【作者】 郭崇慧；

【导师】 唐焕文；

【作者基本信息】 大连理工大学 ， 管理科学与工程， 2002， 博士

【摘要】 围绕地区经济预测和中长期发展规划中的若干定量模型、算法以及应用等方面(其中定量方法主要包括投入产出分析、人工神经网络和系统优化方法)，本文进行了一些研究和讨论。 主要研究成果如下： 1．根据信息论中的最小叉熵原理，按照已知信息的多少，建立了几个可用于修订直接消耗系数的熵优化模型，分析了最小叉熵优化模型与NAIVE方法、最大熵优化模型、RAS方法以及改进的RAS方法的关系，并给出了求解优化模型的对偶算法。计算实例的结果表明，所建立的模型和给出的算法是有效的、可行的。 2．深入研究了动态投入产出系统的渐近稳定性和均衡增长解问题。利用矩阵特征值理论和广义系统理论，在相对弱的条件下(不需要直接消耗系数矩阵不可分解)，证明了动态投入产出系统不是渐近稳定的。并在相同条件下，证明了封闭动态投入产出模型存在具有经济意义的均衡增长解。 3．对动态投入产出模型的解及其灵敏度问题进行了分析。给出了基年的总产出向量和各期的最终净需求向量发生变动时，对计划期内国民经济各部门总产出可能产生的影响的计算公式，利用该公式可以方便地计算初始条件和外生变量变动时对国民经济各部门总产出的传递效应，从而为制定经济政策和计划提供科学依据。 4．将人工神经网络技术应用于地区经济预测领域，设计了一个“基于人工神经网络的地区经济预测系统”，并利用HS-FR共轭梯度算法对网络的学习算法进行了改进。该预测系统已用于某地区中长期经济预测，预测结果比较符合实际，已被当地政府计划部门采用，为制定“十五规划”提供了依据。 5．以地区经济预测和规划为实际背景，提出了一个地区经济智能预测系统框架，以人工神经网络方法、动态投入产出技术和系统优化方法为核心，结合 大连理工大学博士学位论文 实际建立了地区经济智能预测模型体系，分析了综合集成技术在构建地区经 济智能预测系统中的应用。 6．给出了一个混合的HSFR共轭梯度算法，在无充分下降条件下，得到了关于 HSFR算法的两个收敛性定理。讨论了一类共轭梯度算法的收敛性，推广了 1992年 Gilbert和 Nocedal的收敛性结果。 7．讨论了进化策略的收敛性。对于有界闭集上的连续函数，利用作者给出的一 种简洁的方法，证明了一类进化策略以概率为1收敛（几乎处处收敛）于优 化问题的全局极小点，为进化算法的收敛性分析提供了一种新的模式。针对 实值连续函数优化问题，提出了一种改进的进化规划算法并给出了收敛性定 理，数值实验表明，所提出的算法是可行的、有效的。 8．针对某省的公路交通的实际情况，建立了一个干线公路网等级结构优化的目 标规划模型，并给出了算法，计算结果已经应用于某省“十五”时期公路网 的实际规划当中。更多还原

【Abstract】 This dissertation is devoted to some quantitative models, algorithms and applications for regional medium- and long-term development planning. The quantitative methods concern with input-output analysis, artificial neural network and systems optimization methods. The main work of the paper can be summarized as follows:1. Entropy optimization models and algorithms for updating I-O coefficients are considered. Based on the minimum cross-entropy principle of information theory, several entropy optimization models for updating I-O coefficients are established according to prior information. And the relationships between the minimum cross-entropy optimization model and NAIVE, the maximum entropy optimization model, RAS and modified RAS are analyzed, respectively. A dual algorithm for optimization model with an example is presented. Numerical results illustrate that entropy optimization updating method is feasible and effective.2. The asymptotic stability and balanced growth solutions of the dynamic input-output system are studied. Under some natural weak assumptions that do not require the technological coefficients matrix is indecomposable, the fact that the dynamic input-output system is not asymptotically stable and the closed dynamic input-output model exists a balanced growth solution is proved.3. A sensitivity analysis for solutions to dynamic input-output model is performed. The formulations, which compute the changes of national economy sectors’ total output during the planning periods when the base year’s total output vector and each year’s final demand vector vary, are given. Due to the changes of the initial conditions and the exogenous variables, the transfer effects of national economy sectors’ total output are revealed.4. Based on artificial neural network technique, a regional economic forecasting system, which has been applied to practical regional medium- and long-term economic forecasting in certain city, is designed. A mixed HS-FR conjugate gradient algorithm is applied to the regional economic forecasting system to train the neural networks effectively. The proposed forecasting system has been applied to forecast main economic indicator of certain city in "the Tenth Five-Year" period, and the forecasting results are adopted by the regional government plan agency to formulate "the Tenth Five-Year" planning.5. Directed at regional medium- and long-term economic forecasting and planning, a conceptual framework of regional economic intelligent forecasting system based on the previous studies is presented, and the application of meta-synthesis in building the forecasting system is analyzed. A regional economic intelligent forecasting models system, in which the core is input-output models, artificial neural networks methods, and optimization techniques, is established.6. A mixed HS-FR conjugate gradient algorithm is proposed. Two convergence theorems without the sufficient descent condition for the mixed HS-FR algorithm are given. The convergence theorem of a class of conjugate gradient algorithms is proven, which extend the main convergence theorem in Gilbert and Noceda (1992).7. For the purpose of analyzing its asymptotic convergence properties the evolution strategy procedures for real-valued function optimization are described. Two convergence theorems, which show that under suitable conditions the evolution strategy asymptotically converges to a global minimum point with probability one, are given. An improved evolutionary programming algorithm for real-valued function optimization is proposed. Numerical results illustrate that the proposed algorithm is feasible and effective.8. According to the demands of making the medium- and long-term highway networks planning in certain province, a goal-programming model for arterial highway network grade structure optimization is established, and an algorithm with example is given.更多还原