Design And Implementation Of Short-term Load Forecasting System Based On Improved PSO-LSSVM

In recent years,with the continuous deepening of power reform and the gradual improvement of power marketization,the production mode of generating electricity according to plan is gradually being replaced by the mode of on-demand production.Due to the special nature of electric energy,electric power load forecasting has become an important basis for power companies to carry out load dispatching and routine maintenance of the power grid.The traditional load forecasting method has higher prediction accuracy for smooth load data.However,the factors affecting load change are increasingly complex,and the effects of multiple factors on the load show nonlinearity and uncertainty,making it difficult to accurately describe the traditional load forecasting method.The trend of load changes.Therefore,power companies urgently need more accurate load forecasting schemes in order to adopt more scientific and efficient management methods to improve the safety and economy of power systems.Least Square Support Vector Machine(LSSVM)can solve practical problems with small samples,nonlinear,high-dimensional,local minimum points,etc.,and has strong generalization ability.The field of power load forecasting has been widely used.In the actual application process,the choice of LSSVM parameters depends on manual experience,lack of corresponding theoretical support,resulting in the accuracy of the algorithm can not be guaranteed,so it is very important to choose a parameter suitable for the specific problem Moreover,applying the research load forecasting algorithm to practical projects is of great academic and engineering value.This article mainly contains the following three aspects:(1)Analyze the relevant factors affecting the load change;and explain the identification and processing methods of such data for the problem that there may be abnormal data in the load data;in order to avoid the unit and the order of magnitude of the electric load data,meteorological data and holiday data The difference in the impact on the prediction gives a related data normalization treatment.(2)Study the application of LSSVM algorithm in short-term load forecasting.Aiming at the shortcomings of this algorithm,this paper introduces particle cluster optimization algorithm(Panicle Swarm Optimization)(PSO)to optimize its parameters,and establishes PSO-LSSVM short-term load forecasting model.Although the performance of the algorithm is improved compared with the LSSVM algorithm,the original PSO is prone to premature convergence,and the performance of the algorithm still has room for improvement.Therefore,an improved strategy of the original PSO is proposed.Firstly,the inertia weight is improved,and the nonlinear decreasing form is adopted to ensure that the particle has a wider search ability at the beginning of the algorithm.When the algorithm is iterative,the particle can perform a more detailed search.Secondly,it is the improvement of the learning model.In the late stage of the algorithm iteration,when the particles are too similar,the particle diversity can be guaranteed and the excessive invalid search can be avoided.This paper proposes a new learning model,which can effectively improve the search performance of the algorithm.Finally,based on the above work,an improved PSO-LSSVM short-term load forecasting model is established.Through experimental comparison,it is verified that the relevant improvement strategy proposed in this paper is effective.(3)Design and implement a smart energy efficiency management cloud platform.The platform is based on the Java language development and uses the B/S architecture.The load forecasting system is used as the subsystem of the platform,and is implemented based on the improved PSO-LSSVM algorithm proposed in this paper.The system provides functions such as data acquisition,data processing,load forecasting,and data display,and is designed to provide power companies with a more practical and accurate method of load forecasting.