ELM-PSO-FCM Based Missing Values Imputation for Byproduct Gas Flow Data Analysis
Byproduct gas flow data analysis is necessary for scheduling optimization in iron and steel enterprises. However, missing values are inevitable for reasons like sensor fault and data transmission error. Our work focused on data loss problem and proposed a robust method for missing data imputation. Fuzzy c-means (FCM) was employed as the basic principle in our work. In order to improve the robustness of FCM, three strategies were introduced to the approach. Linear interpolation was first adopted to enhance the accuracy of convergence. Parameters of FCM were also optimized by means of Particle Swarm Optimization (PSO). Furthermore, Extreme Learning Machine (ELM) was used to improve the generalization performance of the data imputation model. To fully evaluate the proposed method, several experiments were elaborated and the results proved the superior characteristics.
Missing data, Imputation,Fuzzy c-means,PSO, Extreme learning machine