BP神经网络对烟草销售量预测方法的改进研究

针对烟草销售量传统预测方法的不足,本文结合实际,应用BP神经网络模型对其进行了改进:把实际误差看作一组序列,进行逼近模拟,作为一个单独量加入最终结果,提高了预测的精度.本文的思路和方法可推广到其他社会经济数据序列的预测中去.     

MATLAB实现BP神经网络的煤炭需求预测

当前我国煤炭消费量不断增长,给煤炭供应带来较大压力,因此对未来煤炭消费情况的掌握显得尤为重要.本文采集了与煤炭消费相关的指标数据,采用BP神经网络预测方法,在MATLAB上,利用其中神经网络功能,建立合理训练模型,进行数据测试和预测,最后得出2010年中国煤炭消费量的预测结果,并提出相关结论.     

基于遗传神经网络的煤与瓦斯突出预测研究

为提高煤与瓦斯突出预测的效率和准确率,在煤与瓦斯突出影响因素分析的基础上,将遗传优化算法与BP神经网络相结合,对煤与瓦斯突出进行预测,避免了单纯神经网络易陷入局部最优问题。以平顶山八矿为研究对象,基于地质动力区划方法,搜集影响煤与瓦斯突出的相关数据,选取典型的突出样本对建立的遗传神经网络模型进行训练,对训练好的网络进行仿真预测,结果与实际情况相一致。     

基于BP神经网络的煤矿突水预测系统的设计

BP神经网络是一种具有自适应、自学习、高度非线性的神经网络,只要有对应的输入输出数据样本,无需了解其内部的具体逻辑关系,就可以根据新输入得到相应输出。将神经网络用于突水的预测,使用一组实际数据作为样本,对神经网络进行训练,测试,得到一个能够进行突水预测的网络。     

浅谈优化煤炭企业财务电算化

财务电算化是企业管理信息系统中最重要的组成部分,其作用在于实现会计核算和财务管理功能的最佳的结合和统一。它除了应有完善的会计业务核算功能外,还应有强大的会计数据分析、预测(如原材料需求、销售量、利润等的预测)、计划的制订(如采购计划、销售计划、库存计划、成本利润计划、现金预算等)和控制,支持各种中下层次的管理决策(主要是结构化决策)和某些高层次的半结构化决策,实现或辅助实现财务会计和管理会计的主要功能,并能使它们与企业的日常管理工作紧密有机地结合在一起。从而使会计职能发生根本性的变化,从过去会计主要进行内部收支核算、对外报送报表,转     

结合相似日与改进神经网络的短期光伏发电预测

为提高光伏发电预测精度,本文运用灰色关联理论分析历史气象数据,筛选出与待测日天气数据关联度较高的历史数据组作为相似日集合。建立经思维进化算法优化的BP神经网络预测模型,将上述相似日集合作为训练样本代入预测模型用于预测光伏发电功率。以澳洲某光伏系统的数据为例进行预测,结果表明,相比传统BP神经网络法、RBF神经网络,结合相似日与思维进化算法优化神经网络的光伏短期发电预测方法具有更高的预测精度。     

基于BP神经网络的黑龙江省煤炭产业发展前景分析

运用BP神经网络模型对黑龙江省煤炭市场需求进行了预测,并采用生产收入弹性、需求收入弹性和需求生产弹性差值对黑龙江省煤炭产业发展前景进行了综合分析,分析结果表明,黑龙江省煤炭产业在未来几年具有良好的发展情景和广阔的市场需求。     

基于BP神经网络的变电工程造价分析模型研究与预测

本文使用神经网络对变电站造价与其影响因素进行分析、预测。首先,运用模糊数学理论对变电站工程样本进行优选,选出与待估价工程类似的训练样本;然后,运用BP神经网络实现变电站工程造价和主要影响因素之间的复杂非线性映射,进而用已建的模型对工程进行造价预测;最后,通过比对实际值和预测值,验证所建模型的预测精度。     

灰色神经网络在轴承故障预测中的应用

为提高旋转机械中滚动轴承故障预测的精度,提出灰色神经网络预测模型。利用训练样本数据,使用灰色神经网络模型完成训练过程;基于已训练好的模型对未来时间点的运行状态进行拟合,实现轴承的故障趋势预测。相比采用单一的BP神经网络预测模型,该组合模型具有较高的精度,对轴承故障趋势预测有一定的现实意义。     

