Short Term Forecast Of Wind Power PV Power Output Based On Improved SSA-ELM

With the continuous increase in global energy demand,China is actively exploring the development and application of new energy.Vigorously developing renewable energy can not only alleviate the energy crisis caused by the shortage of oil and gas resources,but also is an important strategy for the construction of China’s modern energy system.Wind and solar power generation technologies,as mature and technologically advanced new energy generation methods in China’s clean energy development,have broad development prospects.However,due to the strong randomness of wind and solar energy,they have posed a huge challenge to the existing high proportion of new energy modern power systems.Therefore,accurate prediction of new energy output power has become the key to the stable operation of large-scale grid connection of new energy.Improved SSA-ELM wind and photovoltaic output power prediction methods have been developed for the respective output characteristics of wind and photovoltaic power generation.The specific research content is as follows:In response to the problem of poor accuracy after random generation of input layer weights and hidden layer biases in Extreme Learning Machine(ELM),the Sparrow Search Algorithm(SSA)is introduced to optimize the input layer weights and hidden layer biases,and the optimal parameter combination of ELM is obtained to achieve the goal of improving the accuracy of wind power and photovoltaic output power prediction.A combination prediction model based on box dimension empirical mode decomposition(BEEMD),permutation entropy(PE),and SSA-ELM is proposed to predict short-term wind power in response to the nonlinear characteristics of strong volatility in wind power data,starting from mining its own power characteristics.Firstly,the complementary set empirical mode decomposition(CEEMD)is used to decompose the original sequence of wind power in frequency order,calculate the box dimension of each sub sequence after decomposition,separate the signal with abnormal box dimension from the original signal,and then perform empirical mode decomposition(EMD)on the remaining signal.This can solve the problem of EMD mode aliasing and avoid unnecessary integration operations in CEEMD.The example demonstrates that BEEMD achieves decomposition effect while saving computational time.Secondly,the entropy value of each submode is calculated using PE,and sequences with similar permutation entropy are merged to form a new sequence.Finally,SSA-ELM is used as the prediction model to predict the new submode and then superimposed to obtain the power prediction value.Taking a wind farm in Jiuquan,Gansu,China as an example,the BEEMD method has been compared and proven to be less time-consuming in decomposition,resulting in a more stable data sequence signal.SSA-ELM is combined with BEEMD and PE to propose a new combined prediction model BEEMD-PE-SSA-ELM,which has higher prediction accuracy than three prediction methods:wavelet neural network(WNN),long and short term convolutional neural network(LSTM-CNN)and ELM under particle swarm optimization(PSO-ELM).A short-term photovoltaic output power prediction model based on a combination of similar day,grey Markov(GM)model,and adaptive enhancement SSA-ELM is proposed to address the significant impact of weather types on the output power prediction results of photovoltaic power generation.Firstly,by combining two similar day clustering methods,Euclidean distance(ED)and grey correlation degree(GRA),and using two similar variables,irradiance and temperature,four sets of similar days were selected for both sunny and non sunny days,providing more complementary and comprehensive input data for subsequent prediction models.Secondly,four sets of similar daily data were used as input data for the grey theory model to predict sunny and non sunny days.Markov chains were introduced to correct the errors of the grey theory model,compensating for the low accuracy of the grey theory model in the field of random data.Finally,the SSA-ELM under adaptive lifting is used to aggregate the four sets of GM prediction results and establish a strong predictor model for GM-SSAELMA.To verify the optimization effect,compared with four prediction methods such as error backpropagation neural network(BP),ELM,LSTM-CNN,and SSA-ELM,it was found that the GM-SSA-ELMA method has higher accuracy and stability on both sunny and non sunny days.