Bio-inspired heuristic dynamic programming for high-precision real-time flow control in a multi-tributary river system

摘要

Rivers are a type of time-varying, highly nonlinear systems, especially with multiple tributaries. To address low precision and poor real-time flow control of multi-tributary river systems, a heuristic dynamic programming algorithm inspired by regulation mechanisms in the human body (Bio-HDP) is proposed. Bio-HDP consists of several control units including the model network (MNCU), the action network (ANCU), the critic network (CNCU), and the central coordination (CCCU), all of which are developed based on the interactions and mechanisms in human’s NEI system (nervous, endocrine, and immune systems). The MNCU approximates the actual system through learning actual data on the control object. The ANCU calculates the control variable according to the error between the current state of the system acquired from the MNCU and the set value. The CNCU is used to evaluate whether the current control variable obtained by the ANCU can meet the control requirements. Finally, the CCCU conducts coordinated regulation of the above three control units in real-time according to the current state of the system, and makes the control mechanism faster and more accurate. By incorporating a biological feedback mechanism into the MNCU, the ANCU, and the CNCU, we adopt a neural network with multi-level feedback for the control mechanism. We select the Chongyang River system in China as the control object to verify the effects of the Bio-HDP control algorithm. The results show that the Bio-HDP algorithm improves speed, adaptability, and accuracy of the algorithm in comparison to other algorithms, and exhibits timely and accurate control effects.