Research On Evidence Regression Modeling And Its Application Of Thermal Objects

The development of power station informationization has made it relatively easy to obtain historical operational data.Therefore,data-driven modeling based on historical operational data has become a new research hotspot for the modeling process of thermal objects.Historical operating data often have characteristics such as many working conditions and uneven distribution.In order to cover all conditions,the scale of historical trainning sample sets is relatively large,which leads to complex data-driven modeling and high hardware requirements.In addition,the expected output value of the model may not always be accurate due to the accuracy of the historical operating data itself.Therefore,in some applications,compared with the expected output value of the model,the output form of the estimated interval can help to understand the accuracy and referability of the model.Evidence regression is a non-parameter modeling method with low computational complexity,strong robustness and high modeling accuracy,which is more suitable for datadriven modeling based on large-scale training sets.In addition,it has an output form of estimation interval,which reflects the confidence and the referability of the predicted output of the model.Therefore,this paper will study the evidence regression modeling method and demonstrate its application of thermal objects.The paper provides new ideas for the modeling process of thermal objects based on historical operating data,which has a high theoretical and practical value.The main contributions of this dissertation can be summarized as followings:(1)Study on evidence regression methodIn order to adapt to the large sample scale and uneven density of historical operating data,this paper proposes an evidence regression multi-model method based on fuzzy weighted charmonic clustering algorithm.This paper improves proposes a fuzzy weighted c-harmonic clustering algorithm.The improved algorithm introduces the concept of "typical degree",which can help to construct the evidence directly for the sample sets.Firstly,this evidence construction is used to domain partitioning of evidence regression multi-model method.The sample sets is divided into smaller sub-domain sample sets,and the evidence regression model is trained separately for different sub-domain to reduce training time and improve modeling accuracy.Then,the evidence construction is also used for sample discrimination of evidence regression multi-model method.The discriminant results include the identification framework,which means a sample set that does not belong to any existing sub-domain,and thus can be conveniently used for the model updating and corrections.(2)Study on sample selection method for evidence regression of thermal objectThis paper proposes a DBS-FCM-KNN sample selection method,which is suitable for evidence regression modeling of thermal objects.The method uses density deviation sampling method and FCM clustering method to coarse-select sample sets,and the KNN wrapper sample selection suitable for evidence regression is designed.The combination of these constitutes the DBS-FCM-KNN sample selection method for the historical operation data.Examples show that this method can provide a high-quality training sample set for the evidence regression process of thermal objects.(3)Establishment of NOx emission modeling based on evidence regression multi-model methodThis paper constructs a NOx emission model based on our evidence regression multi-model method and sample selection methods.Firstly,this paper analyzes the mechanism of the NOx emission process and selects the input parameters for the modeling.Secondly,we builds the NOx emission model based on historical operating data,and compares with traditional methods,our method has less training time,higher model accuracy and an output form of estimation interval.Finally,we applied the NOx emission model established into boiler combustion optimization,which proves the effectiveness and the practicality of our modeling method.(4)Study on state warning of thermal equipment based on evidence regression multi-model methodThis paper proposes a method for state warning of thermal equipment based on the aforementioned evidence regression multi-model method.Evidence regression multi-model can calculate the estimation interval with certain confidence.Based on this,this paper constructs the normal operation state model of high-pressure heater,and calculates the estimated interval of each operating parameter.Then we directly uses this floating dynamic estimation interval as state warning range.The width of this range can directly reflect the distribution of historical operating data,thus avoiding the difficulty in setting the state warning threshold.Examples show that the state warning technology based on evidence regression multi-model has the simplicity and practicality,which can meet the needs of engineering.