Research On Soft Subspace Clustering Algorithm Based On Firefly Optimization And Its Application

Load forecasting is a non-stationary process,which is affected by weather,season,socioeconomic and stochastic effects,so short term load forecasting is a multi-factor nonlinear complex problem.With the development of machine learning,support vector machine for regression has been widely used in the field of short term load forecasting,and effective forecasting results have been obtained.However,the traditional support vector machine for regression prediction method does not screen the training samples which have a large amount of data and many irrelevant data,resulting in poor prediction performance.To solve this problem,training samples are automatically selected by introducing soft subspace clustering algorithm.This paper takes the soft subspace clustering algorithm as the research object.The firefly algorithm and multi-objective firefly algorithm are used to optimize the soft subspace clustering algorithm,and the improved soft subspace clustering algorithm is applied to the short-term load forecasting.The main research results obtained are as follows:(1)A soft subspace clustering algorithm based on improved firefly algorithm is proposed.The algorithm introduces the objective function and a method of calculating affiliation degree to evaluate the bounded weight matrix and cluster the data samples.The weight matrix is regarded as the feasible solution of the clustering problem,and the improved firefly algorithm is used to optimize the solution to obtain the better weight matrix,so as to improve the clustering.(2)A fuzzy soft subspace clustering algorithm based on random learning firefly algorithm is proposed.This paper uses firefly algorithm with global search ability to optimize the objective function of the new algorithm.At the same time,In order to make up for premature convergence and low precision of firefly algorithm,the evolution pattern of firefly population and the learning method of global optimal particle are improved.The new algorithm formulates the weight matrix into firefly population,and transforms equality constraints of variable weighting problem into bound constraints,updating the firefly position to search for optimal weight and to explore the hidden clusters in the subspace.(3)A soft subspace clustering algorithm based on improved multi-objective firefly algorithm is proposed.Firstly,the step factor and initial attractiveness of multi-objective firefly algorithm are dynamically defined to make up for the problem that algorithm is prone to premature convergence.And in order to obtain more uniformly distributed Pareto-optimal set,a firefly single-row random learning mechanism is designed.Secondly,the improved multi-objective firefly algorithm is applied to soft subspace clustering problem,which optimizes three objective functions,i.e.within-class compactness,between-class separation and negative weight entropy simultaneously.The new algorithm formulates the cluster centers and weights into firefly population,searching the optimal cluster centers and weights by updating the location of fireflies to discover the hidden clusters in subspace.This paper takes the power load data and meteorological data of a certain area from October 25,2014 to December 31,2014 as the research object,the short-term load forecasting model based on support vector machine for regression is established.The three improved soft subspace clustering algorithms proposed in this paper are used to select training samples.The experimental results show that the support vector machine for regression model of training samples selected by soft subspace clustering optimized by improved multi-objective firefly algorithm achieves the best prediction results.