文章主要探讨了一种将卷积神经网络、长短时记忆网络以及注意力机制相结合的方法在新能源发电量预测中的应用及其有效性。随着新能源发电量受外部环境影响而表现出较大波动性和复杂性,传统预测模型难以全面捕捉其中的复杂模式和长期依...文章主要探讨了一种将卷积神经网络、长短时记忆网络以及注意力机制相结合的方法在新能源发电量预测中的应用及其有效性。随着新能源发电量受外部环境影响而表现出较大波动性和复杂性,传统预测模型难以全面捕捉其中的复杂模式和长期依赖性。因此,文章提出了一个集成多种深度学习方法的组合模型(CNN-LSTM-Attention模型),该模型首先通过卷积层提取数据中的局部特征,随后利用长短时记忆网络建模时间序列的长期依赖关系,最后通过注意力机制增强对重要信息的关注度。实验选取了2024年1月1日至6月30日期间的甘肃省风力发电和水力发电数据,进行了特征提取和模型训练。结果显示,相较于单独使用卷积神经网络或长短时记忆网络,结合了注意力机制的组合模型在多个评估指标上均表现出更高的预测准确性和更好的拟合效果。此外,还引入了孤立森林算法对预测误差进行异常值检测,并结合风险等级进行了详细的分类分析,进一步验证了该模型在实际应用中的有效性。本研究为新能源发电系统的预测和管理提供了一种新思路和方法,有助于提高系统运行的可靠性和安全性。This paper mainly discusses the application and effectiveness of a method combining convolutional neural networks, long short-term memory networks, and attention mechanisms in new energy power generation predictions. As new energy generation is affected by the external environment and shows great volatility and complexity, it is difficult to fully capture complex patterns and long-term dependencies using traditional forecasting models. Therefore, this paper proposes a combined model integrating multiple deep learning methods (CNN-LSTM-Attention model), which firstly extracts local features from the data through convolutional layers, then models the long-term dependencies of time series by long short-term memory networks and finally enhances the focus on important information through attention mechanisms. The experiment selected the wind power and hydroelectric power generation data of Gansu Province from January 1 to June 30, 2024, and carried out feature extraction and model training. The results show that compared with convolutional neural networks or long short-term memory networks alone, the combined model with attention mechanism has higher prediction accuracy and better fitting effect on multiple evaluation indicators. In addition, this paper introduces the isolation forest algorithm for outlier detection of the prediction errors and carries out a detailed classification analysis combined with the risk level, further verifying the effectiveness of the model in practical application. The research in this paper provides a new idea and method for the prediction and management of a new energy power generation system, which is helpful in improving the reliability and safety of the whole system operation.展开更多
Improving the accuracy of solar power forecasting is crucial to ensure grid stability,optimize solar power plant operations,and enhance grid dispatch efficiency.Although hybrid neural network models can effectively ad...Improving the accuracy of solar power forecasting is crucial to ensure grid stability,optimize solar power plant operations,and enhance grid dispatch efficiency.Although hybrid neural network models can effectively address the complexities of environmental data and power prediction uncertainties,challenges such as labor-intensive parameter adjustments and complex optimization processes persist.Thus,this study proposed a novel approach for solar power prediction using a hybrid model(CNN-LSTM-attention)that combines a convolutional neural network(CNN),long short-term memory(LSTM),and attention mechanisms.The model incorporates Bayesian optimization to refine the parameters and enhance the prediction accuracy.To prepare high-quality training data,the solar power data were first preprocessed,including feature selection,data cleaning,imputation,and smoothing.The processed data were then used to train a hybrid model based on the CNN-LSTM-attention architecture,followed by hyperparameter optimization employing Bayesian methods.The experimental results indicated that within acceptable model training times,the CNN-LSTM-attention model outperformed the LSTM,GRU,CNN-LSTM,CNN-LSTM with autoencoders,and parallel CNN-LSTM attention models.Furthermore,following Bayesian optimization,the optimized model demonstrated significantly reduced prediction errors during periods of data volatility compared to the original model,as evidenced by MRE evaluations.This highlights the clear advantage of the optimized model in forecasting fluctuating data.展开更多
