// Wake Code - Matlab
// Rasmus Christensen
// Control and Automation, Aalborg University
// N_turb = number of turbines.
// X_turb = x-position of turbine.
// Y_turb = y-position of turbine.
#endregion
internal static void ROTATE_corrd(out ILArray<double> out_x, out ILArray<double> out_y, ILArray<double> xTurb, ILArray<double> yTurb, double rotA)
{
#region "Used variables declaration"
ILArray<double> x_out;
ILArray<double> y_out;
int i;
#endregion
x_out = zeros(1, length(xTurb)); // Initialization x-coordinates
y_out = zeros(1, length(yTurb)); // Initialization y-coordinates
rotA = rotA * pi / 180; // Conversion to radians
for (i = 1; i <= length(xTurb); i++)
{
x_out._(i, '=', xTurb._(i) * cos(rotA) - xTurb._(i) * sin(rotA));
y_out._(i, '=', xTurb._(i) * sin(rotA) + yTurb._(i) * cos(rotA));
}
if (min_(x_out) < 0) // Moves the x-points if these are negative.
{
x_out = x_out + 500 + abs(min_(x_out));
}
if (min_(y_out) < 0) // Moves the y-points if these are negative.
{
y_out = y_out + 500 + abs(min_(y_out));
}
out_x = x_out;
out_y = y_out;
}
// Wake Code - Matlab
// Rasmus Christensen
// Control and Automation, Aalborg University
// N_turb = number of turbines.
// X_turb = x-position of turbine.
// Y_turb = y-position of turbine.
#endregion
internal static void ROTATE_corrd(out ILArray <double> out_x, out ILArray <double> out_y, ILArray <double> xTurb, ILArray <double> yTurb, double rotA)
{
#region "Used variables declaration"
ILArray <double> x_out;
ILArray <double> y_out;
int i;
#endregion
x_out = zeros(1, length(xTurb)); // Initialization x-coordinates
y_out = zeros(1, length(yTurb)); // Initialization y-coordinates
rotA = rotA * pi / 180; // Conversion to radians
for (i = 1; i <= length(xTurb); i++)
{
x_out._(i, '=', xTurb._(i) * cos(rotA) - xTurb._(i) * sin(rotA));
y_out._(i, '=', xTurb._(i) * sin(rotA) + yTurb._(i) * cos(rotA));
}
if (min_(x_out) < 0) // Moves the x-points if these are negative.
{
x_out = x_out + 500 + abs(min_(x_out));
}
if (min_(y_out) < 0) // Moves the y-points if these are negative.
{
y_out = y_out + 500 + abs(min_(y_out));
}
out_x = x_out;
out_y = y_out;
}
//P_ref is a vector of power refenreces for tehe wind turbine with dimension 1xN
//v_nac is a vector of wind speed at each wind turbine with dimension 1xN
//P_demand is a scale of the wind farm power demand.
//parm is a struct of wind turbine parameters e.g. NREL5MW
#endregion
internal static void powerDistributionControl(out ILArray <double> P_ref, out ILArray <double> P_a, double[] v_nac, double P_demand, WindTurbineParameters parm)
{
#region "Used variables declaration"
double rho;
ILArray <double> R;
ILArray <double> rated;
ILArray <double> Cp;
int i;
#endregion
rho = parm.rho; //air density for each wind turbine(probably the same for all)
R = parm.radius.C; //rotor radius for each wind turbine(NREL.r=63m)
rated = parm.rated.C; //Rated power for each wind turbine(NREL.Prated=5MW)
Cp = parm.Cp.C; // Max cp of the turbines for each wind turbine(NREL.Cp.max=0.45)
P_a = zeros(parm.N, 1);
P_ref = zeros(parm.N, 1);
// Compute available power at each turbine
for (i = 1; i <= parm.N; i++)
{
P_a._(i, '=', min_(__[rated._(i), (pi / 2) * rho * _p(R._(i), 2) * _p(v_nac[i - 1], 3) * Cp._(i)]));
}
var sum_P_a_ = sum_(P_a);
//Distribute power according to availibility
for (i = 1; i <= parm.N; i++)
{
if (P_demand < sum_P_a_)
{
P_ref._(i, '=', max_(__[0, min_(__[rated._(i), P_demand * P_a._(i) / sum_P_a_])]));
}
else
{
P_ref._(i, '=', P_a._(i));
}
}
}
//P_ref is a vector of power refenreces for tehe wind turbine with dimension 1xN
//v_nac is a vector of wind speed at each wind turbine with dimension 1xN
//P_demand is a scale of the wind farm power demand.
//parm is a struct of wind turbine parameters e.g. NREL5MW
#endregion
internal static void powerDistributionControl(out ILArray<double> P_ref, out ILArray<double> P_a, double[] v_nac, double P_demand, WindTurbineParameters parm)
{
#region "Used variables declaration"
double rho;
ILArray<double> R;
ILArray<double> rated;
ILArray<double> Cp;
int i;
#endregion
rho = parm.rho; //air density for each wind turbine(probably the same for all)
R = parm.radius.C; //rotor radius for each wind turbine(NREL.r=63m)
rated = parm.rated.C; //Rated power for each wind turbine(NREL.Prated=5MW)
Cp = parm.Cp.C; // Max cp of the turbines for each wind turbine(NREL.Cp.max=0.45)
P_a = zeros(parm.N, 1);
P_ref = zeros(parm.N, 1);
// Compute available power at each turbine
for (i = 1; i <= parm.N; i++)
{
P_a._(i, '=', min_(__[ rated._(i), (pi / 2) * rho * _p(R._(i), 2) * _p(v_nac[i - 1], 3) * Cp._(i) ]));
}
var sum_P_a_ = sum_(P_a);
//Distribute power according to availibility
for (i = 1; i <= parm.N; i++)
{
if (P_demand < sum_P_a_)
{
P_ref._(i, '=', max_(__[ 0, min_(__[ rated._(i), P_demand * P_a._(i) / sum_P_a_ ]) ]));
}
else
{
P_ref._(i, '=', P_a._(i));
}
}
}
// Wake Code - Matlab
// Rasmus Christensen
// Control and Automation, Aalborg University
#endregion
internal static void Turb_centr_coord(out ILArray<int> output, int nTurb, int iMax, ILArray<double> x, ILArray<double> xTurb, int gridRes)
{
#region "Used variables declaration"
ILArray<int> xxcTurb;
int i;
int ii;
#endregion
xxcTurb = zeros_(1, nTurb);
for (i = 1; i <= nTurb; i++)
{
for (ii = 1; ii <= iMax - 1; ii++)
{
if (abs(x._(ii)) <= abs(xTurb._(i)) && abs(xTurb._(i)) < abs(x._(ii + 1)))
{
xxcTurb._(i, '=', ii * sign(xTurb._(i)));
break;
}
}
}
output = xxcTurb;
}
// Wake Code - Matlab
// Rasmus Christensen
// Control and Automation, Aalborg University
#endregion
internal static void Turb_centr_coord(out ILArray <int> output, int nTurb, int iMax, ILArray <double> x, ILArray <double> xTurb, int gridRes)
{
#region "Used variables declaration"
ILArray <int> xxcTurb;
int i;
int ii;
#endregion
xxcTurb = zeros_(1, nTurb);
for (i = 1; i <= nTurb; i++)
{
for (ii = 1; ii <= iMax - 1; ii++)
{
if (abs(x._(ii)) <= abs(xTurb._(i)) && abs(xTurb._(i)) < abs(x._(ii + 1)))
{
xxcTurb._(i, '=', ii * sign(xTurb._(i)));
break;
}
}
}
output = xxcTurb;
}
internal static void wakeCalculation(out ILArray <double> v_nac, ILArray <double> Ct, int i, ILArray <double> wind)
{
#region "Original function comments"
//% v_nac = WAKECALCULATION(Ct,i,wind)
//This function calculates the wake
//Currently it is a very very simplified wake calculation. It just serves as
//a placeholder for a correct wake calculation that will come later
#endregion
#region "Used variables declaration"
ILArray <double> scaling;
#endregion
scaling = linspace(0.5, 0.9, length(Ct));
v_nac = scaling * wind._(i, 2);
}
internal static void wakeCalculation(out ILArray<double> v_nac, ILArray<double> Ct, int i, ILArray<double> wind)
{
#region "Original function comments"
//% v_nac = WAKECALCULATION(Ct,i,wind)
//This function calculates the wake
//Currently it is a very very simplified wake calculation. It just serves as
//a placeholder for a correct wake calculation that will come later
#endregion
#region "Used variables declaration"
ILArray<double> scaling;
#endregion
scaling = linspace(0.5, 0.9, length(Ct));
v_nac = scaling * wind._(i, 2);
}
//% v_nac = WAKECALCULATION(Ct,i,wind)
// RLC, Aalborg
// The below is based on the .F90 code developed by ?, and will give a
// better estimate of the actual wake the individual turbines experience.
