Boiler combustion generalized predictive control method based on dual-stage neural network

Abstract

The invention relates to a boiler combustion generalized predictive control method based on a dual-stage neural network. A multi-step predictive model of a nonlinear time delay system is established through the dual-stage neural network and serves as a predictive model in generalized predictive control, a better identification effect is obtained, meanwhile, falling into a local minimum value is avoided, and the excellent prediction of an output value of a boiler combustion system in the future is ensured; and meanwhile, an improved proportional-integral performance index type implicit generalized predictive controller is provided, a proportional-integral structure is adopted to carry out optimization design on an objective function, a control increment selection strategy of the generalized predictive control is improved, a predicted control increment at a future moment is utilized to correct a control increment at a current moment, and the control effect is optimized, so that combustion of a boiler can be better controlled, and the combustion efficiency is improved. Preliminary experiment results show that the combustion efficiency of the boiler combustion system can be improved according to the design scheme.

Description

technical field [0001] The invention relates to the field of boiler combustion systems, in particular to a generalized predictive control method for boiler combustion based on a two-stage neural network. Background technique [0002] my country is the world's largest coal producer and consumer. The annual power supply, the production of steel and other items, and the smelting of metals all require the treatment of industrial boilers. A boiler is a power device that heats water to generate water vapor. The boiler combustion system is a system with complex input and output. Combustion is the core link. The combustion process directly affects equipment use, resource waste, and workers. Therefore, it is particularly important to improve the controllability of the combustion process and improve the combustion efficiency of the boiler. [0003] The essence of the boiler combustion process is the energy conversion process of converting the chemical energy in the fuel into steam he...

AI Technical Summary

Problems solved by technology

When the existing method uses predictive control to solve a class of system control problems such as boiler combustion, most of them use neural networks, fuzzy models, etc. Control the input to realize the control of the system, but there are still some problems: 1. Existing methods such as literature [Song Qingkun, Li Yuansong. RBF neural network boiler combustion system modeling [J]. Journal of Harbin University of Science and Technology, 2016, 21 (01):89-92.] The neural network prediction model of the boiler combustion system is established through the RBF neural network, which improves the prediction accuracy compared with the BP network. The number of nodes in the hidden layer of the network is difficult to determine, the training process is slow, and it is easy to fall into local small values, which will lead to inaccurate predicted model output values, resulting in errors in the calculation of control signals, and inaccurate output values; 2. Existing methods Such as the literature [Wang Sheng, Zhang Jiayan. Application of improved generalized predictive control in the main steam temperature of thermal power boilers [J]. Journal of Chifeng University (Natural Science Edition), 2019,35(12):49-53.] A stepwise generalized predictive control algorithm is used to reduce the amount of calculation and improve the stability of the output value. [Li M, Zhou Y, Wu Q. Generalized predictive control of time-delay nonlinear systems based on extreme learning machine[ C] / / 2018Chinese Control And Decision Conference (CCDC).2018.] designed an implicit generalized predictive control algorithm to achieve the purpose of reducing the amount of calculation, but the generalized predictive controller designed by the above method is based on the predicted future deviation To calculate the control amount, there is a lack of consideration of the impact of current and previous deviations on the control system, and there will be problems such as excessive overshoot and long adjustment time, which will affect the quality of control

Images