为了更迅速、准确地识别出高压电缆的局部放电故障,本文提出一种基于卷积长短期注意力(convolutional long short-term attention,CNN-LSTM-Attention)神经网络的高压电缆局部放电预测方法.首先,对高压电缆局部放电信号进行实时监测,并...为了更迅速、准确地识别出高压电缆的局部放电故障,本文提出一种基于卷积长短期注意力(convolutional long short-term attention,CNN-LSTM-Attention)神经网络的高压电缆局部放电预测方法.首先,对高压电缆局部放电信号进行实时监测,并用小波分析将其离散,把长信号切分成多段信号且提取每段信号的统计特征量;其次,根据特征量构建神经网络分类模型,其由能够提取轮廓特征的卷积层、提取信号时序特征的长短期记忆层以及具有时序重要部分捕捉能力的注意力层构成;最后,通过实际数据进行仿真.结果表明:所提方法能准确识别较高采样率的异常放电信号,且相比传统神经网络,CNN-LSTM-Attention神经网络的故障识别准确率有明显提高,其马修斯相关系数(Matthews correlation coefficient,MCC)为0.87103.展开更多
文摘文章主要探讨了一种将卷积神经网络、长短时记忆网络以及注意力机制相结合的方法在新能源发电量预测中的应用及其有效性。随着新能源发电量受外部环境影响而表现出较大波动性和复杂性,传统预测模型难以全面捕捉其中的复杂模式和长期依赖性。因此,文章提出了一个集成多种深度学习方法的组合模型(CNN-LSTM-Attention模型),该模型首先通过卷积层提取数据中的局部特征,随后利用长短时记忆网络建模时间序列的长期依赖关系,最后通过注意力机制增强对重要信息的关注度。实验选取了2024年1月1日至6月30日期间的甘肃省风力发电和水力发电数据,进行了特征提取和模型训练。结果显示,相较于单独使用卷积神经网络或长短时记忆网络,结合了注意力机制的组合模型在多个评估指标上均表现出更高的预测准确性和更好的拟合效果。此外,还引入了孤立森林算法对预测误差进行异常值检测,并结合风险等级进行了详细的分类分析,进一步验证了该模型在实际应用中的有效性。本研究为新能源发电系统的预测和管理提供了一种新思路和方法,有助于提高系统运行的可靠性和安全性。This paper mainly discusses the application and effectiveness of a method combining convolutional neural networks, long short-term memory networks, and attention mechanisms in new energy power generation predictions. As new energy generation is affected by the external environment and shows great volatility and complexity, it is difficult to fully capture complex patterns and long-term dependencies using traditional forecasting models. Therefore, this paper proposes a combined model integrating multiple deep learning methods (CNN-LSTM-Attention model), which firstly extracts local features from the data through convolutional layers, then models the long-term dependencies of time series by long short-term memory networks and finally enhances the focus on important information through attention mechanisms. The experiment selected the wind power and hydroelectric power generation data of Gansu Province from January 1 to June 30, 2024, and carried out feature extraction and model training. The results show that compared with convolutional neural networks or long short-term memory networks alone, the combined model with attention mechanism has higher prediction accuracy and better fitting effect on multiple evaluation indicators. In addition, this paper introduces the isolation forest algorithm for outlier detection of the prediction errors and carries out a detailed classification analysis combined with the risk level, further verifying the effectiveness of the model in practical application. The research in this paper provides a new idea and method for the prediction and management of a new energy power generation system, which is helpful in improving the reliability and safety of the whole system operation.
基金supported by the State Grid Science&Technology Project(5400-202224153A-1-1-ZN).
文摘Improving the accuracy of solar power forecasting is crucial to ensure grid stability,optimize solar power plant operations,and enhance grid dispatch efficiency.Although hybrid neural network models can effectively address the complexities of environmental data and power prediction uncertainties,challenges such as labor-intensive parameter adjustments and complex optimization processes persist.Thus,this study proposed a novel approach for solar power prediction using a hybrid model(CNN-LSTM-attention)that combines a convolutional neural network(CNN),long short-term memory(LSTM),and attention mechanisms.The model incorporates Bayesian optimization to refine the parameters and enhance the prediction accuracy.To prepare high-quality training data,the solar power data were first preprocessed,including feature selection,data cleaning,imputation,and smoothing.The processed data were then used to train a hybrid model based on the CNN-LSTM-attention architecture,followed by hyperparameter optimization employing Bayesian methods.The experimental results indicated that within acceptable model training times,the CNN-LSTM-attention model outperformed the LSTM,GRU,CNN-LSTM,CNN-LSTM with autoencoders,and parallel CNN-LSTM attention models.Furthermore,following Bayesian optimization,the optimized model demonstrated significantly reduced prediction errors during periods of data volatility compared to the original model,as evidenced by MRE evaluations.This highlights the clear advantage of the optimized model in forecasting fluctuating data.
