I want to write a program that makes use of Newtons Method:
To estimate the x of this integral:
Where X is the total distance.
I have functions to calculate the Time it takes to arrive at a certain distance by using the trapezoid method for numerical integration. Without using trapz.
function T = time_to_destination(x, route, n)
h=(x-0)/n;
dx = 0:h:x;
y = (1./(velocity(dx,route)));
Xk = dx(2:end)-dx(1:end-1);
Yk = y(2:end)+y(1:end-1);
T = 0.5*sum(Xk.*Yk);
end
and it fetches its values for velocity, through ppval of a cubic spline interpolation between a set of data points. Where extrapolated values should not be fetcheable.
function [v] = velocity(x, route)
load(route);
if all(x >= distance_km(1))==1 & all(x <= distance_km(end))==1
estimation = spline(distance_km, speed_kmph);
v = ppval(estimation, x);
else
error('Bad input, please choose a new value')
end
end
Plot of the velocity spline if that's interesting to you evaluated at:
dx= 1:0.1:65
Now I want to write a function that can solve for distance travelled after a certain given time, using newton's method without fzero / fsolve . But I have no idea how to solve for the upper bound of a integral.
According to the fundamental theorem of calculus I suppose the derivative of the integral is the function inside the integral, which is what I've tried to recreate as Time_to_destination / (1/velocity)
I added the constant I want to solve for to time to destination so its
(Time_to_destination - (input time)) / (1/velocity)
Not sure if I'm doing that right.
EDIT: Rewrote my code, works better now but my stopcondition for Newton Raphson doesnt seem to converge to zero. I also tried to implement the error from the trapezoid integration ( ET ) but not sure if I should bother implementing that yet. Also find the route file in the bottom.
Stop condition and error calculation of Newton's Method:
Error estimation of trapezoid:
Function x = distance(T, route)
n=180
route='test.mat'
dGuess1 = 50;
dDistance = T;
i = 1;
condition = inf;
while condition >= 1e-4 && 300 >= i
i = i + 1 ;
dGuess2 = dGuess1 - (((time_to_destination(dGuess1, route,n))-dDistance)/(1/(velocity(dGuess1, route))))
if i >= 2
ET =(time_to_destination(dGuess1, route, n/2) - time_to_destination(dGuess1, route, n))/3;
condition = abs(dGuess2 - dGuess1)+ abs(ET);
end
dGuess1 = dGuess2;
end
x = dGuess2
Route file: https://drive.google.com/open?id=18GBhlkh5ZND1Ejh0Muyt1aMyK4E2XL3C
Observe that the Newton-Raphson method determines the roots of the function. I.e. you need to have a function f(x) such that f(x)=0 at the desired solution.
In this case you can define f as
f(x) = Time(x) - t
where t is the desired time. Then by the second fundamental theorem of calculus
f'(x) = 1/Velocity(x)
With these functions defined the implementation becomes quite straightforward!
First, we define a simple Newton-Raphson function which takes anonymous functions as arguments (f and f') as well as an initial guess x0.
function x = newton_method(f, df, x0)
MAX_ITER = 100;
EPSILON = 1e-5;
x = x0;
fx = f(x);
iter = 0;
while abs(fx) > EPSILON && iter <= MAX_ITER
x = x - fx / df(x);
fx = f(x);
iter = iter + 1;
end
end
Then we can invoke our function as follows
t_given = 0.3; % e.g. we want to determine distance after 0.3 hours.
n = 180;
route = 'test.mat';
f = #(x) time_to_destination(x, route, n) - t_given;
df = #(x) 1/velocity(x, route);
distance_guess = 50;
distance = newton_method(f, df, distance_guess);
Result
>> distance
distance = 25.5877
Also, I rewrote your time_to_destination and velocity functions as follows. This version of time_to_destination uses all the available data to make a more accurate estimate of the integral. Using these functions the method seems to converge faster.
function t = time_to_destination(x, d, v)
% x is scalar value of destination distance
% d and v are arrays containing measured distance and velocity
% Assumes d is strictly increasing and d(1) <= x <= d(end)
idx = d < x;
if ~any(idx)
t = 0;
return;
end
v1 = interp1(d, v, x);
t = trapz([d(idx); x], 1./[v(idx); v1]);
end
function v = velocity(x, d, v)
v = interp1(d, v, x);
end
Using these new functions requires that the definitions of the anonymous functions are changed slightly.
t_given = 0.3; % e.g. we want to determine distance after 0.3 hours.
load('test.mat');
f = #(x) time_to_destination(x, distance_km, speed_kmph) - t_given;
df = #(x) 1/velocity(x, distance_km, speed_kmph);
distance_guess = 50;
distance = newton_method(f, df, distance_guess);
Because the integral is estimated more accurately the solution is slightly different
>> distance
distance = 25.7771
Edit
The updated stopping condition can be implemented as a slight modification to the newton_method function. We shouldn't expect the trapezoid rule error to go to zero so I omit that.
function x = newton_method(f, df, x0)
MAX_ITER = 100;
TOL = 1e-5;
x = x0;
iter = 0;
dx = inf;
while dx > TOL && iter <= MAX_ITER
x_prev = x;
x = x - f(x) / df(x);
dx = abs(x - x_prev);
iter = iter + 1;
end
end
To check our answer we can plot the time vs. distance and make sure our estimate falls on the curve.
...
distance = newton_method(f, df, distance_guess);
load('test.mat');
t = zeros(size(distance_km));
for idx = 1:numel(distance_km)
t(idx) = time_to_destination(distance_km(idx), distance_km, speed_kmph);
end
plot(t, distance_km); hold on;
plot([t(1) t(end)], [distance distance], 'r');
plot([t_given t_given], [distance_km(1) distance_km(end)], 'r');
xlabel('time');
ylabel('distance');
axis tight;
One of the main issues with my code was that n was too low, the error of the trapezoidal sum, estimation of my integral, was too high for the newton raphson method to converge to a very small number.
