自适应遗传算子优化BP神经网络的气温预测

Adaptive Genetic Operator Optimizes Temperature Prediction of BP Neural Network

摘要: 气温变化是由多种非线性因素引起的，传统的数值气温预测模式由于其繁琐的步骤和较低的预测精度已无法满足人们的需求。鉴于BP神经网络对非线性的拟合有很强的映射能力，因此考虑用BP神经网络对气温进行预测，而滨海地区影响气温变化的因素更加复杂，因此在BP神经网络中加入自适应遗传算子对其阈值、权值以及收敛速度进行优化以此提高气温预测的精度。为进一步提高模型预测精度以及数据处理效率，文章初始对输入数据做了降维筛选主成分，归一化等一系列优化处理，在此基础上建立了BP神经网络以及加入自适应算子的BP神经网络两种气温预测模型。之后用MATLAB软件对山东省日照市的气温相关数据进行仿真，对比BP和GA-BP模型的预测结果，改进后模型预测的准确度更高，以此证明了优化算法的有效性。

Abstract: The temperature change is caused by many nonlinear factors, and the traditional numerical temperature prediction model cannot meet people’s demands due to its complicated steps and low prediction accuracy. Given the BP neural network in nonlinear fitting has a strong ability to map, so consider using the BP neural network to forecast the temperature, and the influencing factors of temperature change are more complicated in the coastal area, thus to join in the BP neural network adaptive genetic operators, weights and threshold value to its convergence speed is optimized to improve the accuracy of the temperature prediction. To further improve the prediction accuracy and data processing efficiency of the model, a series of optimization processes, such as normalization of input data, dimension reduction, and principal component screening, was initially carried out in this paper. On this basis, two temperature prediction models, BP neural network, and BP neural network with the adaptive operator were established. Then MATLAB was used to simulate the air temperature-related data of Rizhao city and the prediction results of BP and GA-BP models were compared. The improved model had a higher prediction accuracy, thus proving the effectiveness of the optimization algorithm.