#endregion
internal static void wakeCalculationsRLC(out ILArray <double> vNac, double dTurb, int nTurb, double kWake, ILArray <double> x, int gridX, int gridY, ILArray <double> yOrder, double dy, ILArray <int> xTurbC, ILArray <int> yTurbC, ILArray <double> Ct, ILArray <double> wField, double[] vHub, WindTurbineParameters parm, SimParm simParm)
{
#region "Used variables declaration"
double[,] Velocity;
int j;
#endregion
// Velocity Computation
Compute_Vell(out Velocity, yOrder, xTurbC, yTurbC, x, wField, vHub, kWake, gridX, gridY, nTurb, dTurb, Ct, dy);
// Extracting the individual Nacelle Wind Speeds from the wind velocity matrix.
//Velocity = Velocity';
vNac = zeros(nTurb, 1);
for (j = 1; j <= length(xTurbC); j++)
{
vNac._(j, '=', Velocity[yTurbC._(j) - 1, xTurbC._(j) - 1]);
}
}
//% v_nac = WAKECALCULATION(Ct,i,wind)
// RLC, Aalborg
// The below is based on the .F90 code developed by ?, and will give a
// better estimate of the actual wake the individual turbines experience.
#endregion
internal static void wakeCalculationsRLC(out ILArray<double> vNac, double dTurb, int nTurb, double kWake, ILArray<double> x, int gridX, int gridY, ILArray<double> yOrder, double dy, ILArray<int> xTurbC, ILArray<int> yTurbC, ILArray<double> Ct, ILArray<double> wField, double[] vHub, WindTurbineParameters parm, SimParm simParm)
{
#region "Used variables declaration"
double[,] Velocity;
int j;
#endregion
// Velocity Computation
Compute_Vell(out Velocity, yOrder, xTurbC, yTurbC, x, wField, vHub, kWake, gridX, gridY, nTurb, dTurb, Ct, dy);
// Extracting the individual Nacelle Wind Speeds from the wind velocity matrix.
//Velocity = Velocity';
vNac = zeros(nTurb, 1);
for (j = 1; j <= length(xTurbC); j++)
{
vNac._(j, '=', Velocity[yTurbC._(j) - 1, xTurbC._(j) - 1]);
}
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// DESCRIPTION:
// RLC - 8/9/2014, interpolation, for getting an accurate CP/CT value. The
// reason for a standalone script, is that the built in MATLAB script
// (interp) includes a lot of redundancy checks that are not necessary, and
// not feasible for embedded solutions.
// Based on the NREL5MW Turbine.
// Uses Linear Polynomial Extrapolation, to get a value for CP, given a Beta
// (Revolutional speed) and Lambda (Tip-Speed-Ratio) of the Turbine.
// The interpolation is computed as:
// y = y0 + (y1 - y0)*(x-x0)/(x1-x0)
//
// Beta is the revolutional entry.
// Lambda is the TSR entry ratio.
// turbineTable defines the table to lookup in.
// negYes defines wether the CP value should be allowed to be negative.
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#endregion
internal static void interpTable(out double interpValue, double Beta, double Lambda, ILArray <double> table, ILArray <double> turbineTableBeta, ILArray <double> turbineTableLambda, bool negYes)
{
#region "Used variables declaration"
ILArray <double> turbineTable;
int sizeBt;
int sizeLa;
int Bt0;
int Bt1;
int La0;
int La1;
ILArray <double> tableLookup;
ILArray <double> lambdaIntervals;
double betaIntervals;
#endregion
//% Setup, loads the table, and stores it.
//persistent turbineTable tableLoad
//if isempty(tableLoad)
turbineTable = table.C;
// tableLoad = 1;
//end
size(out sizeBt, out sizeLa, turbineTable);
//% Index Interpolation
// Finds the beta-point of the interpolation.
min(out Bt0, abs(turbineTableBeta - Beta)); // Determines the index of the closest point.
if (Beta > turbineTableBeta._(Bt0))
{
if (Bt0 == sizeBt) //length(turbineTableBeta)
{
Bt1 = Bt0;
Bt0 = Bt0 - 1;
}
else
{
Bt1 = Bt0 + 1;
}
}
else
{
if (Bt0 == 1)
{
Bt1 = Bt0 + 1;
}
else
{
Bt1 = Bt0;
Bt0 = Bt1 - 1;
}
}
// Finds the Lambda-point of the interpolation.
min(out La0, abs(turbineTableLambda - Lambda)); // Determines the index of the closest point.
if (Lambda > turbineTableLambda._(La0))
{
if (La0 == sizeLa) //length(turbineTableLambda)
{
La1 = La0;
La0 = La1 - 1;
}
else
{
La1 = La0 + 1;
}
}
else
{
if (La0 == 1)
{
La1 = La0 + 1;
}
else
{
La1 = La0;
La0 = La1 - 1;
}
}
//% Table Interpolation
// Table lookup using indexes obtained previously:
tableLookup = __[turbineTable._(Bt0, La0), turbineTable._(Bt0, La1), ';',
turbineTable._(Bt1, La0), turbineTable._(Bt1, La1)];
// Interpolating, using the Lambda values first, then the Betas.
lambdaIntervals = __[((tableLookup._(1, 2) - tableLookup._(1, 1)) / (turbineTableLambda._(La1) - turbineTableLambda._(La0))) * (Lambda - turbineTableLambda._(La0)) + tableLookup._(1, 1), ';',
((tableLookup._(2, 2) - tableLookup._(2, 1)) / (turbineTableLambda._(La1) - turbineTableLambda._(La0))) * (Lambda - turbineTableLambda._(La0)) + tableLookup._(1, 2)];
// Interpolation, using the Beta values (using the intervales computed above).
betaIntervals = ((lambdaIntervals._(2) - lambdaIntervals._(1)) / (turbineTableBeta._(Bt1) - turbineTableBeta._(Bt0))) * (Beta - turbineTableBeta._(Bt0)) + lambdaIntervals._(1);
//% Negativity Handling
// If the negYes value is set as true in the system, the functino will only
// give an absolute value of Ct and Cp - have to be modified so it does this
// correctly.
if (negYes)
{
interpValue = 0;
}
else
{
interpValue = betaIntervals;
}
}
// Wake Simulation
// (C) Rasmus Christensen
// Control and Automation, Aalborg University 2014
// Compute the velocity in front of each wind-turbine, with respect to the wind input.
