Fmincon: Unable to perform assignment because the left and right sides have a different number of elements.

Hi, I could use some help please.

I’m working with the matlab optimizer to optimize a x0 speed profile (obtained by simulating a simulink model) respecting nonlinear constraints and a multi-objective J cost function (J = J1+J2+J3). To do this I use FMINCON-SQP-Forward.

x0 is of size 238 but I "extend" it to a fixed size of 300 samples (entering the last values at 0) to match it with the simulation time.

The optimizer starts running quietly and performs over 2000 Function Evaluation and 8/9 iterations, constraints improve but are not respected all of them, until it stops in the middle of an iteration, returning this error:

"Unable to perform assignment because the left and right sides have a different number of elements. 
Error in matlab.internal.optimfun.plotfcns.Factory/appendScalars (line 572)
data(ct) = optimValues.(dataFieldNames(ct));
Error in matlab.internal.optimfun.plotfcns.Factory>@(optimValues)matlab.internal.optimfun.plotfcns.Factory.appendScalars(optimValues,dataFieldNames) (line 753) 
calculationFcn = @(optimValues) matlab.internal.optimfun.plotfcns.Factory.appendScalars(optimValues, dataFieldNames); 
Error in matlab.internal.optimfun.plotfcns.AbstractPlotFunction/update (line 75) 
data = this.CalculationFcn(optimValues, varargin{:}); 
Error in optimplotfirstorderopt (line 28) thePlot.update(optimValues); 
Error in callAllOptimPlotFcns (line 137) state(i) = feval(plotNames{i},x,optimvalues,flag,varargin{:}); 
Error in sqpInterface Error in fmincon (line 672) [X,FVAL,EXITFLAG,OUTPUT,LAMBDA,GRAD,HESSIAN] = sqpInterface(funfcn,X,full(A),full(B),full(Aeq),full(Beq), ... "

I don’t understand why:

1) Ceq = []

2) C = column vector with fixed dimension

3) x0 has a fixed dimension equal to simulation_time and the optimizer never stops before

4) I tried to disable the optimizer plots but nothing changes

5) The components of J are always defined and are scalar

[Walter Roberson]

Is your simulation calling ode45 or similar functions?

[Gabriele]

yes, ode45

[Star Strider]

One option may be to use the simulation time as the ‘tspan’ argument to yourr ode45 call.

If the ODE integrators encounter a singularity, they will stop the integration. This results in the integrated result not matching the size of the previous arrays, or if you are doing parameter estimation, the array it is being compared to, throwing that error.

The solution is to find the cause of the singularity, then see what you can do to avoid it. In my experience, the genetic algorithm (ga) function deals with this by simply going to the next step rather than stopping (although it will throw a Warning), so one option may be to use ga initially, and then use the gradient descent function to fine-tune that result.

[Gabriele]

Sorry, maybe I didn’t say it wrong.

The values to which fmincon refers for operation are acquired directly from a simulink model which spins n-times for each iteration. The simulink model has ode45 as solver. The simulation time is fixed at 300 (simulation_time).

What you said is really helpful because, to try not to get the error, I added some lines of "assert()" in the code and actually receive the warning:

"Warning: Inequality constraints returned NaN or Inf."

The ga algorithm is actually a possible implementation, but at the moment I don’t have enough x0 to use and the problem is complex and the "ga" takes too long. Also use FMINCON for a constraints arrangement and cost function.

[Walter Roberson]

It is likely that at some point your simulink model is ending early, probably complaining about an inability to integrate past a certain time. When this happens, the data returned by simulink will be the data up to the time the integration ended -- which will probably be shorted than the data that was returned before that point.

You need to avoid the situation where the integration ends early. Or at the very least you need to detect the situation and stop looping yourself.

[Gabriele]

All right, thank you for your help. I try to understand the reason for this early termination