基于粒子群优化极限学习机数据预测模型研究

A New Model for Data Prediction Based on Particle Swarm Optimization Extreme Learning Machine

极限学习机(Extreme Learning Machine,ELM)是一种新型的前馈神经网络,该网络由广义逆直接求出输出层权重,使得其具有误差小、速度快的优点.但针对具体问题,ELM不能自动寻找到最佳的网络结构,从而造成该算法模型针对复杂、无规律性的数据精度及稳定性较差.为了提高极限学习机的泛化能力和预测精度,提出利用粒子群优化极限学习机算法对不同数据进行预测.使用粒子群算法(particle swarm optimization,PSO)选择最优的隐含层偏差和输入权值矩阵,计算出输出权值矩阵,从而提高ELM的精度及稳定性.并通过PSO-ELM和ELM分别对复杂程度不同的汽油辛烷值和交通流量数据进行算法预测比较发现,PSO-ELM优化算法对无规律性、复杂程度高的数据可以获得更高的精度,提高了数据预测的拟合能力.实验结果表明,PSO-ELM对于非线性、无规律性等复杂特性的数据预测具有一定的可行性和有效性.

The Extreme Learning Machine (ELM) is a new type of feedforward neural nehvork. The network directly derives the weight of the output layer from the generalized inverse, which makes it have the advantages of small error and fast speed. However, for the specific problem, ELM cant automatically find the best nehvork structure, which makes the algo-rithm model have poor precision and stability for complex and irregular data. In order to improve the generalization ability and prediction accuracy of the extreme learning machine, this paper proposes to use the particle swann optimization extreme learn-ing machine algorithm to predict different data. In this paper, particle swarni optimization ( PSO) is used to select the optimal hidden layer bias and input weight matrix to calculate the output weight matrix, which improves the accuracy and stability of ELM. Tlirough PSO-ELM and ELM, respectively, the algorithm predictions of gasoline octane number and traffic flow data with different complexity are compared. It is found that the PSO-ELM optimization algorithm can obtain higher precision and improve the data with irregularity and high complexity. Tlie experimental results show that PSO-ELM has certain feasibility and effectiveness for the prediction of gasoline octane number with complex characteristics such as nonlinearity and irregularity.