#endregion
internal static void Compute_Vell(out double[,] vel_output, ILArray <double> yTurb, ILArray <int> xTurbC, ILArray <int> yTurbC, ILArray <double> x, ILArray <double> wField, double[] Uhub, double kWake, int iMax, int jMax, int nTurb, double dTurb, ILArray <double> Ct, double dy)
{
#region "Used variables declaration"
double[,] vell_i;
ILArray <double> shadow;
double r0;
double nk;
int k;
int J;
double SS;
double SS0;
int i;
double RR_i;
double Dij;
double Alpha_i;
double Alpha_k;
double Area;
int ii;
double rrt;
double nj;
int jjMin;
int jjMax;
#endregion
//vell_i = wField.C;
vell_i = new double[wField.Size[0], wField.Size[1]];
for (var i0 = 0; i0 < vell_i.GetLength(0); i0++)
{
for (var i1 = 0; i1 < vell_i.GetLength(1); i1++)
{
vell_i[i0, i1] = wField.GetValue(i0, i1);
}
}
shadow = zeros(1, nTurb);
r0 = 0.5 * dTurb;
nk = 2 * ceil(dTurb / dy);
for (k = 1; k <= nTurb; k++)
{
J = 0;
SS = 0;
SS0 = pi * r0 * r0;
for (i = 1; i <= k - 1; i++)
{
RR_i = r0 + kWake * (x._(xTurbC._(k)) - x._(xTurbC._(i)));
Dij = abs(yTurb._(i) - yTurb._(k));
if ((RR_i >= (r0 + Dij)) || (Dij <= dy))
{
SS = SS + ((r0 * r0) / (RR_i * RR_i));
}
else
{
if (RR_i >= (r0 + Dij) && Dij > dy)
{
J = J + 1;
Alpha_i = acos((RR_i * RR_i) + (Dij * Dij) - (r0 * r0) / (2 * RR_i * Dij));
Alpha_k = acos(((r0 * r0) + (Dij * Dij) - (RR_i * RR_i)) / (2 * r0 * Dij));
AArea(out Area, RR_i, r0, Dij);
shadow._(J, '=', (Alpha_i * (RR_i * RR_i) + Alpha_k * (r0 * r0)) - 2 * Area);
SS = SS + ((shadow._(J)) / SS0) * ((r0 * r0) / (RR_i * RR_i));
}
}
}
for (ii = xTurbC._(k); ii <= iMax; ii++)
{
rrt = r0 + kWake * (x._(ii) - x._(xTurbC._(k)));
nj = ceil(rrt / dy);
jjMin = floor_(max(1, yTurbC._(k) - nj));
jjMax = ceil_(min(jMax, yTurbC._(k) + nj));
for (var j = jjMin; j <= jjMax; j++)
{
if (((-vell_i[ii - 1, j - 1] + Uhub[k - 1]) > 0) && (ii > xTurbC._(k) + nk))
{
vell_i[ii - 1, j - 1] = min(vell_i[ii - 2, j - 1], Uhub[k - 1] + Uhub[k - 1] * (sqrt(1 - Ct._(k)) - 1) * ((r0 * r0) / (rrt * rrt)) * (1 - (1 - sqrt(1 - Ct._(k))) * SS));
vell_i[ii - 1, j - 1] = max(0, vell_i[ii - 1, j - 1]);
}
else
{
vell_i[ii - 1, j - 1] = (Uhub[k - 1] + Uhub[k - 1] * (sqrt(1 - Ct._(k)) - 1) * (r0 / rrt) * (r0 / rrt)) * (1 - (1 - sqrt(1 - Ct._(k))) * SS);
vell_i[ii - 1, j - 1] = max(0, vell_i[ii - 1, j - 1]);
}
}
}
}
vel_output = vell_i;
}
internal static void NowCastWFPFunc(out string outMethod, out ILArray<double> outTime, out ILArray<double> outX, out ILArray<double> outXhmsAll, out int outXhmsAllTimeOffset, out int outXhmsLLength, out int outXhmsUOffset, ILArray<double> Data, double TPredict = TPredictDef, string Method = MethodDef, int r = rDef, double Ts = TsDef)
{
#region "Original function comments"
//Nowcasting with model based on total wind farm power
//
// NowCastWFPFunc(Data,TPredict,Method,r,Ts)
//
// Data : Total wind farm power
// TPredict : Time for starting multi step prediction. If TPredict<1 it
// is assumed a fraction of the end time (default 0.5)
// Method : 'AR(1)' or 'Persistence' only first letter count (default 'a')
// r : Decimation with a moving average of order r (default 1)
// Ts : Sampling time (default 0.1)
//
// External input: None
// Time-stamp: <2014-10-17 14:09:40 tk>
// Version 1: Initial version
// Torben Knudsen
// Aalborg University, Dept. of Electronic Systems, Section of Automation
// and Control
// E-mail: [email protected]
#endregion
#region "Used variables declaration"
int IMT;
double q0;
double TauLambdaLRel;
double Lambda0;
double TauLambdaInf;
double TimeScaling;
//string TitleStr;
int NS;
ILArray<double> T;
ILArray<double> TimePlot;
double NWT;
double NomWFP;
ILArray<double> PWF;
double MinWFP;
double TauLambdaL;
double LambdaL;
double LambdaInf;
int TPS;
ILArray<double> Theta;
ILArray<double> Sigma;
ILArray<double> Xh;
ILArray<double> Lambda;
ILArray<int> Time;
ILArray<double> Res;
int TPSEst;
double A;
double B;
ILArray<double> xhms;
ILArray<double> covxhms;
double aux;
int i;
ILArray<double> q;
double xh;
double xt;
double dt;
ILArray<double> sigmaxhms;
ILArray<double> ConIntAWFP;
int NS10Min;
int NSOneHour;
#endregion
//% setting up inputs
//TsDef = 0.1;
//rDef = 1;
//MethodDef = "a";
//TPredictDef = 0.5;
//if nargin < 5; Ts= []; end;
//if nargin < 4; r= []; end;
//if nargin < 3; Method= []; end;
//if nargin < 2; TPredict= []; end;
//if nargin < 1; error('Error TK: To few input arguments'); end;
//if isempty(r); r= rDef; end;
//if isempty(TPredict); TPredict= TPredictDef; end;
//if isempty(Ts); Ts= TsDef; end;
//if isempty(Method); Method= MethodDef; end;
//% Parameters
IMT = 1; // Include measurement time;
q0 = 0;
TauLambdaLRel = 0.1; // TauLambdaL= 10% of the samples
Lambda0 = 0.5; // Initial Lambda
TauLambdaInf= 600; // TauLambdaInf= 10min
TimeScaling = 1.0 / 3600; // From seconds to hours
//% Initialization
// Use Offwind simulation data made by Rasmus 2014-10-14
// The format is:
// [time sumPower sumRef sumAvai] as a matrix NS x 4. Power i MW
// 48 WT are simulated. Data for individual WT are also found e.g. in
// Power, P_ref, PA, beta etc
if (strncmpi(Method, "a", 1))
{
Method = "AR(1)";
}
else
{
Method = "Persistence";
}
//TitleStr = "Nowcasting with " + Method + " model based on total wind farm " //...
// + "power, Offwind simulation";
Data = Data[_(':')];
NS = size(Data, 1); // Number of samples
NS = max_(find(Data > 0)); // Number of samples;
Data = Data[_(1, ':', NS)]; // Limit the data
T = Ts * (_c(1, NS)).T;
TimePlot = T * TimeScaling; // Time in hours
NWT = 48;
NomWFP = NWT * 5e6 * 1e-6; // Power in MW
PWF = Data; // Total Power in MW
if (r > 1)
{
DecimateWMA(out PWF, out _ILArray_double, PWF, r);
NS = size(PWF, 1); // Number of samples
Ts = Ts * r;
T = Ts * (_c(1, NS)).T;
TimePlot = T * TimeScaling; // Time in hours
}
MinWFP = 0; // For real WFs
//% Definitions etc.
// Calculate Lamba* from TauLambda* and dt
TauLambdaL = TauLambdaLRel * (T._(end) - T._(1));// TauLambdaL= 10% of the samples
LambdaL = exp(-Ts / TauLambdaL);
LambdaInf = exp(-Ts / TauLambdaInf);
//% Algorithm
// Initialization
// Prediction from time in TPredict
// if TPredict is a fraction calculate TPredict
if (TPredict < 1)
{
TPredict = round(TPredict * T._(end) / Ts) * Ts;
}
if (TPredict < TauLambdaInf)
{
warning(__[ "TK: Prediction time is so small that the estimator/predictor ", //...
