Research On Short-term Load Forecasting Of Integrated Deep Neural Network Based On Imaging Time-Series

With the rapid growth of power consumption and the rapid expansion of power installed capacity,this has put forward more lean management for the planning and scheduling of power grids.Power generation companies and power departments need to more accurately grasp the law of load changes and development trends.Accurate grid load forecasting is an important prerequisite for the realization of scientific power generation,dispatching,and distribution.It is also an important guarantee for the safe and stable operation of the grid.Enhancing the level of short-term load forecasting has become one of the important tasks of power planning.In this case,we can help the electric power department to formulate emergency plans and safeguard measures scientifically and effectively through the help of power load forecasting.At the same time,in recent years,power load forecasting has also been the focus of research by experts and scholars.This article mainly uses integrated model methods to improve the power load forecasting accuracy.This article first proposes the short-term load forecasting method of SSA-LSTMText CNN.Singular Spectrum Analysis(SSA)is a powerful method that has emerged in recent years to study nonlinear time series data,which can extract signals representing different components of the original time series.Long and short-term memory network(LSTM)can solve the problem of gradient disappearance in recurrent neural network,making it possible to explore the time relationship between discontinuous data,and convolutional neural network(CNN)can effectively extract features in data.This chapter first decomposes the load data through singular spectrum analysis.Inputs the vector obtained after decomposition into the long and short-term memory network,uses the long and short-term memory network to obtain the time relationship in the data,and uses Text CNN to extract the features between the vectors to obtain the final prediction result.Through the analysis of numerical examples,the Text CNN network shows its best accuracy in all indicators,which also verifies the performance of the convolutional neural network in short-term load forecasting.In order to extract the long dependencies and more features in the time series,the GASF-Attention-Res Net model structure is proposed.Markov transition field(MTF)mainly uses Markov transition probability to encode time series data into images.After the time series data is converted into image data,the large receptive field of the convolutional neural network can be used to effectively extract the data in the time series.This chapter uses the Markov transition field to convert long-term historical loads into images.In terms of models,the first stage of the model is a combined model.After all data features pass through this model,a preliminary prediction result is obtained;the second stage of the model is a network architecture composed of a selfattention layer.Can capture long-term dependencies between data.Finally,the classic Res Net is used as the third stage network of the model.Residual networks can well avoid the network degradation problem caused by deep networks,and convolutional neural networks can effectively extract features.In terms of predicting the results of next 24 hours,GASF-Attention-Res Net shows the best accuracy in multiple indicators.Model integration realizes multi-model fusion of test data by fusing multiple trained models,so that the final result can be combined with the advantages of each learner and the generalization ability of the final model can be improved.Snapshot integration can obtain multiple models in one training process,and integrating these models can effectively improve the accuracy of load forecasting.Linear weighting can integrate multiple different model results.According to the basic idea of the integrated model,the proposed model—GSAR/GDAR/MTR is used as the base learner;in terms of integration strategy,weighted voting and snapshot integration are mainly used.Through experimental analysis,after GSAR,GDAR,and MAR are integrated through their respective snapshots,the prediction results of these three models are linearly weighted by logistic regression,and the advantages of each model are combined.The integrated model obtains the best prediction results,which verifies The effectiveness of the short-term load forecasting method proposed in this paper can reasonably analyze and forecast the actual power grid data.