A Neural Network Model of Combustion Characteristics of Power Plant Boiler

Abstract

A power station boiler combustion characteristic neural network model comprises model input, model output and a neural network. The model input comprises output power of all combustors, coal quality parameters of pulverized coal corresponding to the combustors, pulverized coal concentration, flow speed, primary air, secondary air, over fire air, oxygen content, loads, carbon monoxide content, combustion chamber draft, pressure of all air bellows and opening degrees of all regulating mechanisms in the pulverized coal flow. The model output comprises a boiler effect measuring the performance of the boiler or NOx content relative to the boiler effect and surface characteristics. According to the physical characteristics of the model input, the model input is divided into combustor exclusive input and other global influence input which correspond to different input layers and hidden layer nerve cells. According to the power station boiler combustion characteristic neural network model, a power station boiler structure can be converted to a corresponding neural network special structure, the network model is made to comprise information of actual boiler characteristics, unnecessary interaction and coupling are avoided, subnetwork parameters with similar characteristics are made to be reusable at the same time, and high performance and efficiency are achieved.

Description

technical field [0001] The invention relates to the technical field of combustion characteristic modeling of a power station boiler, in particular to a neural network model of the combustion characteristic of a power station boiler. Background technique [0002] The boiler combustion system is the most complex and difficult to model part of the power plant system. It involves complex physical and chemical changes in a large space field, and the heat transfer causes the phase change of the working fluid. Therefore, it is difficult to describe it with classical mathematical methods, and there is no mature and reliable classical mathematical model. Neural network technology has been widely used in the field of modeling of a large number of industrial systems because of its suitability for complex nonlinear systems and self-learning characteristics. [0003] The mathematical models of boilers mostly focus on modeling on the side of steam and water, but such models are difficult...

AI Technical Summary

Problems solved by technology

[0005] 1. There are many input parameters for boiler combustion, the system is complex, and many neurons are required to form a network. Its training and learning require a large number of samples. However, the actual operating conditions of the boiler often fluctuate, and it is difficult to obtain a sufficient number of samples in a short time. ;

[0006] 2. The conventional neural network model has wide adaptability, but it cannot convert the special structure and physical laws of the boiler system into the neural network structure or parameters. The number of hidden layer neurons is difficult to determine, the modeling process is complicated and time-consuming, and the initial weight matrix value Large impact, low training success rate;

[0007] 3. The boiler system has a lot of input, but the correlation between different inputs is either very strong or weak. For the symmetrical structure adopted in the conventional neural network technology (the output of the neuron in the previous layer is connected with each neuron in the next layer elements are connected) model training is very easy to fail due to premature learning

Images