基于Pearson相关系数准则的用电量组合预测研究

Study on Combination Forecast of Electricity Consumption Based on Pearson Correlation Coefficient Criterion

全社会用电量的预测对能源建设规划、电网规划决策等有重要参考价值。基于自回归AR、BP神经网络和灰色GM(1,1)预测模型,构建了以Pearson相关系数为评价准则的加权组合预测模型。通过江苏省用电量2009-2013年预测值与实际值比较,自回归AR模型、BP神经网络模型和灰色GM(1,1)模型预测误差绝对值和分别为4558.5亿千瓦时、3086.5亿千瓦时、和1305.7亿千瓦时,而本组合预测模型预测误差绝对值和为1235.0亿千瓦时,改进预测效果明显。最后依据该组合预测模型对未来江苏省全社会用电量进行了预测。

The forecast of social power consumption has the crucial reference value for energy construction planning and power network plan decision. By using AR model, BP neural network model and GM （1, 1 ） grey forecast model, the combination forecast model based on the evaluation criteria of Pearson correlation coefficient is constructed in this article. The comparison between the sums of error absolute value from three models, 455.85 ×10^9kWh, 308.55 ×10^9kWh, and 130.57 ×10^9kWh, and the sum from the combination forecast model, 123.5×10^9kWh, proves that the combination forecast model significantly improves the result. Meanwhile, the future social power consumption in Jiangsu province is forecast.