The Research Of Multi-model Soft Sensor Based On Data Driven

Industrial production is a complex process with characters of multiple-conditions, nonlinear, high noise, etc. When soft approach is used for unpredictable variables, normally single model can't effectively describe the characteristics of working conditions and is less able to deal with noise. Contrary with single model, multiple models can describe complex production process better and improve the estimated accuracy and generalization ability. According to the engineering application background, four methods for multi-model modeling are proposed in this thesis. Specific results are as follows:1. A multi-model modeling method based on weighted fuzzy clustering is presented. By using the correlation of input and output as weighted coefficient of fuzzy clustering, it is employed to cluster the input data of sample space, and respectively build sub-models. And then using switch mode, the corresponding sub-models estimate output and final output is determined by the corresponding sub-model output. This method can achieve a more rational division of the input data to improve the accuracy of soft-sensor model. The multi-model is applied to a soft sensor for components of BPA in a cracking reactor exports, and the simulation results show its feasibility and effectiveness.2. A novel method of multi-modeling for soft sensor is proposed in the paper. The method divides the input sample set into one sparse part and several dense parts based on the nearest neighbor clustering algorithm which it fully apply the characteristics of global and local kernel function. Meanwhile, global kernel and local kernel are applied to construct corresponding sub-model with Enhanced Support Vector Classifier. Finally, a soft sensor system with multi-model is obtained. Using the proposed algorithm to the soft-sensor model of BPA component in a Phenol evaporator outlet, the result of simulation shows the effectiveness of the algorithm.3. A novel fuzzy C-Mean clustering based on Chaotic Differential Evolution is presented, which is for multiple models soft-sensing modeling. The proposed algorithm optimizes objection function of fuzzy C-Mean clustering by using Chaotic Differential Evolution and gets a global optimal solution, which can effectively address the problems of local optimum for fuzzy C-Mean clustering. The multi-model is applied to estimating the components of BPA in a Rearrange reactor exports, it is shown that the algorithm is effective.4. A Multi-model soft senor based on Improved Locality Preserving Projection is proposed. The proposed approach extracts the features of input sample space by Supervised and Adapt Weighted Locality Preserving Projection. The multi-models can be constructed by Support Vector Machine after using the nearest neighbor classifier to divide input data space. Using the proposed algorithm to the soft-sensor model of BPA component, the result of simulation shows that the proposed approach has better performance compared with conventional Locality Preserving Projection's, and has superior in accurate estimation, and generalization ability.