Process description of chemical production driving based on hybrid system Petri net model

Driving a process plant is an operation that is often required in chemical production, especially for processes that have batch or batch characteristics.

The driving process has procedural and concurrency in the operational process. It can be considered as a discrete event dynamic system, and the process control involves a large number of discrete and continuous parameters. Therefore, it has complex hybrid dynamic system (HDS) 1,2 characteristics. . The driving process often has a great influence on the production and consumption of the device. It is of practical significance to study the analytical modeling techniques of such processes and then implement effective monitoring and control strategies.

As a mathematical model for researching information systems and their mutual relations, Petri net is a "numeric form" combination modeling method, which can effectively deal with discrete and continuous system modeling and performance analysis problems, and is applied in many fields. In recent years, domestic and foreign research on the problems of hybrid systems based on Petri nets has been carried out and certain results have been achieved.

In this paper, a process description model for Hierarchical Petri Nets is developed for the process of chemical production driving. Based on the time division HDS theory, the driving process is determined as a two-tier hierarchical structure consisting of a coordination layer and a monitoring layer. The coordination layer consists of relatively independent driving process blocks. It has the characteristics of discrete event dynamic systems and can be described by a timed Petri net. The monitoring layer contains a large number of discrete and continuous control operations in the driving process. It is a typical type of interactive HDS. , using hybrid Petri net modeling. The paper takes the recovery of DMF solvent from industrial synthetic leather as an example to describe the specific process of driving.

1 Hybrid System Hierarchical Control Model The basic control methods of the process control system can be divided into three types: centralized control, decentralized control, and hierarchical control. Centralized control can achieve optimality of a single subsystem, and decentralized control can solve the control of multiple subsystems of a general large-scale system. For hybrid systems, due to the interaction of discrete events and continuous processes, the optimization control of a single subsystem is not global. Optimal operation scheme, for which a hierarchical control method can be used. As shown, the synthesis of hierarchical control is usually a top-down solution process.

Hierarchical relationships of Hierarchical HDSs based on Petri Nets can be generated by time division or by space division. For different levels of hierarchical HDS model structure HDS, according to its characteristics can be described using different types of Petri net models, general high-level discrete characteristics are more prominent, usually can be treated as discrete event dynamic system, using extended discrete Petri Networks, such as timed Petri nets (OPNs), are modeled; while lower layers often exhibit strong interactions between discrete variables and continuous variables, and are easily described using hybrid Petri nets (HPN). Based on the unity of the Petri net model description, it provides the conditions for real-time control and simulation to solve the information communication between hierarchical HDS layers.

2 Chemical production driving process Petri net description 2.1 Hierarchical Petri net model structure Based on time division to generate hierarchical HDS method, the DMF recovery driving process hierarchical model is determined as a two-tier structure: the upper layer is the coordination layer, mainly related to the driving process. The production equipment is the core. Considering the time and phase of the operation and control process, it is divided into several process blocks that are independent and interconnected. Each process block contains a set of interconnected process devices that operate as a subprocess under the role of a local controller to perform one or more tasks.

Characterize the relationship of the system at the time level.

A TPN can be represented as a group of seven frames: where PN is the basic Petri > 0, a set of delay parameters associated with the transition node or location node.

P-TFN is used here to introduce the delay parameter into the location node. For a given initial marker M, the initial vector is the delay vector Timed contains the delay time dT of each position in making the transition / j can trigger the earliest time, that is, when a token reaches P, after at least stay dr, The time unit can be used to trigger the transition of its enable output.

The driving process is actually a process of manually adjusting the process parameters to the automatic operation of the controller. The central controller of the main monitoring parameters in the driving process is the center. Based on the driving time sequence, the driving process can be divided into discrete process blocks. The discrete position quantity P of the Petri net describes the manual operation of the parameter, the discrete transition t, describes the automatic operation of the controller, and the manual operation time of the parameter can be considered as the delay parameter dr of the discrete transition.

