用响应曲面法和基于人工神经网络的遗传算法对阿特拉津降解菌DNS32发酵培养基中3个重要基质成分(玉米粉、豆饼粉、K2HPO4)进行优化研究。响应曲面法确定3种成分的含量为玉米粉39.494 g/L,豆饼粉25.638 g/L和K2HPO43.265 g/L时,预测发酵活菌最大生物量为7.079×108CFU/mL,实测量为7.194×108CFU/mL;人工神经网络结合遗传算法优化确定3种主要成分含量为玉米粉为39.650 g/L,豆饼粉为25.500 g/L,K2HPO4为2.624 g/L时,预测最大值为7.199×108CFU/mL,实测量为7.244×108CFU/mL;最终确定培养基配方:玉米粉为39.650 g/L,豆饼粉为25.500 g/L,K2HPO4为2.624 g/L,CaCO3为3.000 g/L,MgSO4.7H2O和NaCl均为0.200 g/L;优化后阿特拉津降解菌DNS32发酵生物量比优化前提高了36.6%。结果表明,在阿特拉津降解菌DNS32发酵培养基组分优化方面,响应面法和基于人工神经网络的遗传算法都是可行的,基于人工神经网络的遗传算法具有更好的拟合度和预测准确度。