4 ms·

Even when no communication is allowed during the game, some kind of PID control could make the supply chain stable.

oehpr 7y ago
It's funny you should mention that, because most PID configurations are not very stable. Once you introduce the Integral and the Derivative, if you're not careful that PID will oscillate into the stratosphere. In fact, I'd say that PID's that are functioning most efficiently are very nearly tuned to the point of oscillating. Even worse is the kinds of adjustments you have to make to PID's when there is significant lag between their inputs and outputs.
- rcxdude 7y ago
  Indeed. Control loops are generally limited by the bandwidth of the system they are controlling, and lag in the system limits the bandwidth you can achieve. Trying to get a PID loop to move a system faster than that system's bandwidth will just create oscillations, and the more tightly the PID is tuned the worse its behaviour when exposed to shocks (especially when the system becomes significantly non-linear in the process). You can extract more performance out of a system by having a very good predictive model of it and measuring its inputs: This can really improve how you drive the system but it's much more difficult to achieve, and you're still limited by how quickly information moves through the system in terms of how you can react.
  - hef19898 7y ago
    You described Supply Chains pretty well, I think!
  - _0ffh 7y ago
    Yeah, I think that's a much more promising approach. Have a good model, use it to build an observer, now you can control based on the state of your simulated system.
jacques_chester 7y ago
The Beer Distribution Game is one of the touchstone examples for introductions to Systems Dynamics, a field pioneered at MIT with direct roots in control theory. PID controllers are not news in that part of the world. A major insight of the field is that lags due to accumulation create wildly unexpected behaviours in systems that humans try to control. Many automated systems struggle too.

It's funny you should mention that, because most PID configurations are not very stable. Once you introduce the Integral and the Derivative, if you're not careful that PID will oscillate into the stratosphere. In fact, I'd say that PID's that are functioning most efficiently are very nearly tuned to the point of oscillating. Even worse is the kinds of adjustments you have to make to PID's when there is significant lag between their inputs and outputs.

rcxdude 7y ago
Indeed. Control loops are generally limited by the bandwidth of the system they are controlling, and lag in the system limits the bandwidth you can achieve. Trying to get a PID loop to move a system faster than that system's bandwidth will just create oscillations, and the more tightly the PID is tuned the worse its behaviour when exposed to shocks (especially when the system becomes significantly non-linear in the process). You can extract more performance out of a system by having a very good predictive model of it and measuring its inputs: This can really improve how you drive the system but it's much more difficult to achieve, and you're still limited by how quickly information moves through the system in terms of how you can react.
- hef19898 7y ago
  You described Supply Chains pretty well, I think!
- _0ffh 7y ago
  Yeah, I think that's a much more promising approach. Have a good model, use it to build an observer, now you can control based on the state of your simulated system.

jacques_chester 7y ago

The Beer Distribution Game is one of the touchstone examples for introductions to Systems Dynamics, a field pioneered at MIT with direct roots in control theory. PID controllers are not news in that part of the world. A major insight of the field is that lags due to accumulation create wildly unexpected behaviours in systems that humans try to control. Many automated systems struggle too.