基于BP神经网络的黄金期货价格预测

在期货市场交易策略中,用预测目标作为衡量买卖时机交易的思想,是主流思想中的重要分支之一。为解决期货价格预测问题,文章对使用BP神经网络预测的方法进行探索。笔者以黄金1305期货合约2008年5月23日到2013年4月25日收盘价共857个数据为依据,建立了可自动调节参数的BP神经网络模型,并利用该网络对收盘价和收益率序列进行了滚动预测,根据预测结果进行期货品种的买卖。依据所构造策略进行交易,在模拟交易中获得了不错的收益。     

A big data driven framework for demand-driven forecasting with effects of marketing-mix variables

This study aims to investigate the contributions of promotional marketing activities, historical demand and other factors to predict, and develop a big data-driven fuzzy classifier-based framework, also called “demand-driven forecasting,” that can shape, sense and respond to real customer demands. The availability of timely information about future customer needs is a key success factor for any business. For profit maximization, manufacturers want to sense demand signals and shape future demands using price, sales, promotion and others economic factors so that they can fulfil customer's orders immediately. However, most demand forecasting systems offer limited insight to manufacturers as they fail to capture contemporary market trends, product seasonality and the impact of forecasting on the magnitude of the bullwhip effect. This paper aims to improve the accuracy of demand forecasts. In order to achieve this, a back-propagation neural network-based model is trained by fuzzy inputs and compared with benchmark forecasting methods on a time series data, by using historical demand and sales data in combination with advertising effectiveness, expenditure, promotions, and marketing events data. A statistical analysis is conducted, and the experiments show that the method used in the proposed framework outperforms in optimality, efficiency and other statistical metrics. Finally, some invaluable insights for managers are presented to improve the forecast accuracy of fuzzy neural networks, develop marketing plans for products and discuss their implications in several fields.     

Fashion retail forecasting by evolutionary neural networks

Recent literature on nonlinear models has shown that neural networks are versatile tools for forecasting. However, the search for an ideal network structure is a complex task. Evolutionary computation is a promising global search approach for feature and model selection. In this paper, an evolutionary computation approach is proposed in searching for the ideal network structure for a forecasting system. Two years’ apparel sales data are used in the analysis. The optimized neural networks structure for the forecasting of apparel sales is developed. The performances of the models are compared with the basic fully connected neural networks and the traditional forecasting models. We find that the proposed algorithms are useful for fashion retail forecasting, and the performance of it is better than the traditional SARIMA model for products with features of low demand uncertainty and weak seasonal trends. It is applicable for fashion retailers to produce short-term retail forecasting for apparels, which share these features.     

Extended revenue forecasting within a service industry

Revenue forecasting is an important topic for management to track business performance and support related decision making processes (e.g. headcount or capital expenditure). It focuses on how a business recognises operating revenue, which can differ from the point at which a sales order is won. Whilst there are many publications detailing forecasting theory, in a business context these largely focus on sales order recognition alone.This paper describes the development of a revenue forecasting tool appropriate for service provision. The organisation involved in the development of the revenue forecasting tool will remain anonymous for commercial reasons but will be referred to as “Organisation A”. The targeted outcome was to extend the forecast window from one month to three months with an error rate of no more than ±10%. The tool was required to consolidate supporting data, adopt appropriate analysis/projection techniques and extend the forecast window in a specific and complex business environment.The resulting tool returned high level results that were aligned to the original targets, and was developed with three components using a combination of projection approaches appropriate to the operating environment. Whilst limited to a specific service industry as a trial, the paper provides a useful reference point for revenue forecasting in complex service businesses and provides a basis for further research opportunities for extended revenue forecasting and business analysis approaches within other service industries.     

