基于多层聚类和改进BP神经网络的短期负荷预测

Short term load forecasting based on multilayer clustering and improved BP neural network

随着电网规模的扩大,数据量的急剧增加,漏采、误采等不良数据及数据冗余会对短期负荷预测产生负面影响,造成预测准确度下降,预算结果无法正常使用.针对这些现象,提出一种基于多层聚类和改进BP神经网络的负荷预测模型.该模型基于多层聚类对原始数据进行预处理,选取形成与待预测数据相似的样本数据集,建立基于改进BP神经网络的预测模型.多层聚类模型减小了输入改进BP神经网络的数据量,避免了不良数据对预测模型造成的影响,预测模型更贴近待预测数据特点;改进BP神经网络避免了在训练过程中陷入局部最小解.预测结果表明:相比模糊C均值聚类方法,多层聚类与改进BP神经网络的负荷预测方法提高了预测精度与预测速度.

With the expansion of the scale of the power grid and the giant leap of the amount of data, the bad data and data redundancy, such as missing data, false collecting data and so on, will have a negative impact on short-term load forecasting, which results in the decrease of forecasting accuracy and the unavailability of budget results. In light of these problems, a load forecasting model based on multilayer clustering and improved BP neural network is proposed. The model preprocesses the original data based on multilayer clustering, then selects to form a sample data set similar to the predicted data, and establishes a prediction model based on improved BP neural network. The multilayer clustering model reduces the amount of data input in improved BP neural network, and avoids the effects of bad data on the prediction model;and the prediction model is closer to the characteristics of the data to be predicted. The improved BP neural network avoids getting into the local minimum solution in the process of training. The prediction results show that compared with the fuzzy C-means method, the load forecasting method based on multilayer clustering and improved BP neural network improves the forecasting accuracy and speed.