"might not have converged yet" ]);
}
TPS = min_(find(T >= TPredict)); // Use time from measurements
// Multi step prediction
if (strncmpi(Method, "a", 1))
{
// ARX1 version;
// Recursive parameter estimation
RLSMARX1(out Theta, out Sigma, out Xh, out Lambda, out Time, out _ILArray_double, out _ILArray_double, out _ILArray_double, PWF, 1, LambdaInf);
Res = __[ TimePlot[_(Time)], PWF[_(Time)], Xh, Sigma, Lambda ];
Res = __[ nan(Time._(1) - 1, size(Res, 2)), ';', Res ];
// Notice that length of Time is shorter than length of T if batch RLS
// start is used so TPSEst < TPS in that case
TPSEst = min_(find(_dbl(Time) * Ts >= TPredict)); // Use time from estimates
// Parameter values must be taken for index TPSEst
A = Theta._(TPSEst, 1);
if (abs(A) > 1)
{
warning(__[ "TK: Unstable pole, max(abs(eig)): ", num2str(abs(A)) ]);
}
B = Theta._(TPSEst, 2);
xhms = zeros(NS - TPS, 1);
covxhms = zeros(NS - TPS, 1);
xhms._(1, '=', Xh._(TPSEst + 1));
covxhms._(1, '=', Sigma._(TPSEst + 1));
aux = Sigma._(TPSEst + 1);
for (i = 2; i <= NS - TPS; i++)
{
xhms._(i, '=', A * xhms._(i - 1) + B);
aux = A * aux * A.T();
covxhms._(i, '=', covxhms._(i - 1) + aux);
}
// Prepend xhms with the present measurement so the plot clearly indicates
// the time for the measurement
}
else
{
// Persistence version;
// Initialization
Lambda = Lambda0;
q = q0;
xh = PWF._(1);
Res = __[ T._(1), PWF._(1), xh, q, Lambda0 ];
// Recursive estimation of incremental covariance
for (i = 2; i <= NS; i++)
{
xt = PWF._(i) - xh;
dt = T._(i) - T._(i - 1);
q = _m(Lambda, '*', q) + (1 - Lambda) * _p(xt, 2) / dt;
Res = __[ Res, ';', __[ T._(i), PWF._(i), xh, q, Lambda ] ];
Lambda = LambdaL * Lambda + (1 - LambdaL) * LambdaInf;
xh = PWF._(i);
}
Res[_(':'), _(1)] = Res[_(':'), _(1)] * TimeScaling;
// Persistence version;
xhms = Res._(TPS + 1, 3) * ones(NS - TPS, 1);
covxhms = Res._(TPS + 1, 4) * Ts * ((_c(1, NS - TPS)).T);
}
if (IMT != 0)
{
xhms = __[ PWF._(TPS), ';', xhms ];
covxhms = __[ 0, ';', covxhms ];
}
sigmaxhms = sqrt(covxhms);
ConIntAWFP = _m(xhms, '*', __[ 1, 1, 1 ]) + _m(sigmaxhms, '*', __[ -2, 0, 2 ]);
ConIntAWFP = min(max(ConIntAWFP, MinWFP), NomWFP); // Limit confidence limits
// Plot results
// Plot actual as black solid and lower, prediction and upper confidence
// limits as red, green and blue solid for 10 min, dashed for one hour and
// dotted for the rest.
//figure;
NS10Min = min(NS - TPS - 1, round_(600.0 / Ts));
NSOneHour = min(NS - TPS - 1, round_(3600.0 / Ts));
//set(gcf, "defaultaxescolororder", ILMath.eye(3, 3));
outMethod = Method;
outTime = TimePlot;
outX = Res[_(':'), _(2)];
outXhmsAll = ConIntAWFP;
outXhmsAllTimeOffset = (TPS - IMT);
outXhmsLLength = (NS10Min + IMT);
outXhmsUOffset = (NSOneHour + IMT);
//plot(
// TimePlot, Res[ILMath.full, 2 - 1], 'k',///...
// TimePlot[ILMath.r(TPS + 1 - IMT - 1, TPS + NS10Min - 1)], ConIntAWFP[ILMath.r(1 - 1, NS10Min + IMT - 1), ILMath.full],//...
// TimePlot[ILMath.r(TPS + NS10Min + 1 - 1, TPS + NSOneHour - 1)], ConIntAWFP[_a(NS10Min + 1, 1, NSOneHour) + IMT - 1, ILMath.full], "--",//...
// TimePlot[ILMath.r(TPS + NSOneHour + 1 - 1, ILMath.end)], ConIntAWFP[ILMath.r(NSOneHour + 1 + IMT - 1, ILMath.end), ILMath.full], ':');
//title(TitleStr);
//Legend= {'x' 'xhmsL' 'xhms' 'xhmsU'};
//legend(Legend);
//grid('on');
}
private static void RLSMARX1(out ILArray<double> Theta, out ILArray<double> Sigma, out ILArray<double> Xh, out ILArray<double> Lambda, out ILArray<int> Time, out ILArray<double> AByTime, out ILArray<double> BByTime, out ILArray<double> SigmaByTime, ILArray<double> X, ILArray<double> U = null, ILArray<double> lambda = null, ILArray<double> theta0 = null, ILArray<double> sigma0 = null, int Plot = PlotDef)
{
#region "Original function comments"
//RLS calculates recursive least squares parameter estimates corresponding
// to the output input data X U and the model x(n+1)= A*x(n)+B*u(n)+e(n)
// <=> x(n+1)'= [x(n)' u(n)']*[A'; B']+e(n)
// Duble exponential forgetting can be used where lamba increases
// exponentielly from lamba0 to lambda.
//
// [Theta,Sigma,Xh,Lambda,Time]= RLSMARX1(X,U,lambda,theta0,sigma0)
//
// X: Output Matrix with size number of samples * number of channels
// U: Input Matrix with size number of samples * number of channels
// zero scalar/matrix corresponds to no input, ones corresponds to estimating
// a mean value parameter. (default ones)
// lambda : Forgetting factor (default 0.99) if lambda= [lambda lambda0]
// the forgetting factor increases exponentially from lambda0
// to lambda.
// theta0 : Initial parameter estimate. Notice that theta0= [A0';B0'];
// sigma0 : Initial covvariance estimate.
// If no initial parameters are specifyed a offline/batch
// estimate is used for the start.
// Plot : If 1/true plot informative plots (default: false)
//
// External input:
// Time-stamp: <2014-10-17 11:44:27 tk>
// Version 1: 2014-10-01 app.
// Version 2: 2014-10-02 12:53:07 tk Included LS/offline/batch startup
// Version 3: 2014-10-07 13:57:54 tk Included additional output and
// plotting
// Torben Knudsen
// Aalborg University, Dept. of Electronic Systems, Section of Automation
// and Control
// E-mail: [email protected]
#endregion
#region "Used variables declaration"
double TauLambdaLRel;
double lambdaInf;
double lambda0;
double IniNumSampFrac;
int N;
int n;
int m;
double TauLambdaL;
double lambdaL;
ILArray<double> sigma;
ILArray<double> theta;
int NIni;
ILArray<double> XX;
ILArray<double> YY;
ILArray<double> R;
ILArray<double> Epsilon;
ILArray<double> xh;
ILArray<double> Res;
int i;
ILArray<double> phi;
ILArray<double> epsilon;
#endregion
//% setting up inputs
//UDef = 1;
//lambdaDef = 0.99;
//theta0Def = 0;
//sigma0Def = 0;
//PlotDef = 0;
//if (nargin < 6) { Plot = []; }
if (sigma0 == null) { sigma0 = __[' ']; }
if (theta0 == null) { theta0 = __[' ']; }
if (lambda == null) { lambda = __[' ']; }
if (U == null) { U = __[' ']; }
//if (nargin < 1) { error('Error TK: To few input arguments'); }
if (isempty(U)) { U = UDef; }
//if (isempty(Plot)) { Plot = PlotDef; }
if (isempty(lambda)) { lambda = lambdaDef; }
if (isempty(theta0)) { theta0 = theta0Def; }
if (isempty(sigma0)) { sigma0 = sigma0Def; }
//% Parameters
TauLambdaLRel = 0.1; // TauLambdaL= 10% of the samples
lambdaInf = lambda._(1);
if (length(lambda) > 1) // Initial lambda;
{
lambda0 = lambda._(2);
}
else
{
lambda0 = lambdaInf;
}
IniNumSampFrac = 0.1; // Number of samples to use for init
//% Definitions etc.
