基于策略学习的机票动态定价算法

机票动态定价旨在构建机票售价策略以最大化航班座位收益。现有机票定价算法都建立在提前预测各票价等级的需求量基础之上,会因票价等级需求量的预测偏差而降低模型性能。为此,提出基于策略学习的机票动态定价算法,其核心是不再预测各票价等级的需求量,而是将机票动态定价问题建模为离线强化学习问题。通过设计定价策略评估和策略更新的方式,从历史购票数据上学习具有最大期望收益的机票动态定价策略。同时设计了与现行定价策略和需求量预测方法的对比方法及评价指标。在两趟航班的多组定价结果表明:相比于现行机票销售策略,策略学习算法在座位收益上的提升率分别为30.94%和39.96%,且比基于需求量预测方法提升了6.04%和3.36%。     

一种航空公司品牌机票加座位捆绑销售系统的研究和实现

本文设计并实现了一种全新的品牌机票与座位捆绑销售的系统，使航空公司品牌机票、预付费座位产品的可销售场景更加丰富，为出行用户提供多样的附加服务产品选择机会。通过本系统为航空公司灵活销售定制的动态定价处理模式，航司会提升应对不同附加服务销售场景的产品价格调整能力。该系统提供的附加服务管理模块，支持配置不同的预付费座位产品基本信息和销售规则，还可设置品牌机票与预付费座位产品捆绑销售规则，最后通过调价规则的配置，对不同常旅客等级的出行用户展示个性化的价格和优惠幅度，从而吸引出行用户购买捆绑产品，最大限度保留忠诚用户，增加航司品牌机票和附加服务的收入。     

基于Stackelberg博弈的网络虚拟化资源分配方法

在网络虚拟化环境中,基于Stackelberg博弈模型提出了静态博弈算法下网络运营商对频谱的定价策略及静态和动态博弈算法下各虚拟运营商的频谱分配方法,并推导了各运营商收益最大时的纳什均衡点。该方法能同时满足网络运营商和虚拟网络运营商收益最大化的需求。仿真结果验证了算法的有效性及纳什均衡点的存在性。     

航线团队旅客收益管理双层规划模型

当团队人数大于航线上任一航班的剩余座位时,在散客预测的基础上研究在不挤掉散客的前提下如何进行团队拆分并使航线上的收益最大。文中提出利用双层规划模型对团队旅客进行拆团,使得航空公司在增加收益的同时又不会失去旅行代理商。最后,设计了双层迭代算法求解模型,并结合实例验证了模型和算法的有效性。     

分布式环境中基于市场机制的资源自适应调价策略

针对分布式环境中资源定价面临的资源使用率、价格、收益三者之间的冲突问题,提出一种基于市场机制的资源自适应调价策略。该策略在保障资源提供者收益前提下,通过资源价格自适应动态调整来平衡资源节点上的任务分配与资源提供者收益之间的冲突。理论分析证明了调价策略的有效性,并在此基础上设计了自适应调价算法。仿真实验采用真实分布式系统中资源节点信息作为实验节点的性能参数,在大规模网格任务中检验了自适应调价策略的性能表现。实验结果表明,基于市场机制的“自适应调价策略”在保障资源收益、均衡资源利用率的性能表现方面显著优于传统的定价策略。     

异构认知网络环境下的动态分级资源管理方法

针对异构认知网络中的资源管理问题,提出了基于认知的动态分级资源管理方法(DHRM)。根据不同时间尺度,引入小波神经网络、基于维纳过程的预测方法和增强学习算法获得业务分布变化、切换呼叫资源需求量以及用户喜好等信息,从而动态调配异构多网络各级可用资源。在资源合理分配基础上,根据各网络实时状态以及用户喜好,通过多属性决策算法动态地将业务流分配到最佳接入网络中。仿真结果表明,DHRM相对于网间静态资源管理方法系统容量提高了约20%。     

认知无线电网络中基于随机学习博弈的信道分配与功率控制

传统的认知无线电频谱分配算法往往忽略节点的传输功率对网络干扰的影响,且存在节点间交互成本高的问题.为此,通过量化传输功率等级,以最大化弹性用户收益为目标,构建联合频谱分配与功率控制非合作博弈模型,证明了该博弈为严格潜在博弈且收敛到纳什均衡点.进一步,将随机学习理论引入博弈模型,提出了基于随机学习的策略选择算法,并给出了该算法收敛到纯策略纳什均衡点的充分条件及严格证明.仿真结果表明,所提算法在少量信息交互前提下能获得较高的传输速率,并提升用户满意度.     

