Research On Electricity Price Forecast Based On Extreme Learning Machine And MapReduce

With the increasing demand for various energy sources in the society,new energy sources are continuously connected to the grid,and the power transaction data covers conventional energy and new energy.The scale of data collection is increasing from the current GB to TB or even PB.In the future power market environment,in the face of such massive high-latitude big data,it is necessary to further improve the high-concurrency and high-performance data processing technology requirements for electricity price forecasting.The cloud computing storage model of the smart grid has achieved certain research results,but the parallel algorithm based on the electricity price forecasting model of the big data environment is rarely mentioned.In the context of the rapid development of Internet technology,this paper studies the key issues of power market operation in the big data environment,namely short-term electricity price forecasting.Firstly,it analyzes the operation mode of domestic and international power trading market and the trading characteristics of traditional energy power and distributed energy power,studies electricity price forecasting methods and big data related technologies,and combines integrated empirical mode decomposition algorithm with online nuclear extreme learning machine model.The optimized electricity price forecasting model is obtained.On this basis,according to the limitations of a single extreme learning machine,based on the idea of integrated learning combined with multiple online nuclear extreme learning machines,an online learning based online learning machine model is proposed.Compared with the traditional algorithm,the model algorithm has higher accuracy and generalization ability,and its advantages are more obvious under the condition of big data noise.Secondly,the electricity price prediction model algorithm based on integrated learning combined with multiple extreme learning machine models will increase the time complexity in the serial data processing mode.According to the characteristics of big data and the principle of data processing technology,a parallel algorithm based on MapReduce's multiple online kernel extreme learning machine models is proposed,which can enable a MapReduce job to complete online training of multiple models.The proposed MapReduce-based parallel algorithm can effectively improve the accuracy of the model and the real-time training speed.Finally,based on the MapReduce parallel model algorithm,the power price prediction function is designed and implemented for the distributed architecture power trading system.Set up a remote big data environment,install and configure the corresponding version of the Hadoop server,and configure and run Cloudera and Cloudera Desktop.The power price prediction system function designed and implemented under this architecture mainly includes three functional modules: the preprocessing module is used to calculate the original electricity price data.The pre-processing obtains the corresponding data loaded into the model;the learning training module uses the parallel model algorithm to perform parallel training of multiple online nuclear extreme learning machine models;the prediction module is used for predicting and displaying the data after learning and training.The electricity price forecasting model combined with the big data parallel processing technology makes the model more powerful,faster training and higher predictive stability.For the future development of distributed energy grid,the new generation of power trading platform in the big data environment Applied research has certain practical value.