Here was my final code for this problem:
function x = distance(T, route)
load(route)
n=10e6;
x = mean(distance_km);
i = 1;
maxiter=100;
tol= 5e-4;
condition=inf
fx = #(x) time_to_destination(x, route,n);
dfx = #(x) 1./velocity(x, route);
while condition > tol && i <= maxiter
i = i + 1 ;
Guess2 = x - ((fx(x) - T)/(dfx(x)))
condition = abs(Guess2 - x)
x = Guess2;
end
end
I need to solve the following non-linear problem with constraints but I am not sure to use the appropriate function and solver as changing from the algorithm: 'interior-point' to 'sqp' give different answers. Could you help first to make sure fmincon is the most appropriate function here and also if the options should be set up in a different way to ensure that the optimization does not stop too early. I have tried to set up: 'MaxFunctionEvaluations' to 1000000 and 'MaxIterations'to 10000 and this helps but the time required increases as well dramatically.
w is a vector whose values have to be optimized.
scores is a vector of the same size as w.
the size if w and scores is usually between 10 and 40.
Here is the optimization setup:
objective function: sum(log(abs(w)));
linear inequalities1: w_i > 0 if scores_i > 0
linear inequalities2: w_i < 0 if scores_i < 0
non-linear equalities1: sum(abs(w)) = 1
non-linear inequalities1: sqrt(w * sigma * w') <= 0.1
Here is my code:
N = numel(scores);
%%%LINEAR EQUALITIES CONSTRAINTS
Aeq = [];
beq = [];
%%%LINEAR INEQUALITIES CONSTRAINTS
temp = ones(0,N);
temp(scores >= 0) = -1;
temp(scores < 0) = 1;
A = diag(temp);
b = zeros(N,1);
%%%BOUNDs CONSTRAINTS
lb = [];
ub = [];
w0 = ones(1,N)/N;
nonlcon = #(w)nonlconstr(w,sigma,tgtVol);
options = optimoptions('fmincon','Algorithm','interior-point','MaxFunctionEvaluations',1000000,'MaxIterations',10000);
%options = optimoptions('fmincon','Algorithm','interior-point','OptimalityTolerance',1e-7);
[w,fval,exitflag,output] = fmincon(#(w)objectfun(w),w0,A,b,Aeq,beq,lb,ub,nonlcon,options);
function [c,ceq] = nonlconstr(w,sigma,tgtVol)
c = sqrt(w*sigma*w') - tgtVol;
ceq = sum(abs(w)) - 1;
end
function f = objectfun(w)
f = sum(-log(abs(w)));
end
I am using GA to optimize the parameters of the membership functions in my fuzzy system.
I create a function for fitness:
function y = gafuzzy(x)
global FISsys
global allData
global realResult
FISsys = readfis('aCAess.fis');
allData = importdata('ab.mat');
realResult = importdata('ad.mat');
FISsys.input(1,1).mf(1,1).params = [x(1) x(2) x(3)];
FISsys.input(1,1).mf(1,2).params = [x(4) x(5) x(6)];
FISsys.input(1,2).mf(1,1).params = [x(7) x(8) x(9)];
FISsys.input(1,2).mf(1,2).params = [x(10) x(11) x(12)];
FISsys.output.mf(1,1).params = [x(13) x(14) x(15)];
FISsys.output.mf(1,2).params = [x(16) x(17) x(18)];
c = evalfis(allData,FISsys);
e=sum(abs(c-realResult));
y = e;
end
And A[15*18] matrix for linear inequalities is :
A = [1,-1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0;
0,1,-1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0;
0,0,0,1,-1,0,0,0,0,0,0,0,0,0,0,0,0,0;
0,0,0,0,1,-1,0,0,0,0,0,0,0,0,0,0,0,0;
0,0,0,0,0,0,1,-1,0,0,0,0,0,0,0,0,0,0;
0,0,0,0,0,0,0,1,-1,0,0,0,0,0,0,0,0,0;
0,0,0,0,0,0,0,0,0,1,-1,0,0,0,0,0,0,0;
0,0,0,0,0,0,0,0,0,0,1,-1,0,0,0,0,0,0;
0,0,0,0,0,0,0,0,0,0,0,0,1,-1,0,0,0,0;
0,0,0,0,0,0,0,0,0,0,0,0,0,1,-1,0,0,0;
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,-1,0;
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,-1;
0,1,0,0,-1,0,0,0,0,0,0,0,0,0,0,0,0,0;
0,0,0,0,0,0,0,1,0,0,-1,0,0,0,0,0,0,0;
0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,-1,0]
and b[15*1] vector is:
b = [0;0;0;0;0;0;0;0;0;0;0;0;0;0;0]
but when I run GA, I get this error:
Illegal parameters in fisTriangleMf() --> a > b
why?
Generally, in the triangle MF the first number, here a (shows the left vertex) should be smaller than the second number, here b (the top vertex). So you can have a triangle MF like [-1 0 1] but it cannot be like [0 -1 1].
in your code, I assume sometimes you don't satisfy the inequality in one of those places:
[x(1) < x(2) < x(3)];
[x(4) < x(5) < x(6)];
[x(7) < x(8) < x(9)];
[x(10) < x(11) < x(12)];
....
if the program is randomizing these values, you can bound them in your code easily by checking and replacing, for instance:
if x(1) >= x(2)
tmp = x(1);
x(1) = x(2);
x(2) = tmp;
end