N = size(X, 1);
n = size(X, 2);
m = size(U, 2);
// Calculate Lamba* from TauLambda* assuming Ts= 1
TauLambdaL = TauLambdaLRel * N;
lambdaL = exp(-1.0 / TauLambdaL);
if (all(U == 1))
{
U = ones(N, 1);
}
//% Algorithm
// Initialisation
if (all(all(sigma0 == 0))) // No sigma0 specifyed
{
sigma = 0 * eye(n);
}
else
{
sigma = sigma0;
}
if (all(all(theta0 == 0))) // No sigma0 specifyed
{
theta = zeros(n + m, n);
}
else
{
theta = theta0;
}
// Start with a offline estimate if initial parameters are not specifyed
if (all(__[ theta0, ';', sigma0 ] == 0))
{
NIni = max(n + 1, ceil_(N * IniNumSampFrac));
XX = __[ X[_(1, ':', NIni - 1), _(':')], U[_(1, ':', NIni - 1), _(':')] ];
YY = X[_(_c_(1, NIni - 1) + 1), _(':')];
R = _m(XX.T, '*', XX);
theta = _m(_m(inv(R), '*', (XX.T)), '*', YY);
Epsilon = YY - _m(XX, '*', theta);
sigma = _m(Epsilon.T, '*', Epsilon) / NIni;
xh = _m(XX[_(end), _(':')], '*', theta);
}
else
{
NIni = 1;
xh = X[_(1), _(':')];
R = 1e6 * eye(n);
}
lambda = lambda0;
Res = __[ NIni, (theta[_(':')]).T, (sigma[_(':')]).T, xh, lambda ];
AByTime = zeros(n, n, N - NIni + 1);
AByTime[_(':'), _(':'), _(1)] = (theta[_(1, ':', n), _(1, ':', n)]).T;
BByTime = zeros(n, m, N - NIni + 1);
BByTime[_(':'), _(':'), _(1)] = (theta[_(n + 1, ':', n + m), _(1, ':', n)]).T;
SigmaByTime = zeros(n, n, N - NIni + 1);
SigmaByTime[_(':'), _(':'), _(1)] = sigma;
// Recursion
// Model x(n+1)= A*x(n)+B*u(n)+e(n) <=>
// x(n+1)'= [x(n)' u(n)']*[A'; B']+e(n)
for (i = NIni + 1; i <= N; i++)
{
phi = (__[ X[_(i - 1), _(':')], U[_(i - 1), _(':')] ]).T; // phi(i)
R = (lambda._Scalar()) * R + _m(phi, '*', (phi.T)); // R(i)
xh = _m((phi.T), '*', theta); // xh(i)
epsilon = X[_(i), _(':')] - xh; // epsilon(i)
theta = theta + _m(_s(R, '\\', phi), '*', epsilon); // theta(i)
sigma = _m(lambda, '*', sigma) + _m(_m((1 - lambda), '*', (epsilon.T)), '*', epsilon);
Res = __[ Res, ';', __[ i, theta[_(':')].T, sigma[_(':')].T, xh, lambda ] ];
lambda = lambdaL * lambda + (1 - lambdaL) * lambdaInf;
AByTime[_(':'), _(':'), _(i - NIni + 1)] = (theta[_(1, ':', n), _(1, ':', n)]).T; // Index start at 2
BByTime[_(':'), _(':'), _(i - NIni + 1)] = (theta[_(n + 1, ':', n + m), _(1, ':', n)]).T;
SigmaByTime[_(':'), _(':'), _(i - NIni + 1)] = sigma;
}
// Res is organised like
// i theta(:)' sigma(:)' xh lambda with the dimensions
// 1 (n+m)*n n^2 n 1
Time = _int(Res[_(':'), _(1)]);
Theta = Res[_(':'), _(1 + 1, ':', 1 + (n + m) * n)];
Sigma = Res[_(':'), _(1 + 1 + (n + m) * n, ':', 1 + (n + m) * n + _p_(n, 2))];
Xh = Res[_(':'), _(1 + 1 + (n + m) * n + _p_(n, 2), ':', 1 + (n + m) * n + _p_(n, 2) + n)];
Lambda = Res[_(':'), _(1 + 1 + (n + m) * n + _p_(n, 2) + n, ':', end)];
// Do informative plotting and output
//if (Plot)
//{
// figure;
// plot([U X]);
// title('Input and measurements');
// figure;
// plot(Time,reshape(AByTime,n*n,N-NIni+1)');
// title('A parameters');
// figure;
// plot(Time,reshape(BByTime,n*m,N-NIni+1)');
// title('B parameters');
// figure;
// plot(Time,reshape(SigmaByTime,n*n,N-NIni+1)');
// title('Sigma parameters');
// figure;
// plot(Time,Lambda);
// title('Lambda');
// Epsilon= X(Time,:)-Xh;
// Epsilon= Epsilon(100:end,:);
// figure;
// XCorrtkAll(Epsilon);
// figure;
// normplot(Epsilon);
// gridtk('on',figures);
// Lab= ['Final parameters A n*n B n*m Sigma n*n'];
// RowLab= {''};
// ColLab= {''};
// Res= [AByTime(:,:,end) BByTime(:,:,end) SigmaByTime(:,:,end)];
// printmattk(Res,Lab);
//}
}
internal static void NowCastWFPFunc(out string outMethod, out ILArray <double> outTime, out ILArray <double> outX, out ILArray <double> outXhmsAll, out int outXhmsAllTimeOffset, out int outXhmsLLength, out int outXhmsUOffset, ILArray <double> Data, double TPredict = TPredictDef, string Method = MethodDef, int r = rDef, double Ts = TsDef)
{
#region "Original function comments"
//Nowcasting with model based on total wind farm power
//
// NowCastWFPFunc(Data,TPredict,Method,r,Ts)
//
// Data : Total wind farm power
// TPredict : Time for starting multi step prediction. If TPredict<1 it
// is assumed a fraction of the end time (default 0.5)
// Method : 'AR(1)' or 'Persistence' only first letter count (default 'a')
// r : Decimation with a moving average of order r (default 1)
// Ts : Sampling time (default 0.1)
//
// External input: None
// Time-stamp: <2014-10-17 14:09:40 tk>
// Version 1: Initial version
// Torben Knudsen
// Aalborg University, Dept. of Electronic Systems, Section of Automation
// and Control
// E-mail: [email protected]
#endregion
#region "Used variables declaration"
int IMT;
double q0;
double TauLambdaLRel;
double Lambda0;
double TauLambdaInf;
double TimeScaling;
//string TitleStr;
int NS;
ILArray <double> T;
ILArray <double> TimePlot;
double NWT;
double NomWFP;
ILArray <double> PWF;
double MinWFP;
double TauLambdaL;
double LambdaL;
double LambdaInf;
int TPS;
ILArray <double> Theta;
ILArray <double> Sigma;
ILArray <double> Xh;
ILArray <double> Lambda;
ILArray <int> Time;
ILArray <double> Res;
int TPSEst;
double A;
double B;
ILArray <double> xhms;
ILArray <double> covxhms;
double aux;
int i;
ILArray <double> q;
double xh;
double xt;
double dt;
ILArray <double> sigmaxhms;
ILArray <double> ConIntAWFP;
int NS10Min;
int NSOneHour;
#endregion
//% setting up inputs
//TsDef = 0.1;
//rDef = 1;
//MethodDef = "a";
//TPredictDef = 0.5;
//if nargin < 5; Ts= []; end;
//if nargin < 4; r= []; end;
//if nargin < 3; Method= []; end;
//if nargin < 2; TPredict= []; end;
//if nargin < 1; error('Error TK: To few input arguments'); end;
//if isempty(r); r= rDef; end;
//if isempty(TPredict); TPredict= TPredictDef; end;
//if isempty(Ts); Ts= TsDef; end;
//if isempty(Method); Method= MethodDef; end;
//% Parameters
IMT = 1; // Include measurement time;
q0 = 0;
TauLambdaLRel = 0.1; // TauLambdaL= 10% of the samples
Lambda0 = 0.5; // Initial Lambda
TauLambdaInf = 600; // TauLambdaInf= 10min
TimeScaling = 1.0 / 3600; // From seconds to hours
//% Initialization
// Use Offwind simulation data made by Rasmus 2014-10-14
// The format is:
// [time sumPower sumRef sumAvai] as a matrix NS x 4. Power i MW
// 48 WT are simulated. Data for individual WT are also found e.g. in
// Power, P_ref, PA, beta etc
if (strncmpi(Method, "a", 1))
{
Method = "AR(1)";
}
else
{
Method = "Persistence";
}
//TitleStr = "Nowcasting with " + Method + " model based on total wind farm " //...
