Dynamic neural network control (DNNC)

Dynamic Neural Network Control (DNNC) is a model predictive control strategy potentially applicable to a large class of nonlinear systems. It uses a neural network to model the process and its mathematical inverse to control the process. The DNNC strategy differs from previous neural network controllers because the network structure is very simple and offers potential for fewer weights and better initialization of network weights. DNNC's ability to model nonlinear process behavior does not appear to suffer as a result of its simplicity.; In this research, the performance of the DNNC strategy for controlling highly non-linear CSTR with time varying parameters including activation energy (i.e., deactivation of catalyst) and heat transfer coefficient (i.e. fouling) is demonstrated. First, the DNNC controller performance was compared with PID control and a recently proposed neural network controller (NIMC) under conditions of constant parameters. In comparison to PID and NIMC, DNNC showed excellent controller performance in spite of its simplicity (total number of weights and bias terms = 15) compared to NIMC (total number of weights and bias terms = 70). The DNNC strategy is also able to reject the unmodeled disturbance more effectively than either the PID or the NIMC strategy.; For the nonlinear, time varying case the performance of DNNC was compared to the PID control strategy. DNNC (without on-line adaptation) showed excellent performance in controlling the exothermic CSTR in the region where the PID controller failed. It has been shown that the DNNC controller strategy is robust enough to perform well over a wide range of operating conditions. A first order filter was introduced into the DNNC structure in order to ensure stability and robustness.; The DNNC design technique for nonlinear dynamic systems is closely related to its stability properties. The local and global stability analysis in the DNNC framework is much easier than conventional neural networks. The results from the DNNC stability analysis will be used to define the Neural Network Stability Index (NNSI). NNSI can be used to determine the optimal network structure, to analyze the controller performance, and to design an optimal controller.; For time varying processes, updating conventional neural network models which are complex and consist of several nodes and weights, is difficult and time consuming. Since the DNNC model is very simple, the updating routine is, in fact, much faster than that of any other neural network hybrid model that has been proposed previously. DNNC has the ability to rapidly and recursively learn from plant data.; The design of multivariable controllers in the DNNC control strategy is also straightforward. The DNNC weights represent the process model and interaction between input/output and it is possible to determine the severity of the interaction by interpreting the network weights.