/** Solves a standard form LP problem in the form of min(c) s.t. A.x = b lb <= x <= ub */
protected int optimizeStandardLP(int nOfSlackVariables) throws Exception {
log.info("optimizeStandardLP");
LPOptimizationRequest lpRequest = getLPOptimizationRequest();
if (log.isDebugEnabled() && lpRequest.isDumpProblem()) {
log.debug("LP problem: " + lpRequest.toString());
}
LPOptimizationResponse lpResponse;
if (lpRequest.isPresolvingDisabled()) {
// optimization
LPPrimalDualMethod opt = new LPPrimalDualMethod(minLBValue, maxUBValue);
opt.setLPOptimizationRequest(lpRequest);
if (opt.optimizePresolvedStandardLP() == OptimizationResponse.FAILED) {
return OptimizationResponse.FAILED;
}
lpResponse = opt.getLPOptimizationResponse();
setLPOptimizationResponse(lpResponse);
} else {
// presolving
LPPresolver lpPresolver = new LPPresolver();
lpPresolver.setAvoidScaling(lpRequest.isRescalingDisabled());
lpPresolver.setAvoidFillIn(lpRequest.isAvoidPresolvingFillIn());
lpPresolver.setAvoidIncreaseSparsity(lpRequest.isAvoidPresolvingIncreaseSparsity());
testPresolver = lpPresolver; // just for testing
lpPresolver.setNOfSlackVariables((short) nOfSlackVariables);
lpPresolver.presolve(getC(), getA(), getB(), getLb(), getUb());
int presolvedDim = lpPresolver.getPresolvedN();
if (presolvedDim == 0) {
// deterministic problem
log.debug("presolvedDim : " + presolvedDim);
log.debug("deterministic LP problem");
lpResponse = new LPOptimizationResponse();
lpResponse.setReturnCode(OptimizationResponse.SUCCESS);
lpResponse.setSolution(new double[] {});
} else {
// solving the presolved problem
DoubleMatrix1D presolvedC = lpPresolver.getPresolvedC();
DoubleMatrix2D presolvedA = lpPresolver.getPresolvedA();
DoubleMatrix1D presolvedB = lpPresolver.getPresolvedB();
if (log.isDebugEnabled()) {
if (lpPresolver.getPresolvedYlb() != null) {
log.debug("Ylb: " + ArrayUtils.toString(lpPresolver.getPresolvedYlb().toArray()));
log.debug("Yub: " + ArrayUtils.toString(lpPresolver.getPresolvedYub().toArray()));
}
if (lpPresolver.getPresolvedZlb() != null) {
log.debug("Zlb: " + ArrayUtils.toString(lpPresolver.getPresolvedZlb().toArray()));
log.debug("Zub: " + ArrayUtils.toString(lpPresolver.getPresolvedZub().toArray()));
}
}
// new LP problem (the presolved problem)
LPOptimizationRequest presolvedLPRequest = lpRequest.cloneMe();
presolvedLPRequest.setC(presolvedC);
presolvedLPRequest.setA(presolvedA);
presolvedLPRequest.setB(presolvedB);
presolvedLPRequest.setLb(lpPresolver.getPresolvedLB());
presolvedLPRequest.setUb(lpPresolver.getPresolvedUB());
presolvedLPRequest.setYlb(lpPresolver.getPresolvedYlb());
presolvedLPRequest.setYub(lpPresolver.getPresolvedYub());
presolvedLPRequest.setZlb(lpPresolver.getPresolvedZlb());
presolvedLPRequest.setZub(lpPresolver.getPresolvedZub());
if (getInitialPoint() != null) {
presolvedLPRequest.setInitialPoint(lpPresolver.presolve(getInitialPoint().toArray()));
}
if (getNotFeasibleInitialPoint() != null) {
presolvedLPRequest.setNotFeasibleInitialPoint(
lpPresolver.presolve(getNotFeasibleInitialPoint().toArray()));
}
// optimization
// NB: because of rescaling during the presolving phase, minLB and maxUB could have been
// rescaled
double rescaledMinLBValue =
(Double.isNaN(lpPresolver.getMinRescaledLB()))
? this.minLBValue
: lpPresolver.getMinRescaledLB();
double rescaledMaxUBValue =
(Double.isNaN(lpPresolver.getMaxRescaledUB()))
? this.maxUBValue
: lpPresolver.getMaxRescaledUB();
LPPrimalDualMethod opt = new LPPrimalDualMethod(rescaledMinLBValue, rescaledMaxUBValue);
opt.setLPOptimizationRequest(presolvedLPRequest);
if (opt.optimizePresolvedStandardLP() == OptimizationResponse.FAILED) {
return OptimizationResponse.FAILED;
}
lpResponse = opt.getLPOptimizationResponse();
}
// postsolving
double[] postsolvedSolution = lpPresolver.postsolve(lpResponse.getSolution());
lpResponse.setSolution(postsolvedSolution);
setLPOptimizationResponse(lpResponse);
}
return lpResponse.getReturnCode();
}
// formulate and generate the solution, if necessary
private void solve() {
if (solved) return;
// min obj.x s.t. a.x=b, lb <= x <= ub
// x is energy use per block for size hours, b is slack var per block.
// Block is a shift, or portion of shift with constant price.
// For multi-hour blocks, energy use is evenly distributed across hours
// after solution.
Date start = new Date();
int shifts = needs.length;
// Create blocks that break on both shift boundaries and tariff price
// boundaries.
