Improved neural network boiler combustion system modeling method based on object combustion mechanism

Abstract

The invention discloses an improved neural network boiler combustion system modeling method based on an object combustion mechanism. A neural network comprises seven nerve cells which sequentially include a mechanism decomposition layer, a fuzzy input layer, a fuzzification layer, a fuzzy reasoning layer, a reasoning compensation layer, a normalization layer and an output layer, and network structure parameters are learned and identified by the aid of an EKF (extended Kalman filter) algorithm and an improved particle swarm algorithm. Compared with a traditional fuzzy neural network, the neuralnetwork solves the problem of high model calculation accuracy randomness caused by indeterminate fuzzy rules and the like, and the stability and the generalization ability of a model are improved. Compared with an adaptive fuzzy neural network, calculated amount of an algorithm determined fuzzy rule number is reduced, actual physical significance is given to fuzzy rules, readability and cognitionof a model are improved, and theoretical reference can be provided for on-site combustion reconstruction of a boiler and online adjustment of combustion parameters.

Description

technical field [0001] The invention relates to an improved neural network boiler combustion system modeling method based on an object combustion mechanism, which belongs to the fields of thermal power engineering and automation. Background technique [0002] Due to the high dimensionality of the input parameters of the boiler combustion model itself and the complexity of the model, the neural network is currently a widely used boiler combustion modeling tool, and the fuzzy neural network, as a combination of neural network and fuzzy algorithm, has become the neural network field. a research focus. However, it will face many problems to directly apply the traditional fuzzy neural network to the modeling of the combustion model. For example, the use of traditional fuzzy neural networks will face problems such as the difficulty of determining the number of fuzzy rules and the large randomness of model accuracy. Although the fuzzy neural network using fuzzy rule adaptive pruni...

AI Technical Summary

Problems solved by technology

But directly applying the traditional fuzzy neural network to the modeling of the combustion model will face many problems

For example, the use of traditional fuzzy neural networks will face problems such as the difficulty of determining the number of fuzzy rules and the large randomness of model accuracy. Although the fuzzy neural network using fuzzy rule adaptive pruning algorithm can reduce the number of fuzzy rules, the relationship between different fuzzy rules It is not clear, which increases the difficulty for the understanding of the model structure and the later on-site application

Images