// + "power, Offwind simulation";
Data = Data[_(':')];
NS = size(Data, 1); // Number of samples
NS = max_(find(Data > 0)); // Number of samples;
Data = Data[_(1, ':', NS)]; // Limit the data
T = Ts * (_c(1, NS)).T;
TimePlot = T * TimeScaling; // Time in hours
NWT = 48;
NomWFP = NWT * 5e6 * 1e-6; // Power in MW
PWF = Data; // Total Power in MW
if (r > 1)
{
DecimateWMA(out PWF, out _ILArray_double, PWF, r);
NS = size(PWF, 1); // Number of samples
Ts = Ts * r;
T = Ts * (_c(1, NS)).T;
TimePlot = T * TimeScaling; // Time in hours
}
MinWFP = 0; // For real WFs
//% Definitions etc.
// Calculate Lamba* from TauLambda* and dt
TauLambdaL = TauLambdaLRel * (T._(end) - T._(1));// TauLambdaL= 10% of the samples
LambdaL = exp(-Ts / TauLambdaL);
LambdaInf = exp(-Ts / TauLambdaInf);
//% Algorithm
// Initialization
// Prediction from time in TPredict
// if TPredict is a fraction calculate TPredict
if (TPredict < 1)
{
TPredict = round(TPredict * T._(end) / Ts) * Ts;
}
if (TPredict < TauLambdaInf)
{
warning(__["TK: Prediction time is so small that the estimator/predictor ", //...
"might not have converged yet"]);
}
TPS = min_(find(T >= TPredict)); // Use time from measurements
// Multi step prediction
if (strncmpi(Method, "a", 1))
{
// ARX1 version;
// Recursive parameter estimation
RLSMARX1(out Theta, out Sigma, out Xh, out Lambda, out Time, out _ILArray_double, out _ILArray_double, out _ILArray_double, PWF, 1, LambdaInf);
Res = __[TimePlot[_(Time)], PWF[_(Time)], Xh, Sigma, Lambda];
Res = __[nan(Time._(1) - 1, size(Res, 2)), ';', Res];
// Notice that length of Time is shorter than length of T if batch RLS
// start is used so TPSEst < TPS in that case
TPSEst = min_(find(_dbl(Time) * Ts >= TPredict)); // Use time from estimates
// Parameter values must be taken for index TPSEst
A = Theta._(TPSEst, 1);
if (abs(A) > 1)
{
warning(__["TK: Unstable pole, max(abs(eig)): ", num2str(abs(A))]);
}
B = Theta._(TPSEst, 2);
xhms = zeros(NS - TPS, 1);
covxhms = zeros(NS - TPS, 1);
xhms._(1, '=', Xh._(TPSEst + 1));
covxhms._(1, '=', Sigma._(TPSEst + 1));
aux = Sigma._(TPSEst + 1);
for (i = 2; i <= NS - TPS; i++)
{
xhms._(i, '=', A * xhms._(i - 1) + B);
aux = A * aux * A.T();
covxhms._(i, '=', covxhms._(i - 1) + aux);
}
// Prepend xhms with the present measurement so the plot clearly indicates
// the time for the measurement
}
else
{
// Persistence version;
// Initialization
Lambda = Lambda0;
q = q0;
xh = PWF._(1);
Res = __[T._(1), PWF._(1), xh, q, Lambda0];
// Recursive estimation of incremental covariance
for (i = 2; i <= NS; i++)
{
xt = PWF._(i) - xh;
dt = T._(i) - T._(i - 1);
q = _m(Lambda, '*', q) + (1 - Lambda) * _p(xt, 2) / dt;
Res = __[Res, ';', __[T._(i), PWF._(i), xh, q, Lambda]];
Lambda = LambdaL * Lambda + (1 - LambdaL) * LambdaInf;
xh = PWF._(i);
}
Res[_(':'), _(1)] = Res[_(':'), _(1)] * TimeScaling;
// Persistence version;
xhms = Res._(TPS + 1, 3) * ones(NS - TPS, 1);
covxhms = Res._(TPS + 1, 4) * Ts * ((_c(1, NS - TPS)).T);
}
if (IMT != 0)
{
xhms = __[PWF._(TPS), ';', xhms];
covxhms = __[0, ';', covxhms];
}
sigmaxhms = sqrt(covxhms);
ConIntAWFP = _m(xhms, '*', __[1, 1, 1]) + _m(sigmaxhms, '*', __[-2, 0, 2]);
ConIntAWFP = min(max(ConIntAWFP, MinWFP), NomWFP); // Limit confidence limits
// Plot results
// Plot actual as black solid and lower, prediction and upper confidence
// limits as red, green and blue solid for 10 min, dashed for one hour and
// dotted for the rest.
//figure;
NS10Min = min(NS - TPS - 1, round_(600.0 / Ts));
NSOneHour = min(NS - TPS - 1, round_(3600.0 / Ts));
//set(gcf, "defaultaxescolororder", ILMath.eye(3, 3));
outMethod = Method;
outTime = TimePlot;
outX = Res[_(':'), _(2)];
outXhmsAll = ConIntAWFP;
outXhmsAllTimeOffset = (TPS - IMT);
outXhmsLLength = (NS10Min + IMT);
outXhmsUOffset = (NSOneHour + IMT);
//plot(
// TimePlot, Res[ILMath.full, 2 - 1], 'k',///...
// TimePlot[ILMath.r(TPS + 1 - IMT - 1, TPS + NS10Min - 1)], ConIntAWFP[ILMath.r(1 - 1, NS10Min + IMT - 1), ILMath.full],//...
// TimePlot[ILMath.r(TPS + NS10Min + 1 - 1, TPS + NSOneHour - 1)], ConIntAWFP[_a(NS10Min + 1, 1, NSOneHour) + IMT - 1, ILMath.full], "--",//...
// TimePlot[ILMath.r(TPS + NSOneHour + 1 - 1, ILMath.end)], ConIntAWFP[ILMath.r(NSOneHour + 1 + IMT - 1, ILMath.end), ILMath.full], ':');
//title(TitleStr);
//Legend= {'x' 'xhmsL' 'xhms' 'xhmsU'};
//legend(Legend);
//grid('on');
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// DESCRIPTION:
// RLC - 8/9/2014, interpolation, for getting an accurate CP/CT value. The
// reason for a standalone script, is that the built in MATLAB script
// (interp) includes a lot of redundancy checks that are not necessary, and
// not feasible for embedded solutions.
// Based on the NREL5MW Turbine.
// Uses Linear Polynomial Extrapolation, to get a value for CP, given a Beta
// (Revolutional speed) and Lambda (Tip-Speed-Ratio) of the Turbine.
// The interpolation is computed as:
// y = y0 + (y1 - y0)*(x-x0)/(x1-x0)
//
// Beta is the revolutional entry.
// Lambda is the TSR entry ratio.
// turbineTable defines the table to lookup in.
// negYes defines wether the CP value should be allowed to be negative.
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#endregion
internal static void interpTable(out double interpValue, double Beta, double Lambda, ILArray<double> table, ILArray<double> turbineTableBeta, ILArray<double> turbineTableLambda, bool negYes)
{
#region "Used variables declaration"
ILArray<double> turbineTable;
int sizeBt;
int sizeLa;
int Bt0;
int Bt1;
int La0;
int La1;
ILArray<double> tableLookup;
ILArray<double> lambdaIntervals;
double betaIntervals;
#endregion
//% Setup, loads the table, and stores it.
