BP神经网络和数学模型在电量预测中的应用

Application of BP Neural Network and Mathematical Model in Electricity Quantity Prediction

阐述BP神经网络与X12、ARIMA数学结合算法,提高不同维度售电量预报精度。探讨影响售电量预报参数,对历史售电量数据分解为趋势分量、季节周期分量和随机分量,且各分量随时间的变化规律不一样,针对各分量序列不同特征分别分析,其结果选用乘法模型进行叠加处理。在保证计算精度和运算时间满足实际应用的前提下,合理计算BP网络隐层数量,以及输入层参数。通过大量实际数据与预报数据的误差分析,确定传统数学模型与BP神经网络相结合的方法预报精度误差小于5%。

This paper describes the combination of BP neural network with X12 and ARIMA mathematical algorithms to improve the accuracy of electricity sales forecasting in different dimensions.It explores the parameters that affect electricity sales forecasting,decomposing historical electricity sales data into trend components,seasonal cycle components,and random components,and the variation patterns of each component over time are different.Different characteristics of each component sequence are analyzed separately,and the results are stacked using a multiplication model.Reasonably calculate the number of hidden layers and input layer parameters of the BP network while ensuring that the calculation accuracy and computation time meet practical applications.Through error analysis of a large amount of actual data and forecast data,it is determined that the prediction accuracy error of the method combining traditional mathematical models with BP neural networks is less than 5%.