特高压输电线路工程造价预测模型研究——基于因子分析及BP 神经网络

本文在对特高压输电线路工程进行项目划分的基础上,分析识别出影响特高压输电线路工程造价的主要因素,并使用因子分析法对其中18 个主要影响因素进行度量,将得到的5 个因子作为输入、工程单位造价作为输出,构建了基于BP 神经网络的造价预测模型,以国内已建和在建的9 个特高压输变电工程中的75 段输电线路工程数据为样本进行了实证研究,结果证明预测模型对特高压输电线路工程造价进行预测具有可行性和较高的准确性,为特高压工程全生命周期内管理优化提供了一种新的思路和实现方法。     

Developing and Using Forecasting Models of Consumer Durables The Case of Color Television

Predicting the future sales of new and established products is a critical activity for companies to be able to plan and control their operations. Forecasting consumer durable sales is an especially difficult and challenging task since the marketplace is always changing. Barry Bayus, Saman Hong and Russell Labe present the concept of a market-driven forecasting model and discuss an application involving the forecasting of color television industry sales for RCA's Consumer Electronics Division. An important aspect of this application is that a single approach or model was inadequate to accurately forecast sales over the entire period since the introduction of color TV. Econometric and simulation models which were developed are described, along with their forecasting performance and management acceptance and use.     

Factors Affecting New Product Forecasting Accuracy in New Firms

This study identifies factors that seem to influence a new firm's ability to accurately forecast new product sales. William Gartner and Robert Thomas present a conceptual model and develop hypotheses that specify antecedent factors prior to new product launch, such as the founder's expertise and the marketing research methods used, as well as environmental factors occurring after product launch, such as competitive factors and market volatility, that influence new product forecasting accuracy. The hypotheses were tested with data collected from a survey of 113 new U.S. software firms. Some tentative guidelines for improving sales forecast accuracy among new firms are offered. Directions for future research are discussed.     

Robust Testing of the Utility‐based High‐technology Product Sales Forecasting Models Proposed by Decker and Gnibba‐Yukawa (2010)

Decker and Gnibba‐Yukawa (2010) propose an elegant utility‐based model for forecasting the sales of high‐technology products and suggest that the model yields forecasts that are highly accurate. However, this finding is based on forecasts for a total of only six holdout observations shared across three products. This number of observations is insufficient for reliable inferences to be drawn about the accuracy of a method and the use of such a small data set runs counter to an accepted principle of forecast evaluation. The authors’ proposed model was tested on more extensive data and sensitivity analysis applied to the results. No evidence was found that the utility‐based model could outperform a relatively simple extrapolative model despite the much greater effort involved in applying the proposed model. In addition, the utility‐based model is only applicable for forecasting sales during a narrow interval in a product's life cycle and requires several periods of historic sales data before it can be implemented. It also depends heavily on the accurate estimates of parameters that are determined outside the model (and which may depend on difficult judgments by managers) and assumes that consumers or households will only purchase the product once between the launch date and the forecast horizon. In light of this, it is argued that the utility‐based model is likely to have limited usefulness as a sales forecasting tool.     

支持向量机模型在断层破碎带围岩变形预测中的应用

围岩变形预测是隧道安全评价及其指导后期施工的重要依据,为提高变形预测精度,结合工程实践,提出了PSO-SVM-BP预测模型的思路。首先,利用三次样条插值及二次平滑法对变形数据进行预处理,为后期变形预测奠定基础;其次,利用粒子群算法对支持向量机进行参数优化,建立PSO-SVM模型,并对围岩变形进行初步预测;最后,利用BP神经网络进行误差修正,达到综合预测的目的,并利用工程实例进行检验,以验证预测模型的有效性。结果表明:初步预测结果的相对误差均小于5%,而误差修正后的预测精度被提高到0.97%,预测精度较高,验证了预测模型的有效性,可为类似研究提供参考。     

An exploratory Investigation of new product forecasting practices

To guide new product forecasting efforts, the following study offers preliminary data on new product forecasting practices during the commercialization stage (prelaunch and launch stage). Data on department responsibility for and involvement in the new product forecasting process, technique usage, forecast accuracy, and forecast time horizon across different types of new products are reported. Comparisons of new product forecasting practices for consumer firms versus industrial firms are also reported.
Overall, study results show that the marketing department is predominantly responsible for the new product forecasting effort, there is a preference to employ judgmental forecasting techniques, forecast accuracy is 58% on average across the different types of new products, and two to four forecasting techniques are typically employed during the new product forecasting effort. Compared to consumer firms, industrial firms appear to have longer forecast time horizons and rely more on the sales force for new product forecasting. Additional analyses show that there does not appear to be a general relationship between a particular department's involvement and higher forecast accuracy or greater satisfaction, nor does it appear that use of a particular technique relates to higher forecast accuracy and greater satisfaction. Countering previous research findings, the number of forecasting techniques employed also does not appear to correlate to higher forecasting accuracy or greater satisfaction. Managerial and research implications are discussed.