//persistent turbineTable tableLoad
//if isempty(tableLoad)
turbineTable = table.C;
// tableLoad = 1;
//end
size(out sizeBt, out sizeLa, turbineTable);
//% Index Interpolation
// Finds the beta-point of the interpolation.
min(out Bt0, abs(turbineTableBeta - Beta)); // Determines the index of the closest point.
if (Beta > turbineTableBeta._(Bt0))
{
if (Bt0 == sizeBt) //length(turbineTableBeta)
{
Bt1 = Bt0;
Bt0 = Bt0 - 1;
}
else
{
Bt1 = Bt0 + 1;
}
}
else
{
if (Bt0 == 1)
{
Bt1 = Bt0 + 1;
}
else
{
Bt1 = Bt0;
Bt0 = Bt1 - 1;
}
}
// Finds the Lambda-point of the interpolation.
min(out La0, abs(turbineTableLambda - Lambda)); // Determines the index of the closest point.
if (Lambda > turbineTableLambda._(La0))
if (La0 == sizeLa) //length(turbineTableLambda)
{
La1 = La0;
La0 = La1 - 1;
}
else
{
La1 = La0 + 1;
}
else
{
if (La0 == 1)
{
La1 = La0 + 1;
}
else
{
La1 = La0;
La0 = La1 - 1;
}
}
//% Table Interpolation
// Table lookup using indexes obtained previously:
tableLookup = __[ turbineTable._(Bt0, La0), turbineTable._(Bt0, La1), ';',
turbineTable._(Bt1, La0), turbineTable._(Bt1, La1) ];
// Interpolating, using the Lambda values first, then the Betas.
lambdaIntervals = __[ ( (tableLookup._(1, 2) - tableLookup._(1, 1)) / (turbineTableLambda._(La1) - turbineTableLambda._(La0)) ) * (Lambda - turbineTableLambda._(La0)) + tableLookup._(1, 1), ';',
( (tableLookup._(2, 2) - tableLookup._(2, 1)) / (turbineTableLambda._(La1) - turbineTableLambda._(La0)) ) * (Lambda - turbineTableLambda._(La0)) + tableLookup._(1, 2) ];
// Interpolation, using the Beta values (using the intervales computed above).
betaIntervals = ( (lambdaIntervals._(2) - lambdaIntervals._(1)) / (turbineTableBeta._(Bt1) - turbineTableBeta._(Bt0)) ) * (Beta - turbineTableBeta._(Bt0)) + lambdaIntervals._(1);
//% Negativity Handling
// If the negYes value is set as true in the system, the functino will only
// give an absolute value of Ct and Cp - have to be modified so it does this
// correctly.
if (negYes)
{
interpValue = 0;
}
else
{
interpValue = betaIntervals;
}
}
private static void RLSMARX1(out ILArray <double> Theta, out ILArray <double> Sigma, out ILArray <double> Xh, out ILArray <double> Lambda, out ILArray <int> Time, out ILArray <double> AByTime, out ILArray <double> BByTime, out ILArray <double> SigmaByTime, ILArray <double> X, ILArray <double> U = null, ILArray <double> lambda = null, ILArray <double> theta0 = null, ILArray <double> sigma0 = null, int Plot = PlotDef)
{
#region "Original function comments"
//RLS calculates recursive least squares parameter estimates corresponding
// to the output input data X U and the model x(n+1)= A*x(n)+B*u(n)+e(n)
// <=> x(n+1)'= [x(n)' u(n)']*[A'; B']+e(n)
// Duble exponential forgetting can be used where lamba increases
// exponentielly from lamba0 to lambda.
//
// [Theta,Sigma,Xh,Lambda,Time]= RLSMARX1(X,U,lambda,theta0,sigma0)
//
// X: Output Matrix with size number of samples * number of channels
// U: Input Matrix with size number of samples * number of channels
// zero scalar/matrix corresponds to no input, ones corresponds to estimating
// a mean value parameter. (default ones)
// lambda : Forgetting factor (default 0.99) if lambda= [lambda lambda0]
// the forgetting factor increases exponentially from lambda0
// to lambda.
// theta0 : Initial parameter estimate. Notice that theta0= [A0';B0'];
// sigma0 : Initial covvariance estimate.
// If no initial parameters are specifyed a offline/batch
// estimate is used for the start.
// Plot : If 1/true plot informative plots (default: false)
//
// External input:
// Time-stamp: <2014-10-17 11:44:27 tk>
// Version 1: 2014-10-01 app.
// Version 2: 2014-10-02 12:53:07 tk Included LS/offline/batch startup
// Version 3: 2014-10-07 13:57:54 tk Included additional output and
// plotting
// Torben Knudsen
// Aalborg University, Dept. of Electronic Systems, Section of Automation
// and Control
// E-mail: [email protected]
#endregion
#region "Used variables declaration"
double TauLambdaLRel;
double lambdaInf;
double lambda0;
double IniNumSampFrac;
int N;
int n;
int m;
double TauLambdaL;
double lambdaL;
ILArray <double> sigma;
ILArray <double> theta;
int NIni;
ILArray <double> XX;
ILArray <double> YY;
ILArray <double> R;
ILArray <double> Epsilon;
ILArray <double> xh;
ILArray <double> Res;
int i;
ILArray <double> phi;
ILArray <double> epsilon;
#endregion
//% setting up inputs
//UDef = 1;
//lambdaDef = 0.99;
//theta0Def = 0;
//sigma0Def = 0;
//PlotDef = 0;
//if (nargin < 6) { Plot = []; }
if (sigma0 == null)
{
sigma0 = __[' '];
}
if (theta0 == null)
{
theta0 = __[' '];
}
if (lambda == null)
{
lambda = __[' '];
}
if (U == null)
{
U = __[' '];
}
//if (nargin < 1) { error('Error TK: To few input arguments'); }
if (isempty(U))
{
U = UDef;
}
//if (isempty(Plot)) { Plot = PlotDef; }
if (isempty(lambda))
{
lambda = lambdaDef;
}
if (isempty(theta0))
{
theta0 = theta0Def;
}
if (isempty(sigma0))
{
sigma0 = sigma0Def;
}
//% Parameters
TauLambdaLRel = 0.1; // TauLambdaL= 10% of the samples
lambdaInf = lambda._(1);
if (length(lambda) > 1) // Initial lambda;
{
lambda0 = lambda._(2);
}
else
{
lambda0 = lambdaInf;
}
IniNumSampFrac = 0.1; // Number of samples to use for init
//% Definitions etc.
