Examples with LevenbergMarquardtOptimizer org.hipparchus.optim.nonlinear.vector.leastsquares.LevenbergMarquardtOptimizer used on opensource projects

Example 1 with LevenbergMarquardtOptimizer

use of org.hipparchus.optim.nonlinear.vector.leastsquares.LevenbergMarquardtOptimizer in project Orekit by CS-SI.

the class HarmonicParametricAccelerationTest method testCoefficientsDetermination.

@Test
public void testCoefficientsDetermination() throws OrekitException {
    final double mass = 2500;
    final Orbit orbit = new CircularOrbit(7500000.0, 1.0e-4, 1.0e-3, 1.7, 0.3, 0.5, PositionAngle.TRUE, FramesFactory.getEME2000(), new AbsoluteDate(new DateComponents(2004, 2, 3), TimeComponents.H00, TimeScalesFactory.getUTC()), Constants.EIGEN5C_EARTH_MU);
    final double period = orbit.getKeplerianPeriod();
    AttitudeProvider maneuverLaw = new LofOffset(orbit.getFrame(), LOFType.VNC);
    SpacecraftState initialState = new SpacecraftState(orbit, maneuverLaw.getAttitude(orbit, orbit.getDate(), orbit.getFrame()), mass);
    double dP = 10.0;
    double minStep = 0.001;
    double maxStep = 100;
    double[][] tolerance = NumericalPropagator.tolerances(dP, orbit, orbit.getType());
    // generate PV measurements corresponding to a tangential maneuver
    AdaptiveStepsizeIntegrator integrator0 = new DormandPrince853Integrator(minStep, maxStep, tolerance[0], tolerance[1]);
    integrator0.setInitialStepSize(60);
    final NumericalPropagator propagator0 = new NumericalPropagator(integrator0);
    propagator0.setInitialState(initialState);
    propagator0.setAttitudeProvider(maneuverLaw);
    ForceModel hpaRefX1 = new HarmonicParametricAcceleration(Vector3D.PLUS_I, true, "refX1", null, period, 1);
    ForceModel hpaRefY1 = new HarmonicParametricAcceleration(Vector3D.PLUS_J, true, "refY1", null, period, 1);
    ForceModel hpaRefZ2 = new HarmonicParametricAcceleration(Vector3D.PLUS_K, true, "refZ2", null, period, 2);
    hpaRefX1.getParametersDrivers()[0].setValue(2.4e-2);
    hpaRefX1.getParametersDrivers()[1].setValue(3.1);
    hpaRefY1.getParametersDrivers()[0].setValue(4.0e-2);
    hpaRefY1.getParametersDrivers()[1].setValue(0.3);
    hpaRefZ2.getParametersDrivers()[0].setValue(1.0e-2);
    hpaRefZ2.getParametersDrivers()[1].setValue(1.8);
    propagator0.addForceModel(hpaRefX1);
    propagator0.addForceModel(hpaRefY1);
    propagator0.addForceModel(hpaRefZ2);
    final List<ObservedMeasurement<?>> measurements = new ArrayList<>();
    propagator0.setMasterMode(10.0, (state, isLast) -> measurements.add(new PV(state.getDate(), state.getPVCoordinates().getPosition(), state.getPVCoordinates().getVelocity(), 1.0e-3, 1.0e-6, 1.0)));
    propagator0.propagate(orbit.getDate().shiftedBy(900));
    // set up an estimator to retrieve the maneuver as several harmonic accelerations in inertial frame
    final NumericalPropagatorBuilder propagatorBuilder = new NumericalPropagatorBuilder(orbit, new DormandPrince853IntegratorBuilder(minStep, maxStep, dP), PositionAngle.TRUE, dP);
    propagatorBuilder.addForceModel(new HarmonicParametricAcceleration(Vector3D.PLUS_I, true, "X1", null, period, 1));
    propagatorBuilder.addForceModel(new HarmonicParametricAcceleration(Vector3D.PLUS_J, true, "Y1", null, period, 1));
    propagatorBuilder.addForceModel(new HarmonicParametricAcceleration(Vector3D.PLUS_K, true, "Z2", null, period, 2));
    final BatchLSEstimator estimator = new BatchLSEstimator(new LevenbergMarquardtOptimizer(), propagatorBuilder);
    estimator.setParametersConvergenceThreshold(1.0e-2);
    estimator.setMaxIterations(20);
    estimator.setMaxEvaluations(100);
    for (final ObservedMeasurement<?> measurement : measurements) {
        estimator.addMeasurement(measurement);
    }
    // we will estimate only the force model parameters, not the orbit
    for (final ParameterDriver d : estimator.getOrbitalParametersDrivers(false).getDrivers()) {
        d.setSelected(false);
    }
    setParameter(estimator, "X1 γ", 1.0e-2);
    setParameter(estimator, "X1 φ", 4.0);
    setParameter(estimator, "Y1 γ", 1.0e-2);
    setParameter(estimator, "Y1 φ", 0.0);
    setParameter(estimator, "Z2 γ", 1.0e-2);
    setParameter(estimator, "Z2 φ", 1.0);
    estimator.estimate();
    Assert.assertTrue(estimator.getIterationsCount() < 15);
    Assert.assertTrue(estimator.getEvaluationsCount() < 15);
    Assert.assertEquals(0.0, estimator.getOptimum().getRMS(), 1.0e-5);
    Assert.assertEquals(hpaRefX1.getParametersDrivers()[0].getValue(), getParameter(estimator, "X1 γ"), 1.e-12);
    Assert.assertEquals(hpaRefX1.getParametersDrivers()[1].getValue(), getParameter(estimator, "X1 φ"), 1.e-12);
    Assert.assertEquals(hpaRefY1.getParametersDrivers()[0].getValue(), getParameter(estimator, "Y1 γ"), 1.e-12);
    Assert.assertEquals(hpaRefY1.getParametersDrivers()[1].getValue(), getParameter(estimator, "Y1 φ"), 1.e-12);
    Assert.assertEquals(hpaRefZ2.getParametersDrivers()[0].getValue(), getParameter(estimator, "Z2 γ"), 1.e-12);
    Assert.assertEquals(hpaRefZ2.getParametersDrivers()[1].getValue(), getParameter(estimator, "Z2 φ"), 1.e-12);
}

