Research On The Feature Selection And Model Optimization Methods Of Renewable Energy Prediction Based On Artificial Intelligence

With the increasing proportion of renewable energy grid access,accurate prediction is of great significance to improve the new energy consumption and grid stability,and reduce the system planning and operation cost.Therefore,this paper proposes a two-stage optimization model based on extreme learning machine(ELM)and evolutionary algorithm for renewable energy power prediction and its feature input selection and network parameters.Firstly,a feature selection algorithm based on binary genetic algorithm(BGA)is proposed.How to select the input layer data of renewable energy power prediction model has always been an important topic.The coding characteristics of binary genetic algorithm make it very suitable for the problem of feature input selection.In the process of feature selection proposed in this paper,a large number of input features are encoded by binary code and then go into the algorithm for gene optimization.Through 100 repeated experiments,the final frequency of each gene was recorded to represent the correlation between the gene and the predicted object,and the input feature combination with low prediction error was successfully selected.Compared with the traditional prediction model,this method can reduce the cost of calculation and storage without reducing the learning performance.Secondly,this paper introduces the natural evolution strategy NES algorithm to optimize the network parameters of the prediction model in two stages.In the first stage,extreme learning machine(ELM)and generalized matrix inverse(GMI)are used to initialize the model network parameters.Because the process of the first initialization of the network parameters of the extreme learning machine model will cause the initial model to over fit the training set,it is necessary to adjust the parameters obtained in the first stage by introducing the optimization algorithm.In the second stage,the natural evolution strategy algorithm is used to optimize and adjust the parameters for the second time on the basis of the original model training results.The results show that the optimization algorithm can effectively improve the prediction accuracy of the prediction model on the test set.Finally,combined with a case of solar power output prediction,this paper carries out a joint optimization experiment based on feature input and two-stage parameter training,and makes a comparison with the results of traditional optimization methods.The comparison results show that the proposed optimization method is superior to the traditional prediction models in all aspects of prediction error.