ShiftBlock[] blocks = makeBlocks(shifts);
int columns = blocks.length;
int blockIndex = -1;
double[] obj = new double[columns + shifts];
double[][] a = new double[shifts][columns + shifts];
double[] b = new double[shifts];
double[] lb = new double[columns + shifts];
double[] ub = new double[columns + shifts];
int column = 0;
double cumulativeMin = 0.0; // this is the primary constraint
// construct the problem
for (int i = 0; i < shifts; i++) {
// one iteration per shift
// double kwh =
// needs[i].getEnergyNeeded() + needs[i].getMaxSurplus();
// for (int j = 0; j < needs[i].getDuration(); j++) {
while ((blockIndex < blocks.length - 1)
&& (blocks[blockIndex + 1].getShiftEnergy() == needs[i])) {
blockIndex += 1;
// one iteration per block within a shift
// fill in objective function
obj[column] = blocks[blockIndex].getCost();
lb[column] = 0.0;
ub[column] =
(needs[i].getEnergyNeeded() + needs[i].getMaxSurplus())
* (double) blocks[blockIndex].getDuration()
/ needs[i].getDuration();
column += 1;
// construct cumulative usage constraints
// a[i][column] = -1.0;
// time = time.plus(TimeService.HOUR);
}
// fill a row up to column
for (int j = 0; j < column; j++) {
a[i][j] = -1.0;
}
// b vector - one entry per constraint
double need = needs[i].getEnergyNeeded();
if (needs[i].getMaxSurplus() < 0.0) need += needs[i].getMaxSurplus();
cumulativeMin += need;
b[i] = -cumulativeMin;
// fill in slack values, one per constraint
obj[columns + i] = 0.0;
a[i][columns + i] = 1.0;
lb[columns + i] = 0.0;
// upper bound is max possible energy for shift
ub[columns + i] = (needs[i].getEnergyNeeded() + needs[i].getMaxSurplus());
}
// run the optimization
LPOptimizationRequest or = new LPOptimizationRequest();
log.debug("Obj: " + Arrays.toString(obj));
or.setC(obj);
log.debug("a:");
for (int i = 0; i < a.length; i++) log.debug(Arrays.toString(a[i]));
or.setA(a);
log.debug("b: " + Arrays.toString(b));
or.setB(b);
or.setLb(lb);
log.debug("ub: " + Arrays.toString(ub));
or.setUb(ub);
or.setTolerance(1.0e-2);
LPPrimalDualMethod opt = new LPPrimalDualMethod();
opt.setLPOptimizationRequest(or);
try {
int returnCode = opt.optimize();
if (returnCode != OptimizationResponse.SUCCESS) {
log.error(getName() + "bad optimization return code " + returnCode);
}
double[] sol = opt.getOptimizationResponse().getSolution();
Date end = new Date();
log.info("Solution time: " + (end.getTime() - start.getTime()));
log.debug("Solution = " + Arrays.toString(sol));
recordSolution(sol, blocks);
} catch (Exception e) {
log.error(e.toString());
}
// we call it solved whether or not the solution was successful
solved = true;
}
/** Solves an LP in the form of: min(c) s.t. A.x = b G.x < h lb <= x <= ub */
@Override
public int optimize() throws Exception {
log.info("optimize");
LPOptimizationRequest lpRequest = getLPOptimizationRequest();
if (log.isDebugEnabled() && lpRequest.isDumpProblem()) {
log.debug("LP problem: " + lpRequest.toString());
}
// standard form conversion
LPStandardConverter lpConverter =
new LPStandardConverter(); // the slack variables will have default unboundedUBValue
lpConverter.toStandardForm(getC(), getG(), getH(), getA(), getB(), getLb(), getUb());
int nOfSlackVariables = lpConverter.getStandardS();
log.debug("nOfSlackVariables: " + nOfSlackVariables);
DoubleMatrix1D standardC = lpConverter.getStandardC();
DoubleMatrix2D standardA = lpConverter.getStandardA();
DoubleMatrix1D standardB = lpConverter.getStandardB();
DoubleMatrix1D standardLb = lpConverter.getStandardLB();
DoubleMatrix1D standardUb = lpConverter.getStandardUB();
// solve the standard form problem
LPOptimizationRequest standardLPRequest = lpRequest.cloneMe();
standardLPRequest.setC(standardC);
standardLPRequest.setA(standardA);
standardLPRequest.setB(standardB);
standardLPRequest.setLb(
ColtUtils.replaceValues(
standardLb,
lpConverter.getUnboundedLBValue(),
minLBValue)); // substitute not-double numbers
standardLPRequest.setUb(
ColtUtils.replaceValues(
standardUb,
lpConverter.getUnboundedUBValue(),
maxUBValue)); // substitute not-double numbers
if (getInitialPoint() != null) {
standardLPRequest.setInitialPoint(
lpConverter.getStandardComponents(getInitialPoint().toArray()));
}
if (getNotFeasibleInitialPoint() != null) {
standardLPRequest.setNotFeasibleInitialPoint(
lpConverter.getStandardComponents(getNotFeasibleInitialPoint().toArray()));
}
// optimization
LPPrimalDualMethod opt = new LPPrimalDualMethod(minLBValue, maxUBValue);
opt.setLPOptimizationRequest(standardLPRequest);
if (opt.optimizeStandardLP(nOfSlackVariables) == OptimizationResponse.FAILED) {
return OptimizationResponse.FAILED;
}
// back to original form
LPOptimizationResponse lpResponse = opt.getLPOptimizationResponse();
double[] standardSolution = lpResponse.getSolution();
double[] originalSol = lpConverter.postConvert(standardSolution);
lpResponse.setSolution(originalSol);
setLPOptimizationResponse(lpResponse);
return lpResponse.getReturnCode();
}