Research On Energy Demand And Consumption Forecasting Based On Hybrid Genetic Algorithm-Localized Support Vector Regression

Primarily, the paper demonstrates the necessity of the energy security for the high-speed developing economy and the political steady, which reveals the importance of energy demand/consumption forecasting for the domestic economical development. Even the SVR has been applied successfully in many domains, it lacks the utilization of the trend between training data, and this will leads to the decrease of forecasting accuracy directly. Consequently, Localized Support Vector Regression (LSVR), which can capture this trend and then seek for the better optimal hyper-plane, is proposed in this paper. However, the definite expression of the nonlinear mapping can not be got. Hence, the decomposition algorithm and equivalent substitution method are used to kernelize LSVR model, which makes the construction of LSVR model possibly in practice.As the availability of the LSVR mentioned above, the modeling method that using LSVR to model the energy demand/consumption system is proposed in this paper to figure out the nonlinearity and small sample of energy demand/consumption forecasting effectively. However, the forecasting accuracy of the LSVR model is highly depends on proper parameters. For the above reason, the paper applies hybrid genetic algorithm to optimize the integer and real parameters simultaneously.Finally, after the analysis of the main factors that affect energy demand/consumption, the LSVR method is used to build the energy demand/consumption forecasting model based on the energy-related data provided by Shanghai Development and Reform Institute. The performance comparison between the simulation of LSVR-based and the SVR-based energy demand/consumption forecasting model proves the validity and advantages of LSVR method. To improve the modeling ability of LSVR further, hybrid genetic algorithm (HGA) is proposed to optimize the parameters of LSVR globally. And in the simulation, HGA-LSVR modeling method performs a lower error than LSVR, which testifies the advantage of HGA-LSVR modeling method comparing with LSVR. Meanwhile, the comparision also illustrates the research-worthy and social meaning of the HGA-LSVR based energy demand/consumption forecasting model.