Example 2 with LevenbergMarquardtOptimizer

use of org.hipparchus.optim.nonlinear.vector.leastsquares.LevenbergMarquardtOptimizer in project Orekit by CS-SI.

the class BatchLSEstimatorTest method testKeplerRangeRate.

/**
 * Perfect range rate measurements with a perfect start
 * @throws OrekitException
 */
@Test
public void testKeplerRangeRate() throws OrekitException {
    Context context = EstimationTestUtils.eccentricContext("regular-data:potential:tides");
    final NumericalPropagatorBuilder propagatorBuilder = context.createBuilder(OrbitType.KEPLERIAN, PositionAngle.TRUE, true, 1.0e-6, 60.0, 1.0);
    // create perfect range rate measurements
    final Propagator propagator = EstimationTestUtils.createPropagator(context.initialOrbit, propagatorBuilder);
    final List<ObservedMeasurement<?>> measurements1 = EstimationTestUtils.createMeasurements(propagator, new RangeRateMeasurementCreator(context, false), 1.0, 3.0, 300.0);
    final List<ObservedMeasurement<?>> measurements = new ArrayList<ObservedMeasurement<?>>();
    measurements.addAll(measurements1);
    // create orbit estimator
    final BatchLSEstimator estimator = new BatchLSEstimator(new LevenbergMarquardtOptimizer(), propagatorBuilder);
    for (final ObservedMeasurement<?> rangerate : measurements) {
        estimator.addMeasurement(rangerate);
    }
    estimator.setParametersConvergenceThreshold(1.0e-3);
    estimator.setMaxIterations(10);
    estimator.setMaxEvaluations(20);
    EstimationTestUtils.checkFit(context, estimator, 2, 3, 0.0, 1.6e-2, 0.0, 3.4e-2, // we only have range rate...
    0.0, // we only have range rate...
    170.0, 0.0, 6.5e-2);
}

Example 3 with LevenbergMarquardtOptimizer

use of org.hipparchus.optim.nonlinear.vector.leastsquares.LevenbergMarquardtOptimizer in project Orekit by CS-SI.

the class BatchLSEstimatorTest method testKeplerPV.