2.3 HPN model of monitoring layer Because the process block has procedural and concurrency in operation, the coordination layer can be considered as a discrete event dynamic system. Its meters are modeled using timed Petri nets, and the status of each event is given with a The discrete position of the delay is expressed.

The lower layer is the monitoring layer whose task is to perform specific operations within the process block. The monitoring layer is characterized by the interaction between continuous control and discrete control, continuous state and discrete state. It is a typical interactive HDS. An effective method to describe such problems is the hybrid Petri net theory and technology, and the temperature, pressure, flow and Level-based simulation parameters and discrete states generated by process valves, transfer pumps, and other control devices are represented as continuous and discrete positions and transitions, respectively.

The above two-tier model describes the process monitoring features of the DMF drive system. Adding a third-tier model at the high end can further describe the entire car's start-up, operation, load increase/decrease, and parking scheduling characteristics.

2.2 Change of the TPN model of the coordination layer The amount of time introduced in the node or location node allows it to be reasonably PrePost, hTM). Where P is a finite set of positions, T is a finite transition set, and PR h is mapping PUT=CD to indicate whether the node (position or transition) is discrete (D) or continuous (C); Pre is the input relation mapping: If h(P)=CPre(PXT)*R one; if h(P)=D,Pre(PXT)*N.,R* is a non-negative real number set, N is a non-negative integer set; Post is the output relation mapping : If h (P) = C, Pst (PXT) * R one; If h (P) = D, Post (PXT) * N; t is mapping T * R + to specify a non-negative real number d for each transition. , where for D-transition jd. delay time; for C-transition jdj to reflect the maximum transmission rate, the maximum transmission rate is V / = 1 / d; M. Petri net initial identification.

The interaction of discrete parts (positions and transitions) and successive parts (positions and transitions) of the HPN can be divided into several cases: 1 the effect of discreteness on continuity; 2 the effect of continuous on dispersion; 3 conversion of continuous and discrete tokens. The process control unit's output variables such as temperature, pressure, flow, and liquid level characterize the process control index, which is generally expressed as a continuous part of the HPN, while discrete control devices such as pumps and valves act as input to influence the evaporation tank operation process. H Hongguang Li, male, born in 1963, received a master's degree from East China University of Science and Technology in 1988, and is currently a Ph.D. candidate; engaged in research on hybrid systems, Petri nets, and intelligent control.

Sensor on-line fault diagnosis based on adaptive neural-fuzzy inference Weng Guirong (School of Mechanical and Electrical Engineering, Soochow University, Suzhou 215021) Barrier on-line diagnosis principle. The simulation results show that this method has the characteristics of high convergence speed, high diagnosis accuracy and strong generalization ability, and it can diagnose online faults of sensors in various working processes of the system.

With the rapid development of modern science and technology, production equipment is becoming larger, more automated and more intelligent, and people are increasingly demanding for their safety, reliability and effectiveness. To this end, a variety of sensors are used to monitor the status of equipment in real time. Because the sensor is in a bad working environment, it is easier to damage. When the sensor fails, the system performance may be degraded or the system may be paralyzed. At present, most of the sensor fault diagnosis methods are based on the analysis of redundancy method, neural network payment 2~4 and so on. In recent years, neural network diagnostic methods have become the mainstream, especially BP neural network 23 and radial basis neural network. 4. The relationship between neural networks and fuzzy logic is the relationship between human brain structure and function, and each has its own strengths and symbiosis. The former uses the biological neural network as the simulation basis to approach the self-organizing and parallel processing functions of the human brain from the topological structure “hardware” aspect; the latter uses the fuzzy logic as the basis to understand the human information obfuscation through the way of thinking “software”. Thinking function.

In practical applications, we find that the above methods have advantages and disadvantages, especially in terms of speed, accuracy, versatility, and overall situation. These aspects will seriously affect the performance of fault diagnosis. For this reason, the author proposes to use an adaptive neuro-fuzzy inference system to solve the sensor's on-line fault diagnosis and make full use of the respective superiority of neural network and fuzzy logic.