Research On Cloud Computing And Machine Learning Algorithms In The Power Load Forecasting

Short-term load forecasting is the basement of achieving optimal operation in thepower system, it has a significant impact for the security, reliability and economy ofpower system. As the development of intelligent power grid, some super-big cities duringthe period of peak demand will face millions of records of the electric power data, oneyear of data storage size will increase from the current level of GB to terabytes, evenpetabytes. At the same time, data dimension also increase from dozens to hundreds oftransition. If do a traditional load forecasting algorithm in such a mass ofhigh-dimensional data, the PC will encounter the bottleneck of computing resources. Insuch a mass of high-dimensional data on traditional load forecasting will encounter thebottleneck of computing resources. Even though the storage model of the smart gridbased on cloud computing has made some certain developments, but some parallelalgorithms of field of electric power based on cloud computing are few studies. At thisbackground, this paper research on above problems.Firstly, In order to improve the classification accuracy and effectiveness, andprovides effective reference for data preprocessing stage of load forecasting, this paperpropose a Parallel Quantum-Behaved Particle Swarm Optimization Fuzzy C-Meansclustering algorithm. Quantum particle swarm intelligence algorithm (QPSO) isintroduced into the traditional fuzzy c-means (FCM) clustering algorithm, using QPSO‘sstronger global search ability overcome FCM algorithm‘s weakness of falling into localoptimum easily and sensitive to initial clustering center. Secondly, In the stage of loadforecasting, this paper propose a new algorithm which introduce sequential minimaloptimization algorithm (SMO) into the support vector electric load forecastingalgorithms (ε SVR) to realize the fast training of ε SVR. In addition, for the practicalapplication of power load forecasting, a distributed power load forecasting algorithmbased on cloud computing and extreme learning machine was proposed. Distributed andmulti-agent technology were introduced to the forecasting algorithm, load forecastingalgorithm prediction accuracy was improved. At last, The MapReduce programmingframework and the HBase of cloud computing are adopted to the proposed algorithm forparallel improvements, so the ability to deal with large amounts of high dimensional datawas improved.Finally, do some experiment tests and case analysis to prove the algorithms. SomeStandard test data sets of UCI and European Network on Intelligent Technologies’ realpower load data are used to do some experiments on the cloud computing cluster whichis set up and configured in the lab and compared with the traditional load forecastingalgorithm experiment. Experimental results show that the proposed model for load forecasting accuracy is superior to power load forecasting algorithm, so it can not onlyprovide effective basis for short-term power load forecasting, but also have better parallelperformance.