/**
 * Perfect PV measurements with a perfect start
 * @throws OrekitException
 */
@Test
public void testKeplerPV() throws OrekitException {
    Context context = EstimationTestUtils.eccentricContext("regular-data:potential:tides");
    final NumericalPropagatorBuilder propagatorBuilder = context.createBuilder(OrbitType.KEPLERIAN, PositionAngle.TRUE, true, 1.0e-6, 60.0, 1.0);
    // create perfect PV measurements
    final Propagator propagator = EstimationTestUtils.createPropagator(context.initialOrbit, propagatorBuilder);
    final List<ObservedMeasurement<?>> measurements = EstimationTestUtils.createMeasurements(propagator, new PVMeasurementCreator(), 0.0, 1.0, 300.0);
    // create orbit estimator
    final BatchLSEstimator estimator = new BatchLSEstimator(new LevenbergMarquardtOptimizer(), propagatorBuilder);
    for (final ObservedMeasurement<?> measurement : measurements) {
        estimator.addMeasurement(measurement);
    }
    estimator.setParametersConvergenceThreshold(1.0e-2);
    estimator.setMaxIterations(10);
    estimator.setMaxEvaluations(20);
    EstimationTestUtils.checkFit(context, estimator, 1, 4, 0.0, 2.2e-8, 0.0, 1.1e-7, 0.0, 1.4e-8, 0.0, 6.3e-12);
    RealMatrix normalizedCovariances = estimator.getOptimum().getCovariances(1.0e-10);
    RealMatrix physicalCovariances = estimator.getPhysicalCovariances(1.0e-10);
    Assert.assertEquals(6, normalizedCovariances.getRowDimension());
    Assert.assertEquals(6, normalizedCovariances.getColumnDimension());
    Assert.assertEquals(6, physicalCovariances.getRowDimension());
    Assert.assertEquals(6, physicalCovariances.getColumnDimension());
    Assert.assertEquals(0.00258, physicalCovariances.getEntry(0, 0), 1.0e-5);
}

Example 4 with LevenbergMarquardtOptimizer

use of org.hipparchus.optim.nonlinear.vector.leastsquares.LevenbergMarquardtOptimizer in project Orekit by CS-SI.

the class BatchLSEstimatorTest method testKeplerRangeAndRangeRate.

/**
 * Perfect range and range rate measurements with a perfect start
 * @throws OrekitException
 */
@Test
public void testKeplerRangeAndRangeRate() throws OrekitException {
    Context context = EstimationTestUtils.eccentricContext("regular-data:potential:tides");
    final NumericalPropagatorBuilder propagatorBuilder = context.createBuilder(OrbitType.KEPLERIAN, PositionAngle.TRUE, true, 1.0e-6, 60.0, 1.0);
    // create perfect range measurements
    final Propagator propagator = EstimationTestUtils.createPropagator(context.initialOrbit, propagatorBuilder);
    final List<ObservedMeasurement<?>> measurementsRange = EstimationTestUtils.createMeasurements(propagator, new RangeMeasurementCreator(context), 1.0, 3.0, 300.0);
    final List<ObservedMeasurement<?>> measurementsRangeRate = EstimationTestUtils.createMeasurements(propagator, new RangeRateMeasurementCreator(context, false), 1.0, 3.0, 300.0);
    // concat measurements
    final List<ObservedMeasurement<?>> measurements = new ArrayList<ObservedMeasurement<?>>();
    measurements.addAll(measurementsRange);
    measurements.addAll(measurementsRangeRate);
    // create orbit estimator
    final BatchLSEstimator estimator = new BatchLSEstimator(new LevenbergMarquardtOptimizer(), propagatorBuilder);
    for (final ObservedMeasurement<?> meas : measurements) {
        estimator.addMeasurement(meas);
    }
    estimator.setParametersConvergenceThreshold(1.0e-3);
    estimator.setMaxIterations(10);
    estimator.setMaxEvaluations(20);
    // we have low correlation between the two types of measurement. We can expect a good estimate.
    EstimationTestUtils.checkFit(context, estimator, 1, 2, 0.0, 0.16, 0.0, 0.40, 0.0, 2.1e-3, 0.0, 8.1e-7);
}

Example 5 with LevenbergMarquardtOptimizer

use of org.hipparchus.optim.nonlinear.vector.leastsquares.LevenbergMarquardtOptimizer in project Orekit by CS-SI.

the class GroundStationTest method testEstimateStationPosition.

