Applying MPC in Non-Refinery Settings
- 1. Applying MPC in Non- Refinery Settings Giles Laurier Capstone Technology
- 2. Agenda Oil refining MPC in oil refining MPC in other process industries
- 3. The Origin of Species
- 5. Optimization Hierarchy Short Term Plan Advanced Control Regulatory Control Operating Objectives, Component Prices, Constraints Operating Targets Controller Setpoints Valve Positions
- 6. MPC History Oil refining started using linear programming in 1940’s First adaption to nonlinear problems (SLP) published by Shell in 1950’s Use of LP to solve control problems Exxon 1970’s DMC (Shell) 1980’s
- 7. Key features Definition of “good” control Time domain model Rules Control is online solution of optimization problem
- 8. Input 1 Input 2 Input 3 Input 4 Out 1 Out 2 Out 3 Out 4 Knowledge Base MPC Model Matrix
- 9. MPC vs PI Discrete PI Control Law Δ𝐶𝑂 = 𝐾( 𝑒 𝑡 𝑇𝑖 + (𝑒 𝑡 − 𝑒 𝑡 − 1 ) Control expressed solely in terms of current and past error
- 10. MPC Methodology PlantsTests Model Fitting Tuning Commissioning
- 11. Process models Process tests to identify the relationship Model identified for every input/output relationship Required for future prediction Models embedded directly in controller Input Output Model: How does output change to a 1 unit change in input
- 12. Refinery testing We know the answer from first principles, simulation experience Stable feed stock, stable plant Common to test multiple inputs simultaneously 2 2 )( tA n gainStdDev noise
- 13. Testing in other industries Heterogeneous feeds Transportation delay Harsh environments Less instrumentation There is not a single “best” model that converges with longer testing Recognize when model is “good enough” to satisfy the objective
- 14. Chip Digester Plug flow reactor with four hour transportation time delay
- 15. External Delay Handling U(t) Delay U(t-d) Identification Commercial time series analysis packages developed to meet the needs of oil refineries do not handle situations where the time delay varies during the plant tests External delay stacks are required to preprocess the inputs
- 16. Equation transformation Theory: Q= 𝑉 𝐹 𝑘𝑒 𝑎𝑇+𝑏 Highly nonlinear But: ln(Q) linear inT and ln(F) Transforms of both controlled and manipulated variables often required
- 17. Performance Refining Rules of thumb for variance reduction 25-40% reduction Are these realistic elsewhere?
- 18. Regulatory Performance Y(t) = G(s)u(t)+Noise(t) Process model identification focuses on G(s) Variance reduction is by undoing or correcting for the Noise(t)
- 19. Two noise samples with identical variance
- 20. Variance decomposed by frequency 0 5 10 15 20 25 30 35 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Cycles/minute Power Spectrum Power Spectrum 1 Power Specttrum 2
- 21. The noise vs control lineup
- 24. Control Performance Feedback at least as important as Predictive component
- 25. Robustness Can’t control what you can’t measure Can’t control what you can’t model
- 26. Ore Crushing
- 28. Iron ore pellet induration
- 29. 225 245 265 285 305 325 345 V.Low Low Medium High V.High (kg/p) Firing intensity Compressive strength
- 30. Calculations to infer state Unusual Controlled/Constraints Variance Dominant cycle period Continuous plant tests to detect gain reversals Logistic regression
- 31. Logistic Regression Regression method to predict the likelihood of a binary outcome as f(x) 𝑃 = 1 1 + 𝑒− 𝑎+𝑏𝑋 Similar to classical neuron Allows including constraints such as keep probability of upset event to less than z%
- 32. Bullet Proofing Instrument failures Trips Blockages Safe Parking strategies essential Pure “MPC” of successful MPC project may be less than 40% of total engineering configuration
- 33. MPC contrast Refining Stable systems with deep theoretical knowledge Other industries Dynamic/variable/frustrating environments MPC + “training wheels”
- 34. Conclusions The suitability of technology can’t be considered in isolation from the environment that created it Refining MPC evolved to address specific situations and challenges Its fitness in other industries depends on careful adaptations