基于成本约束的虚拟网映射策略及竞争分析

为实现物理网提供商长期收益的最大化,单个虚拟网的映射成本和接入控制策略最为关键,但在之前的研究中,资源价格定义不能反映资源供求关系,不利于物理网资源的有效利用,且接入控制策略没有综合考虑成本和收益的关系.为此,首先基于凸二次规划松弛方法,设计以映射成本最小化为目标的单虚拟网映射方案求解的近似算法;然后,针对动态到达的单虚拟网构建请求,基于影子价格的物理网资源定价策略,用上述近似算法求出映射方案,并基于映射成本约束的虚拟网接入控制策略,完成竞争算法设计,并给出算法的竞争比分析.实验表明,所提方法能使物理网资源得到有效利用,进而提高虚拟网构建请求的接受率和物理网提供商的长期收益.     

无线mesh网络中可信协同信道资源分配策略

为了有效提升无线mesh网络信道资源的利用率和网络服务质量,提出基于可信协同的信道资源分配策略.针对节点自适应特点,引入博弈理论、建立节点的信誉机制以实现节点可信协同并优化信道分配结果.仿真实验分别对节点服务等级、网络收益结果作相应评价.实验结果发现节点服务等级对节点网络收益有直接影响,当协同服务等级达到3时,网络收益状况最佳,此时节点跳数与服务等级呈协同关系;对比经典协同算法,在相同网络拓扑环境下,可信协同信道资源分配策略分别是UACRR算法、DMP-MBA算法的1.04倍、1.069倍,明显占优.     

基于CBAM-CondenseNet的航班延误波及预测模型

针对航班延误衍生的航班延误波及问题,该文提出一种基于CBAM-CondenseNet的航班延误波及预测模型。首先,通过分析航班延误在航空网络内产生的延误波及现象,确定会受前序延误航班影响的航班链;其次,对选定的航班链数据进行清洗,将航班信息与机场信息进行数据融合;最后,提出改进的CBAM-CondenseNet算法对融合后的数据进行特征提取,构建Softmax分类器对首班离港航班延误波及的后续离港航班延误等级进行预测。该文提出的CBAM-CondenseNet算法融合了CondenseNet和CBAM的优势,采用通道和空间注意力机制来加强网络结构深层信息的传递。实验结果表明,算法改进后有效提升网络性能,预测准确率可达97.55%。     

Optimal Nonlinear Pricing for a Monopolistic Network Service Provider with Complete and Incomplete Information

In the communication network pricing literature, it is the linear pricing schemes that have been largely adopted as the means of controlling network usage or generating profits for network service providers. This paper extends the framework to nonlinear pricing and investigates optimal nonlinear pricing policy design for a monopolistic service provider. The problem is formulated as an incentive-design problem, and incentive (pricing) policies are obtained for a many-users regime, which enable the service provider to approach arbitrarily close to Pareto- optimal solutions. Under the assumption that the service provider knows the true user types, analytical and numerical results indicate a profit improvement exceeding 38% over linear pricing by the introduction of nonlinear pricing. We also consider the scenario where the service provider has incomplete information on user types. A comparative study of the results for complete information and incomplete information is carried out as well, with numerical results pointing to 25%-40% loss of profit by the service provider due to incompleteness of information on the user types.     

Enhancing Performance with a Learnable Strategy for Multiple Question Answering Modules

A question answering (QA) system can be built using multiple QA modules that can individually serve as a QA system in and of themselves. This paper proposes a learnable, strategy‐driven QA model that aims at enhancing both efficiency and effectiveness. A strategy is learned using a learning‐based classification algorithm that determines the sequence of QA modules to be invoked and decides when to stop invoking additional modules. The learned strategy invokes the most suitable QA module for a given question and attempts to verify the answer by consulting other modules until the level of confidence reaches a threshold. In our experiments, our strategy learning approach obtained improvement over a simple routing approach by 10.5% in effectiveness and 27.2% in efficiency.     

Forecasting Demand for Telecommunications Products from Cross-Sectional Data

A crucial planning problem for many telecommunications companies is how best to forecast changes in demand for their products over the next several years. This paper presents a new approach to demand forecasting that performs well compared to even sophisticated time series methods but that requires far less data. It is based on the following simple idea: Divide units of analysis (census blocks, customers, etc.) into groups with relatively homogeneous behaviors, forecast the behavior of each group (which can be done easily, by construction), and sum over all groups to obtain aggregate forecasts. Identifying groupings of customers to minimize forecast errors is a difficult combinatorial challenge that we address via the data-mining technique of classification tree analysis. Product acquisition rates are modeled as transition rates in a multi-state simulation model. The dynamic simulation model is used to integrate the transition rate and covariate information and to predict the resulting changes in product demands over time.     

Predicting cell phone adoption metrics using machine learning and satellite imagery

Approximately half of the global population does not have access to the internet, even though digital connectivity can reduce poverty by revolutionizing economic development opportunities. Due to a lack of data, Mobile Network Operators and governments struggle to effectively determine if infrastructure investments are viable, especially in greenfield areas where demand is unknown. This leads to a lack of investment in network infrastructure, resulting in a phenomenon commonly referred to as the ‘digital divide’. In this paper we present a machine learning method that uses publicly available satellite imagery to predict telecoms demand metrics, including cell phone adoption and spending on mobile services, and apply the method to Malawi and Ethiopia. Our predictive machine learning approach consistently outperforms baseline models which use population density or nightlight luminosity, with an improvement in data variance prediction of at least 40%. The method is a starting point for developing more sophisticated predictive models of infrastructure demand using machine learning and publicly available satellite imagery. The evidence produced can help to better inform infrastructure investment and policy decisions.     

