Research On Chaotic Time Series Prediction Of Power System Based On Deep Learning

The stable operation of the power system is an important guarantee for national economic development and people’s lives.One of the important factors of power system instability is the chaotic oscillation of the system under certain parameters and working conditions.In a chaotic power system,the output voltage,frequency,and speed of the generator will oscillate randomly,which may lead to grid collapse.It is of great significance to predict the chaotic phenomenon of the power system,take protective measures in advance,and maintain the stable operation of the power system.On other hand,chaotic time series widely exists in complex nonlinear systems such as finance,climate,energy,transportation,etc.Due to the initial value sensitivity,inherent randomness and non-periodicity of chaotic systems,chaotic time series is difficult to predict and analyze.With the development of deep learning,chaos prediction based on deep learning is an effective method.In this paper,the research on the prediction algorithm and model of power system chaotic time series based on deep learning is carried out.The specific research work is as follows:（1）In this paper,a deep learning algorithm is proposed to predict the chaotic behavior of the power system using deep long short-term memory（DLSTM）networks,which have two forms,deep long short-term memory with static scenario（DLSTM-s）and deep long-term memory with dynamic scenario（DLSTM-d）.Genetic Algorithm is used to optimize the hyperparameters of networks.Then,taking the interconnected power systems as an example,the effectiveness of the proposed DLSTM is verified by numerical simulation.Finally,the experimental results of DLSTM are compared with those of the echo state network,multi-recurrent neural network,deep gated recurrent unit,long short-term memory The results show that a trained DLSTM network can predict chaotic behavior of the power systems only based on time series data of one state variable.It is also found that the DLSTM-s network proposed in this paper is superior to other network models.（2）In this paper,a deep learning algorithm is proposed to predict the chaotic behavior of the power system using deep long-short term memory and genetic attention mechanism（DLSTM-GA）.By passing the shared parameters,adding attention mechanism to optimize DLSTM model based on genetic algorithm,one can find potential characteristics in time sequence and avoid the local optimization.This method is inspired by evolutionary computation method of optimization method and is a good way to learn the parameters in the attention layer.The results show that a trained DLSTM-GA network has higher prediction accuracy and long-term prediction ability than other reference models in predicting chaotic power system time series.（3）A time spectrum neural network based on optimization is proposed for chaos prediction of power system.Firstly,the potential correlation layer is used to mine the potential correlation between multivariate time series,and then the time series are converted into frequency domain signals through the sequence conversion unit to learn their characteristics.Finally,a variety of algorithms are combined to optimize the model to achieve better prediction effect.Experimental results illustrated that the optimized time spectrum neural network can not only predict the multivariable chaos of power system,but also has higher prediction accuracy and stability than other baseline models.