Soft Sensor Modeling Based on Extreme Learning Machine and Case-Based Reasoning

Neural network (NN) and Case-based reasoning (CBR) have common advantages over other learning strategies. NN and CBR can be directly applied to the classification and regression problem without additional transform mechanisms. However, they all have disadvantages. The knowledge representation of NN is unreadable and this black box property restricts the application of NN to areas which needs proper explanations. Meanwhile CBR suffers from the feature-weighting problem, when CBR measures the distance between cases, some input features should be treated more importantly than others. This paper, we propose a hybrid prediction system of extreme learning machine (ELM) and Case-based reasoning (ELM-CBR). In our hybrid system, the feature weight set calculated from the trained ELM network plays the core role in connecting both the learning strategies, and the explanation on prediction can be given by presenting the most similar cases from the case base. Moreover, the prediction value of the Online Sequential Extreme Learning Machine also utilized in conjunction with the neighborhood information. This provides extended information for the query with most similar cases in the database. Finally, we present an application in the sugarcane juice clarification, experiments show that the hybrid system has a better recognition rate compared the k-NN and GA-CBR method.