Research On Distributed Energy Grid-connected Massive Data Processing And Decision-making Assistance Technology

Under the current dual-carbon target strategy,distributed photovoltaic power generation has become an important direction for energy transformation layout due to its flexibility in deployment and advantages of small transmission and distribution losses.However,photovoltaic power generation is strongly affected by many uncontrollable factors,such as meteorological environment,and has a high degree of randomness and intermittency.These characteristics lead to significant impacts on the stable operation of the entire power grid system after large-scale access of photovoltaic power generation.Accurate photovoltaic power prediction can assist the power grid system in flexible scheduling and timely decision-making,improving the reliability of the power grid.Based on the National Grid Corporation of China’s scientific research project"Distributed Energy Resources Integration:Massive Data Processing and Decision Support" and combined with the current practical scenarios of distributed photovoltaics,this paper designs and implements a distributed photovoltaic big data platform,which systematically solves the difficulties of data collection,storage,and monitoring caused by the explosive growth of data after massive distributed photovoltaics access to the grid.At the same time,a distributed photovoltaic power plant ultra-short-term power prediction model based on the fusion of multiple features is proposed,which realizes the overall modeling of all distributed power stations in the region.After adding an online learning mechanism,the prediction accuracy and real-time performance of the model are further improved.The work of this paper can be summarized as follows:(1)In order to give full play to the role of historical data collected by the platform in power prediction research,this paper proposes a DeepFM&LSTM distributed photovoltaic ultra-short-term power prediction model based on the fusion of multiple features.Through the correlation analysis of the original features,a series of features strongly related to power generation are excavated.At the same time,related encoding features are processed through embedding and input to the model together to realize the overall modeling of the distributed power station.Finally,the effectiveness of the proposed model and its effect on the power prediction of newly constructed photovoltaic power stations are verified through experiments.(2)Due to the randomness of photovoltaic power generation,existing models have difficulty ensuring the accuracy of photovoltaic power prediction in scenarios with complex meteorological conditions.In this paper,we propose to use an online learning mechanism to assist prediction.By using the XGBoost model that incorporates latitude and longitude and power distribution interval features for rapid iterative training on the data within a relatively short time window of the power station,we predict the prediction residual or correlation coefficient of the basic model(DeepFM&LSTM).Finally,we verify the effectiveness of the model’s assisted prediction through experiments,and analyze the impact of two prediction methods and the size of the time window on the prediction performance.(3)To address the difficulties of distributed photovoltaic data collection and storage,this paper designs and implements a photovoltaic big data platform based on the Hadoop platform,which automates photovoltaic data collection,storage,and computing tasks.While solving the problems of scattered data sources and diversified data types,the integrity and validity of the data are ensured,and efficient storage,query,and multi-angle summary visualization of photovoltaic data are realized.