Research On Short-term Electric Load Forecasting And Wind/photovoltaic Power Prediction

With the proposal of carbon neutral,carbon peak and other policies,new energy power systems with high proportion of wind power and photovoltaic power generation are becoming increasingly popular in the national power grid.The randomness and discontinuity of the output of wind power and photovoltaic fields have a significant impact on the frequency control of the power system,bringing great challenges to the safe,economic,and stable operation of the power system.Accurate forecasting of wind and solar power and load is very important for economic and stable operation as well as safe and efficient dispatching of power systems.This paper is proposed a variety of methods to predict power load and wind and solar power and committed to improving the accuracy of their prediction.The main work and contributions are as follows:In short-term load forecasting,this paper proposes a hybrid neural network forecasting model based on Convolutional Neural Networks(CNN)and Gated Recurrent Unit(GRU)with dual attention mechanism,which improves the accuracy of load forecasting by combining multiple models to effectively learn the rich features and patterns of various input data.The impact of meteorological factors(temperature and humidity)on the load are firstly considered,followed by further consideration of the impact of economic factors on the load,such as the level of electricity price.Both factors and historical load data are used as input together to build a CNN Attention module.CNN fully extracts the feature information of time series data in different time dimensions,and the attention module allocates the weight of the feature information extracted by CNN,so that the model can focus on the most effective information under limited resources and reduce the amount of calculation.The GRU Attention module is constructed.GRU extracts the temporal information of features and gives the importance of information through the attention mechanism to avoid the problem of information loss caused by long sequence of GRU.The experiments are carried out based on the power load data of New South Wales,Australia.The proposed model has been verified to have better accuracy after comparison and analyzation.In wind power prediction,this paper proposes a hybrid neural network prediction method based on attention mechanism,Temporal Convolutional Network(TCN)and Bi Gated Recurrent Unit(Bi GRU).First,from the perspective of data preprocessing,through the analysis and processing of the data characteristics of wind power generation,the interpolation method is used to fill the missing data.Secondly,the time convolution network is used to mine the association between multi-dimensional time series data,extract high-dimensional time series features,accelerate the convergence speed of the residual module,establish a two-way gated cycle unit and introduce attention mechanism at the output end to build a time series attention module,enhance the ability of Bi GRU to mine two-way time series relations,improve the impact of important historical time series information,and obtain the final wind power prediction value.The experiment results show that compared with the classical model,the proposed model can effectively improve the accuracy of wind power prediction.In photovoltaic power prediction,considering that there are many meteorological factors affecting photovoltaic,and although model stacking can improve the accuracy,it will greatly increase the number of network layers and increase the prediction time.In this paper,a short-term photovoltaic power combination prediction model based on improved time convolution network and bidirectional gated cycle unit is proposed.Firstly,the characteristics of photovoltaic output are analyzed,and the Pearson coefficient method is used to determine the more important related factors affecting photovoltaic power generation.Then,the new data set is integrated into the improved time convolution network,and the internal spatial dynamic change law of the learning characteristics is modeled.The dynamic change of the learning time sequence is input into Bi GRU,and the attention mechanism is introduced to give different weights to the implicit state of Bi GRU through mapping weights and learning parameter matrix,reducing the loss of historical information and enhance the impact of important information.Compared with traditional methods,the prediction results of the proposed model have higher accuracy.