(264f) Distributed, Asynchronous Simulation of Power Generation Systems Using Recording Controllers

It is challenging to model and control a whole power plant system. Power generation systems are highly integrated internally and the dynamics change considerably due to strong integration with the power grid. The dynamics display a wide range of time-scales ranging from millisecond in the turbine and electrical systems to seconds and minutes in the steam temperature, combustion and emission control systems. All the time-scales must be modeled and controlled to ensure stable and safe power plant operation. In a power plant system, each process may itself be very complex and specialized integration routines may be needed to solve the dynamics. Connecting the units together and developing an integration scheme for the entire process may be very difficult. This property prevents the dynamic simulation of very complex systems and simplifications are needed to enable reasonable computation times and numerical robustness to prevent instability and convergence to wrong solutions. Several methods have been proposed and they are often used in combination

1.Model reduction

2.Use of simplified unit operations

3.Decouple mass and energy balances (steady state pressure flow matrix)

4.Distributed (or decentralized) simulation

The purpose of this paper is to develop a new approach to distributed simulation. In our distributed simulation, a power plant system is treated as a network consisting of physically distinct process units. Each of these units solved using their own integration routines. Completely decentralized simulation is not feasible since the dynamics of one unit depends dynamically on the states of neighboring process units in the network. The information between the different sub-systems must therefore be synchronized at some fixed or variable communication interval. The states are then reconciled and the simulation can proceed to the next step. All the sub-units are synchronized at the end of the simulation step (which may be quite short). Thus, the method converges (uniformly) to the correct solution as the simulation step decreases. The pattern of information exchanged at regular intervals is like in a sampled data system with a fixed or variable sampling rate.

In order to improve the accuracy of the method, we design components which record and control the simulation errors which accumulate due to finite communication intervals. These components are called recording controllers and they ensure that mass and energy balances are satisfied for the total system by recording and controlling mass balances at the interconnections between the sub-systems.

The main advantages of decentralized simulation in large scale systems are the following

1.Complex unit process models can be used provided they have efficient solution methods

2.Computations can be carried out in parallel and be distributed on several processing units

3.Robust solutions can be achieved since there is no higher level numerical integration needed. Problems of extreme stiffness (fast and slow time-scales) due to interaction between fast and slow processes can be avoided

4.Flexible since sub-components can be exchanged without need for developing new integration scheme's

5.Guaranteed convergence and little sensitivity with respect to choice of initial conditions.

The algorithm proceeds as follows: a time horizon is chosen or a method is developed to adaptively choose the time-horizon; each process integrates its sub-system until the time horizon is reached. This can be done in parallel or sequentially, in a single processor or on multiple processors. The interconnection variables between the different sub-systems are then updated by changing the boundary conditions.

The method introduces an integration error because the inputs to each sub-system come from an estimate instead of the actual sub-system outputs between the communication instances. The recording controllers are introduced to accumulate and control the integration errors during the dynamic simulation. At each synchronization time, new estimates for the interconnection variables are broadcast to the peer sub-systems.

The recording controllers have the following three functions:

1.Accumulate errors between communication intervals;

2.Perform error checking;

3.Control the integration errors.

These recording controllers can be designed to guarantee the convergence of the simulation trajectories so that they are very close those of the real system during transients and at steady state. The main assumption is that each individual sub-component in the network should be stable. According to the passivity theorem, the difference between the distributed and central integrations will dissipate over time. This guarantees the convergence of the state variables.

The proposed distributed simulation with recording controllers has been tested in a prototype furnace-boiler-turbine system. The simulation results show that the recording controllers have good performances in term of the simulation error and the computational complexity.