Piecewise Linear

Piece-wise Linear Approximation

A piece-wise linear function is an approximation of a nonlinear relationship. For more nonlinear relationships, additional linear segments are added to refine the approximation.

As an example, the piecewise linear form is often used to approximate valve characterization (valve position (% open) to flow). This is a single input-single output function approximation.

In Situ Adaptive Tabulation (ISAT) is an example of a multi-dimensional piecewise linear approximation. The piecewise linear segments are built dynamically as new data becomes available. This way, only regions that are accessed in practice contribute to the function approximation.

APMonitor PWL Object

  File p.txt
    1,   1
    2,   0
    3,   2
    3.5, 2.5
    4,   2.8
    5,   3
  End File

  Objects
    p = pwl
  End Objects

  Connections
    x = p.x
    y = p.y
  End Connections

  Parameters
    ! independent variable
    x = 6.0
  End Parameters

  Variables
    ! dependent variable
    y
  End Variables
[$[Get Code]]

Equivalent Problem with Slack Variables

  Parameters
    ! independent variable
    x = 6.0

    ! data points
    xp[1] = 1
    yp[1] = 1

    xp[2] = 2
    yp[2] = 0

    xp[3] = 3
    yp[3] = 2

    xp[4] = 3.5
    yp[4] = 2.5

    xp[5] = 4
    yp[5] = 2.8

    xp[6] = 5
    yp[6] = 3
  End Parameters

  Variables
    ! piece-wise linear segments
    x[1]             <=xp[2]
    x[2:4] >xp[2:4], <=xp[3:5]
    x[5]   >xp[5]

    ! dependent variable
    y

    ! slack variables
    slk_u[1:4]
    slk_l[2:5]
  End Variables

  Intermediates
    slope[1:5] = (yp[2:6]-yp[1:5]) / (xp[2:6]-xp[1:5])
    y[1:5] = (x[1:5]-xp[1:5])*slope[1:5]
  End Intermediates

  Equations
    minimize slk_u[1:4]
    minimize slk_l[2:5]

    x = x[1]   + slk_u[1]
    x = x[2:4] + slk_u[2:4] - slk_l[2:4]
    x = x[5]                - slk_l[5]

    y = yp[1] + y[1] + y[2] + y[3] + y[4] + y[5]
  End Equations
[$[Get Code]]

PWL in GEKKO Python

import matplotlib.pyplot as plt
from gekko import GEKKO
import numpy as np

m = GEKKO()
m.options.SOLVER = 1

xz = m.FV(value = 4.5)
yz = m.Var()

xp_val = np.array([1, 2, 3, 3.5,   4, 5])
yp_val = np.array([1, 0, 2, 2.5, 2.8, 3])
xp = [m.Param(value=xp_val[i]) for i in range(6)]
yp = [m.Param(value=yp_val[i]) for i in range(6)]

x = [m.Var(lb=xp[i],ub=xp[i+1]) for i in range(5)]
x[0].lower = -1e20
x[-1].upper = 1e20

# Variables
slk_u = [m.Var(value=1,lb=0) for i in range(4)]
slk_l = [m.Var(value=1,lb=0) for i in range(4)]

# Intermediates
slope = []
for i in range(5):
    slope.append(m.Intermediate((yp[i+1]-yp[i]) / (xp[i+1]-xp[i])))

y = []
for i in range(5):
    y.append(m.Intermediate((x[i]-xp[i])*slope[i]))

for i in range(4):
    m.Obj(slk_u[i] + slk_l[i])

m.Equation(xz == x[0]   + slk_u[0])
for i in range(3):
    m.Equation(xz == x[i+1] + slk_u[i+1] - slk_l[i])
m.Equation(xz == x[4]                - slk_l[3])

m.Equation(yz == yp[0] + y[0] + y[1] + y[2] + y[3] + y[4])

m.solve()

plt.plot(xp,yp,'rx-',label='PWL function')
plt.plot(xz,yz,'bo',label='Data')
plt.show()
[$[Get Code]]

Piece-wise Function with Lookup Object

! Piece-wise function with the LOOKUP object

! create the csv file
File m.csv
  input,  y[1],  y[2]
      1,     2,    4
      3,     4,    6
      5,    -5,   -7
     -1,     1,  0.5
End File

! define lookup object m
Objects
  m = lookup
End Objects

! connect m properties with model parameters
Connections
  x = m.input
  y[1] = m.y[1]
  y[2] = m.y[2]
End Connections

! simple model
Model n
  Parameters
    x = 1
    y[1]
    y[2]
  End Parameters

  Variables
    y
  End Variables

  Equations
    y = y[1]+y[2]
  End Equations
End Model
[$[Get Code]]