N = size(X, 1);
n = size(X, 2);
m = size(U, 2);
// Calculate Lamba* from TauLambda* assuming Ts= 1
TauLambdaL = TauLambdaLRel * N;
lambdaL = exp(-1.0 / TauLambdaL);
if (all(U == 1))
{
U = ones(N, 1);
}
//% Algorithm
// Initialisation
if (all(all(sigma0 == 0))) // No sigma0 specifyed
{
sigma = 0 * eye(n);
}
else
{
sigma = sigma0;
}
if (all(all(theta0 == 0))) // No sigma0 specifyed
{
theta = zeros(n + m, n);
}
else
{
theta = theta0;
}
// Start with a offline estimate if initial parameters are not specifyed
if (all(__[theta0, ';', sigma0] == 0))
{
NIni = max(n + 1, ceil_(N * IniNumSampFrac));
XX = __[X[_(1, ':', NIni - 1), _(':')], U[_(1, ':', NIni - 1), _(':')]];
YY = X[_(_c_(1, NIni - 1) + 1), _(':')];
R = _m(XX.T, '*', XX);
theta = _m(_m(inv(R), '*', (XX.T)), '*', YY);
Epsilon = YY - _m(XX, '*', theta);
sigma = _m(Epsilon.T, '*', Epsilon) / NIni;
xh = _m(XX[_(end), _(':')], '*', theta);
}
else
{
NIni = 1;
xh = X[_(1), _(':')];
R = 1e6 * eye(n);
}
lambda = lambda0;
Res = __[NIni, (theta[_(':')]).T, (sigma[_(':')]).T, xh, lambda];
AByTime = zeros(n, n, N - NIni + 1);
AByTime[_(':'), _(':'), _(1)] = (theta[_(1, ':', n), _(1, ':', n)]).T;
BByTime = zeros(n, m, N - NIni + 1);
BByTime[_(':'), _(':'), _(1)] = (theta[_(n + 1, ':', n + m), _(1, ':', n)]).T;
SigmaByTime = zeros(n, n, N - NIni + 1);
SigmaByTime[_(':'), _(':'), _(1)] = sigma;
// Recursion
// Model x(n+1)= A*x(n)+B*u(n)+e(n) <=>
// x(n+1)'= [x(n)' u(n)']*[A'; B']+e(n)
for (i = NIni + 1; i <= N; i++)
{
phi = (__[X[_(i - 1), _(':')], U[_(i - 1), _(':')]]).T; // phi(i)
R = (lambda._Scalar()) * R + _m(phi, '*', (phi.T)); // R(i)
xh = _m((phi.T), '*', theta); // xh(i)
epsilon = X[_(i), _(':')] - xh; // epsilon(i)
theta = theta + _m(_s(R, '\\', phi), '*', epsilon); // theta(i)
sigma = _m(lambda, '*', sigma) + _m(_m((1 - lambda), '*', (epsilon.T)), '*', epsilon);
Res = __[Res, ';', __[i, theta[_(':')].T, sigma[_(':')].T, xh, lambda]];
lambda = lambdaL * lambda + (1 - lambdaL) * lambdaInf;
AByTime[_(':'), _(':'), _(i - NIni + 1)] = (theta[_(1, ':', n), _(1, ':', n)]).T; // Index start at 2
BByTime[_(':'), _(':'), _(i - NIni + 1)] = (theta[_(n + 1, ':', n + m), _(1, ':', n)]).T;
SigmaByTime[_(':'), _(':'), _(i - NIni + 1)] = sigma;
}
// Res is organised like
// i theta(:)' sigma(:)' xh lambda with the dimensions
// 1 (n+m)*n n^2 n 1
Time = _int(Res[_(':'), _(1)]);
Theta = Res[_(':'), _(1 + 1, ':', 1 + (n + m) * n)];
Sigma = Res[_(':'), _(1 + 1 + (n + m) * n, ':', 1 + (n + m) * n + _p_(n, 2))];
Xh = Res[_(':'), _(1 + 1 + (n + m) * n + _p_(n, 2), ':', 1 + (n + m) * n + _p_(n, 2) + n)];
Lambda = Res[_(':'), _(1 + 1 + (n + m) * n + _p_(n, 2) + n, ':', end)];
// Do informative plotting and output
//if (Plot)
//{
// figure;
// plot([U X]);
// title('Input and measurements');
// figure;
// plot(Time,reshape(AByTime,n*n,N-NIni+1)');
// title('A parameters');
// figure;
// plot(Time,reshape(BByTime,n*m,N-NIni+1)');
// title('B parameters');
// figure;
// plot(Time,reshape(SigmaByTime,n*n,N-NIni+1)');
// title('Sigma parameters');
// figure;
// plot(Time,Lambda);
// title('Lambda');
// Epsilon= X(Time,:)-Xh;
// Epsilon= Epsilon(100:end,:);
// figure;
// XCorrtkAll(Epsilon);
// figure;
// normplot(Epsilon);
// gridtk('on',figures);
// Lab= ['Final parameters A n*n B n*m Sigma n*n'];
// RowLab= {''};
// ColLab= {''};
// Res= [AByTime(:,:,end) BByTime(:,:,end) SigmaByTime(:,:,end)];
// printmattk(Res,Lab);
//}
}
/// <summary>
/// (i1,i2) =
/// </summary>
/// <param name="equalitySign">'='</param>
public static void _(this ILArray <double> ilArray, int index1, int index2, char equalitySign, double value)
{
ilArray._ <double>(index1, index2, equalitySign, value);
}
// Wake Simulation
// (C) Rasmus Christensen
// Control and Automation, Aalborg University 2014
// Compute the velocity in front of each wind-turbine, with respect to the wind input.
#endregion
internal static void Compute_Vell(out double[,] vel_output, ILArray<double> yTurb, ILArray<int> xTurbC, ILArray<int> yTurbC, ILArray<double> x, ILArray<double> wField, double[] Uhub, double kWake, int iMax, int jMax, int nTurb, double dTurb, ILArray<double> Ct, double dy)
{
#region "Used variables declaration"
double[,] vell_i;
ILArray<double> shadow;
double r0;
double nk;
int k;
int J;
double SS;
double SS0;
int i;
double RR_i;
double Dij;
double Alpha_i;
double Alpha_k;
double Area;
int ii;
double rrt;
double nj;
int jjMin;
int jjMax;
#endregion
//vell_i = wField.C;
vell_i = new double[wField.Size[0], wField.Size[1]];
for (var i0 = 0; i0 < vell_i.GetLength(0); i0++)
{
for (var i1 = 0; i1 < vell_i.GetLength(1); i1++)
{
vell_i[i0, i1] = wField.GetValue(i0, i1);
}
}
shadow = zeros(1, nTurb);
r0 = 0.5 * dTurb;
nk = 2 * ceil(dTurb / dy);
for (k = 1; k <= nTurb; k++)
{
J = 0;
SS = 0;
SS0 = pi * r0 * r0;
for (i = 1; i <= k - 1; i++)
{
RR_i = r0 + kWake * (x._(xTurbC._(k)) - x._(xTurbC._(i)));
Dij = abs(yTurb._(i) - yTurb._(k));
if ((RR_i >= (r0 + Dij)) || (Dij <= dy))
{
SS = SS + ((r0 * r0) / (RR_i * RR_i));
}
else
{
if (RR_i >= (r0 + Dij) && Dij > dy)
{
J = J + 1;
Alpha_i = acos((RR_i * RR_i) + (Dij * Dij) - (r0 * r0) / (2 * RR_i * Dij));
Alpha_k = acos(((r0 * r0) + (Dij * Dij) - (RR_i * RR_i)) / (2 * r0 * Dij));
AArea(out Area, RR_i, r0, Dij);
shadow._(J, '=', (Alpha_i * (RR_i * RR_i) + Alpha_k * (r0 * r0)) - 2 * Area);
SS = SS + ((shadow._(J)) / SS0) * ((r0 * r0) / (RR_i * RR_i));
}
}
}
for (ii = xTurbC._(k); ii <= iMax; ii++)
{
rrt = r0 + kWake * (x._(ii) - x._(xTurbC._(k)));
nj = ceil(rrt / dy);
jjMin = floor_(max(1, yTurbC._(k) - nj));
jjMax = ceil_(min(jMax, yTurbC._(k) + nj));
for (var j = jjMin; j <= jjMax; j++)
{
if (((-vell_i[ii - 1, j - 1] + Uhub[k - 1]) > 0) && (ii > xTurbC._(k) + nk))
{
vell_i[ii - 1, j - 1] = min(vell_i[ii - 2, j - 1], Uhub[k - 1] + Uhub[k - 1] * (sqrt(1 - Ct._(k)) - 1) * ((r0 * r0) / (rrt * rrt)) * (1 - (1 - sqrt(1 - Ct._(k))) * SS));
vell_i[ii - 1, j - 1] = max(0, vell_i[ii - 1, j - 1]);
}
else
{
vell_i[ii - 1, j - 1] = (Uhub[k - 1] + Uhub[k - 1] * (sqrt(1 - Ct._(k)) - 1) * (r0 / rrt) * (r0 / rrt)) * (1 - (1 - sqrt(1 - Ct._(k))) * SS);
vell_i[ii - 1, j - 1] = max(0, vell_i[ii - 1, j - 1]);
}
}
}
}
vel_output = vell_i;
}
/// <summary>
/// = (i)
/// </summary>
public static int _(this ILArray <int> ilArray, int index)
{
return(ilArray._ <int>(index));
}
/// <summary>
/// (i) =
/// </summary>
/// <param name="equalitySign">'='</param>
public static void _(this ILArray <int> ilArray, int index, char equalitySign, int value)
{
ilArray._ <int>(index, equalitySign, value);
}
/// <summary>
/// = (i)
/// </summary>
public static double _(this ILArray <double> ilArray, int index)
{
return(ilArray._ <double>(index));
}