Pricing network resources for adaptive applications

The Differentiated Services framework (DiffServ) has been proposed to provide multiple Quality of Service (QoS) classes over IP networks. A network supporting multiple classes of service also requires a differentiated pricing structure. In this work, we propose a pricing algorithm in a DiffServ environment based on the cost of providing different levels of services, and on long-term average user resource demand of a service class. We integrate the proposed service-dependent pricing scheme with a dynamic pricing and service negotiation environment by considering a dynamic and congestion-sensitive pricing component. Pricing network services dynamically based on the level of service, usage, and congestion allows a more competitive price to be offered, allows the network to be used more efficiently, and provides a natural and equitable incentive for applications to adapt their service requests according to network conditions. We also develop the demand behavior of adaptive users based on a physically reasonable user utility function. Simulation results show that a congestion-sensitive pricing policy coupled with user rate adaptation is able to control congestion and allows a service class to meet its performance assurances under large or bursty offered loads, even without explicit admission control. Users are able to maintain a stable expenditure, and allowing users to migrate between service classes in response to price increases further stabilizes the individual service prices. When admission control is enforced, congestion-sensitive pricing still provides an advantage in terms of a much lower connection blocking rate at high loads.     

Is the Price Right?

We conclude that contribution analysis is adaptable and useful for pricing new products. Pricing strategy should be devised during the product development stage, and both price and cost forecasting are important for this purpose. Further, to avoid the often repeated mistake of basing a new product's price on unrealistic introductory production and marketing costs, the experience curve should be used for realistic cost estimates. It is important to know: ? competitive prices (actual and expected), cost structures, and capacity utilization ? buyers' reactions to price ? the firm's cost structure and corporate objectives. Unless the pricing process is firmly imbedded within a dynamic, long-run marketing strategy, it is likely that a pricing policy unrelated to the marketing strategy will evolve. Correct pricing decisions are vital as today's pricing environment increases pressure for better, faster, and more frequent pricing decisions. Better research concerning the competitive market and customer responses to prices and price changes is urgently needed. Within the industrial firm, a price-research budget is needed. We must avoid a common occurrence, i.e., worrying about the market response to a price decision after the decision has been made.     

Pricing in multiservice loss networks: static pricing, asymptoticoptimality and demand substitution effects

We consider a communication network with fixed routing that can accommodate multiple service classes, differing in bandwidth requirements, demand pattern, call duration and routing. The network charges a fee per call which can depend on the current congestion level and which affects user's demand. Building on the single-node results of I.Ch. Paschalidis and J.N. Tsitsiklis (see IEEE/ACM Trans. Networking, vol.8, p.171-84, 2000), we consider both problems of revenue and of welfare maximization, and show that static pricing is asymptotically optimal in a regime of many, relatively small, users. In particular, the performance of an optimal (dynamic) pricing strategy is closely matched by a suitably chosen class-dependent static price, which does not depend on instantaneous congestion. This result holds even when we incorporate demand substitution effects into the demand model. More specifically, we model the situation where price increases for a class of service might lead users to use another class as an imperfect substitute. For both revenue and welfare maximization objectives we characterize the structure of the asymptotically optimal static prices, expressing them as a function of a parsimonious number of parameters. We employ a simulation-based approach to tune those parameters and to compute efficiently an effective policy away from the limiting regime. Our approach can handle large, realistic, instances of the problem     

基于深度强化学习的云边协同计算迁移研究

基于单一边缘节点计算、存储资源的有限性及大数据场景对高效计算服务的需求,本文提出了一种基于深度强化学习的云边协同计算迁移机制.具体地,基于计算资源、带宽和迁移决策的综合性考量,构建了一个最小化所有用户任务执行延迟与能耗权重和的优化问题.基于该优化问题提出了一个异步云边协同的深度强化学习算法,该算法充分利用了云边双方的计算能力,可有效满足大数据场景对高效计算服务的需求;同时,面向边缘云中边缘节点所处环境的多样及动态变化性,该算法能自适应地调整迁移策略以实现系统总成本的最小化.最后,大量的仿真结果表明本文所提出的算法具有收敛速度快、鲁棒性高等特点,并能够以最低的计算成本获得近似贪心算法的最优迁移决策.     

一种新颖的多agent强化学习方法

提出了一种综合了模块化结构、利益分配学习以及对手建模技术的多agent强化学习方法,利用模块化学习结构来克服状态空间的维数灾问题,将Q-学习与利益分配学习相结合以加快学习速度,采用基于观察的对手建模来预测其他agent的动作分布.追捕问题的仿真结果验证了所提方法的有效性.