Research On Prediction Method Of Renewable Energy Based On Spatiotemporal Information

The energy hunger of cloud data centers costs great electricity resource,while making such centers responsive of major carbon emission.To save energy and reduce emissions,cloud service providers are starting to supplement their cloud data centers with renewable energy.Wind energy has become a critical fountainhead of green power for cloud data centers,so it is imperative to accurately predict the availability of wind energy.As wind speed directly determines the power volume of wind energy,the problem of wind energy forecast can be transformed into wind speed prediction.At present,wind energy forecast is predominantly based on the time series prediction of wind speed,which ignores the spatial correlation of wind speed.For solve this problem,this paper proposes a model of wind speed prediction based on the wind speed of spatial-temporal information,considering the spatial-temporal features of wind speed.This method integrates the historical wind speed time series data and the spatial information of the neighboring areas to forecast the wind speed at a future moment,so as to obtain the development trend of wind energy generation amounts,and thus to optimize the scheduling and energy consumption management of cloud data centers.The major contribution of this paper comes as the following aspects.First of all,collect and process wind speed data from multiple wind speed monitoring stations.In this paper,the wind speed data from several wind speed monitoring stations located in Colorado,US,selects as the research object,and supplements the missing values in the data pre-processing stage,and performs data standardization processing.In addition,as the abnormal values and noise exist in the wind speed data,the model would not delineate the correct pattern of the wind speed data,so this paper adopts appropriate filtering methods as the data smoothing process.Secondly,the establishment and training of the wind speed temporal and spatial prediction model.In order to capture the spatial-temporal characteristics of wind speed,this paper presents a new type of spatial-temporal graph convolutional networks prediction model.The model learns the spatial topology of the graph convolution networks to obtain the spatial correlation of adjacent wind speeds,and uses the gated recurrent unit model to capture the temporal characteristics of the dynamic changes of wind speed.On this basis,several groups of comparative experiments are conducted to determine the optimal value of the model’s supra-parameters,and the model is iteratively trained and optimized to generate the final version of spatial-temporal wind speed forecast model.Finally,use the established spatio-temporal prediction model to conduct experiments and analysis on wind speed prediction.This paper selects multiple baseline models for comparative experiments,and uses multiple evaluation indicators to evaluate the performance of the models in plural dimensions.In order to verify the robustness of the model,this paper also adds random noise to analyze the interference experiment.To sum up,this paper treats the wind speed data with Savitzky-Golay filter smoothing methods,and establishes a spatio-temporal prediction model that combines graph convolutional networks and gated recurrent unit to forecast wind speed in the next time.The experimental results show that the performance of wind speed prediction at future time of this proposed method is better than other baseline models.