基于改进GRU的电力大数据分析

针对电力大数据存在冲击数据、无效数据等异常值导致真实规律难以挖掘的问题，提出一种基于改进GRU的电力大数据分析模型。该模型首先分析了异常值导致数据规律失真的情况，提出利用自适应阈值的小波滤波进行数据清洗；其次以单数据周期为分段点对数据进行分段，以各数据段同一时刻的记忆求和均值为标准记忆；最后，根据数据段的质量改进GRU记忆能力，即保留质量好的数据段记忆、删除质量差的数据段记忆。为了验证模型的性能，选择光伏发电数据进行实验，结果表明：本模型在数据质量较高时的预测准确率比ARIMA、LSTM和标准GRU分别提高了61%、30%和25%，数据质量较差时的预测准确率分别提高了76%、16%和11%。

Power Big Data Analysis Based on Improved GRU

Aiming at the problem that abnormal values of mutation, invalidation and so on in electric power big data analysis results in the difficulty to mine the true law, this paper proposes a big data analysis model based on improved GRU. The model first analyzes the effects of abnormal values leading to data redundancy, errors, etc., and use adaptive threshold wavelet filtering to eliminate the above effects. Then, the data is segmented with a period into several data segments. The memory summation at the same time points, and the average value obtained by the summation result is used as the standard memory. Finally, the GRU memory capacity is improved according to the quality of the data segment, that is, the memory of good quality data segment is retained, and the data segment memory of poor quality is deleted. In order to verify the performance of the model, experiments were performed on photovoltaic power generation data. The results show that the prediction accuracy of this model when the data quality is high is 61%, 30% and 25% higher than that of ARIMA, LSTM and standard GRU respectively. The data quality The prediction accuracy rates of poorer forecasts increased by 76%, 16%, and 11%, respectively.