基于粒子群优化神经网络的GDP预测

发布时间：2019-03-09 09:02

【摘要】：现如今世界经济发展的一个主要趋势便是经济全球化。经济全球化不仅有助于当下中国市场经济体制的加速组建,而且对于国内一些企业的成长及发展起到了推动作用。当然经济全球化也会给我们带来很多挑战。准确把握经济发展的短期走势和长期趋势,为我国国民经济的快速发展出谋献策自然变得尤为重要。由此对于那些未来事物发展情况的预测便越来越受到人们的重视。为了使经济决策的错误降到最低,就要保证未来事物发展情况的预测高度准确,以此来为经济决策提供可靠依据。因此准确预测国内生产总值(Gross Domestic Product,GDP)对政府进行经济结构调整及宏观经济提供决策支撑意义重大。本文首先介绍了粒子群(Paeticle Swarm Optimization,PSO)算法,它属于群智能优化算法中的一种。该算法源于对鸟群觅食过程中的迁徙和聚集的模拟,它的优点在于收敛速度快,同时只需要调整一些少量参数,简单且易于实现。标准粒子群算法能够用非常快的速率找到局部最好解,但它也有局限性,在寻找全局最优解时往往有些逊色,易陷入局部最优的状态。由此针对标准粒子群算法的一些局限性进行改进,提出了加入加速常数1c和2c,然后对比标准粒子群算法和改进的粒子群算法分别对Ackley、Rastrigin、Rosenbrock和Schaffer函数进行寻优测试。实验结果可以看出,改进后的粒子群算法寻找全局最优解的能力得到明显改善,收敛速度显著提高,适合应用于优化问题的求解。其次针对静态的前馈型BP神经网络在时序序列预测问题中精度低、不具有记忆性等问题,提出了采用动态的、有局部记忆功能的Elman神经网络建立GDP预测模型。同时用改进的PSO对ElmanNN的权值和阈值进行优化,以此来提高ElmanNN的训练效率。最后以安徽省为实验对象,详细地分析了GDP的影响因子,在此基础上建立了WCPSO优化ElmanNN的GDP预测模型。将WCPSO优化ElmanNN模型、ElmanNN模型和BPNN模型的预测结果分别与GDP实际数据进行对比。结果表明,所提出的WCPSO-ElmanNN模型的预测精度相比其它两个模型预测精度更高。
[Abstract]:Today, one of the main trends of world economic development is economic globalization. Economic globalization not only helps to accelerate the establishment of the present market economy system in China, but also promotes the growth and development of some domestic enterprises. Of course, economic globalization will also bring us a lot of challenges. It is of great importance to accurately grasp the short-term and long-term trends of economic development and to offer suggestions for the rapid development of our national economy. As a result, people pay more and more attention to the prediction of the development of those things in the future. In order to minimize the errors in economic decision-making, it is necessary to ensure that the prediction of future development is highly accurate, so as to provide a reliable basis for economic decision-making. Therefore, it is of great significance for the government to adjust its economic structure and provide decision-making support for macro-economy by accurately forecasting the gross domestic product (Gross Domestic Product,GDP). In this paper, we first introduce particle swarm optimization (Paeticle Swarm Optimization,PSO) algorithm, which belongs to one of the swarm intelligent optimization algorithms. The algorithm is derived from the simulation of migration and aggregation of birds during foraging. It has the advantage of fast convergence, and only a few parameters need to be adjusted, so it is simple and easy to realize. The standard particle swarm optimization (PSO) algorithm can find the local best solution at a very fast rate, but it also has its limitations. When searching for the global optimal solution, it is often somewhat inferior and easy to fall into the local optimal state. In view of some limitations of the standard particle swarm optimization algorithm, an acceleration constant of 1c and 2C is proposed, and then the Ackley,Rastrigin,Rosenbrock and Schaffer functions are tested by comparing the standard particle swarm optimization algorithm and the improved particle swarm optimization algorithm. The experimental results show that the ability of the improved PSO algorithm to find the global optimal solution is obviously improved, and the convergence rate is significantly increased, which is suitable for the solution of the optimization problem. Secondly, aiming at the problems such as low precision and no memory of the static feedforward BP neural network, a dynamic, local memory based Elman neural network is proposed to build the GDP prediction model. At the same time, the improved PSO is used to optimize the weight and threshold of ElmanNN in order to improve the training efficiency of ElmanNN. Finally, taking Anhui Province as the experimental object, the influence factors of GDP are analyzed in detail. On the basis of this, the GDP prediction model of WCPSO optimized ElmanNN is established. The prediction results of WCPSO optimized ElmanNN model, ElmanNN model and BPNN model are compared with the actual data of GDP respectively. The results show that the prediction accuracy of the proposed WCPSO-ElmanNN model is higher than that of the other two models.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F124;TP18

【参考文献】