Short-Term Wind Power Forecasting Based On Big Data From Adjacent Wind Fields

In the context of energy shortage and environmental problems,countries around the world began to seek low-carbon development path,compete to develop renewable energy,one of which is wind power.Wind power can be a good substitute for fossil energy,to ensure better protection of environment under the premise of energy supply beacse of its rich resources,flexible installed style,mature technology compared to other renewable energy generation and high efficiency.After rapid development in recent years,China has become the world's largest wind power installed capacity of the country.However,wind power has a lot of randomness,intermittent and reverse load characteristics,which highly affect large-scale wind power integration and cause serious wind power curtailment in our country.Therefore,the research on wind power short-term forecasting can make up for the shortcomings of instability,help grid make reasonable scheduling plan,make more wind power accommodated,effectively alleviate wind power curtailment,is of great significance for China's wind power industry healthy and sustainable development.On the other hand,with the gradual rise of large wind data,the use of large data for wind power forecasting is a trend in the future.Deep learning is playing a more and more important role in the data mining.Among them,convolutional neural networks(CNNs)is the most mature one,successful in the field of image recognition and pattern recognition.Firstly,based on the structure of big data of adjacent wind field,the three-dimensional experimental data set is constructed through real data.And the characteristics of the experimental data set are researched by statistical distribution and dynamic correlation analysis methods,laying the foundation for the subsequent prediction modeling.Then,the CNNs model with multi output for short-term wind power forecasting is built,achieved by multiple CNNs networks' independent operation.A key explanation the CNNs model's process and a detailed analysis of the forecasting effect of the model are conducted to verify the applicability and reliability of the model.The results indicated that the CNNs model has good performance in error control.Compared to traditional methods,the CNNs model promotes the overall forecasting accuracy,at the same time achieves a more balanced forecasting effect among different time points and different sample.Finally,the combination of established CNNs model and physical forecasting model is conducted to reduce furtherly the short-term wind power forecasting error.While,structural weight for classification are selected to take full use of the advantages of the two methods in different samples.In practical work,the weight determination problem of combined model is transformed into parameter optimization problem solved quickly and efficiently by genetic algorithm(SC).And the experimental results proves that the combination forecasting model reduces error by about 5% than CNNs model,the structural weight method achieves slightly small error than a single weight.Through the research of this paper,it is proved that the CNNs network method has good application prospect in dealing with big data of wind power short-term forecasting filed to some extent.