Research On Application Of ADP Algorithm To Heat Station Control

The quality of urban heating is related to urban residents’ daily life and study.Scientific and rational control programs and management plays a vital role in urban heating. In the centralized heating system heat network, heating network control needs supply and demand balance and uniform heating, and on this basis, was environmentally friendly and efficient.Approximate Dynamic Programming (ADP) is a algorithm suited to optimal control for complex and nonlinear systems. It consists of a critic neural network, a model neural network and a action neural network. The BP neural network structures were constructed as an approximation of these three networks. Critic network approaches performance index function, model network approaches input-output relationship of system, action network given the appropriate control strategy. By way of learning online, ADP can continue to adjust the critic network and improve the implementation of network output control strategy to keep close to the optimal performance. based on the approximate solution online, it can be a good deal of dynamic programming in the "curse of dimensionality" problem. This paper proposed a predictive control algorithm based on approximate dynamic programming. The performance index function J reflected the deviation between the reference trajectory and the predicting output in system prediction time domain. The algorithm treated the J function optimization problem as a dynamic programming problem, uses the approximate dynamic programming,which can avoid "dimension disaster problem", to solve of the approximate value of J and obtain the best(or suboptimal) control law.By adjusting parameters of the three networks online, the system output is approached to the reference trajectory.In order to make the secondary network temperature reaches the set value, the way to control of heat stations are qualitative adjustment and volume adjustment. The paper used them to control. Firstly, system identification methods were established to adjust the channel system model. According to the control requirements, design the initial parameters of the algorithm. The simulation is in Matlab. The results showed good results in optimal control of heat stations.