@Test
public void testEstimateStationPosition() throws OrekitException, IOException, ClassNotFoundException {
    Context context = EstimationTestUtils.eccentricContext("regular-data:potential:tides");
    final NumericalPropagatorBuilder propagatorBuilder = context.createBuilder(OrbitType.KEPLERIAN, PositionAngle.TRUE, true, 1.0e-6, 60.0, 0.001);
    // create perfect range measurements
    final Propagator propagator = EstimationTestUtils.createPropagator(context.initialOrbit, propagatorBuilder);
    final List<ObservedMeasurement<?>> measurements = EstimationTestUtils.createMeasurements(propagator, new RangeMeasurementCreator(context), 1.0, 3.0, 300.0);
    // move one station
    final RandomGenerator random = new Well19937a(0x4adbecfc743bda60l);
    final TopocentricFrame base = context.stations.get(0).getBaseFrame();
    final BodyShape parent = base.getParentShape();
    final Vector3D baseOrigin = parent.transform(base.getPoint());
    final Vector3D deltaTopo = new Vector3D(2 * random.nextDouble() - 1, 2 * random.nextDouble() - 1, 2 * random.nextDouble() - 1);
    final Transform topoToParent = base.getTransformTo(parent.getBodyFrame(), (AbsoluteDate) null);
    final Vector3D deltaParent = topoToParent.transformVector(deltaTopo);
    final String movedSuffix = "-moved";
    final GroundStation moved = new GroundStation(new TopocentricFrame(parent, parent.transform(baseOrigin.subtract(deltaParent), parent.getBodyFrame(), null), base.getName() + movedSuffix), context.ut1.getEOPHistory(), context.stations.get(0).getDisplacements());
    // create orbit estimator
    final BatchLSEstimator estimator = new BatchLSEstimator(new LevenbergMarquardtOptimizer(), propagatorBuilder);
    for (final ObservedMeasurement<?> measurement : measurements) {
        final Range range = (Range) measurement;
        final String name = range.getStation().getBaseFrame().getName() + movedSuffix;
        if (moved.getBaseFrame().getName().equals(name)) {
            estimator.addMeasurement(new Range(moved, range.getDate(), range.getObservedValue()[0], range.getTheoreticalStandardDeviation()[0], range.getBaseWeight()[0]));
        } else {
            estimator.addMeasurement(range);
        }
    }
    estimator.setParametersConvergenceThreshold(1.0e-3);
    estimator.setMaxIterations(100);
    estimator.setMaxEvaluations(200);
    // we want to estimate station offsets
    moved.getEastOffsetDriver().setSelected(true);
    moved.getNorthOffsetDriver().setSelected(true);
    moved.getZenithOffsetDriver().setSelected(true);
    EstimationTestUtils.checkFit(context, estimator, 2, 3, 0.0, 5.6e-7, 0.0, 1.4e-6, 0.0, 4.8e-7, 0.0, 2.6e-10);
    Assert.assertEquals(deltaTopo.getX(), moved.getEastOffsetDriver().getValue(), 4.5e-7);
    Assert.assertEquals(deltaTopo.getY(), moved.getNorthOffsetDriver().getValue(), 6.2e-7);
    Assert.assertEquals(deltaTopo.getZ(), moved.getZenithOffsetDriver().getValue(), 2.6e-7);
    GeodeticPoint result = moved.getOffsetGeodeticPoint(null);
    GeodeticPoint reference = context.stations.get(0).getBaseFrame().getPoint();
    Assert.assertEquals(reference.getLatitude(), result.getLatitude(), 1.4e-14);
    Assert.assertEquals(reference.getLongitude(), result.getLongitude(), 2.9e-14);
    Assert.assertEquals(reference.getAltitude(), result.getAltitude(), 2.6e-7);
    RealMatrix normalizedCovariances = estimator.getOptimum().getCovariances(1.0e-10);
    RealMatrix physicalCovariances = estimator.getPhysicalCovariances(1.0e-10);
    Assert.assertEquals(9, normalizedCovariances.getRowDimension());
    Assert.assertEquals(9, normalizedCovariances.getColumnDimension());
    Assert.assertEquals(9, physicalCovariances.getRowDimension());
    Assert.assertEquals(9, physicalCovariances.getColumnDimension());
    Assert.assertEquals(0.55431, physicalCovariances.getEntry(6, 6), 1.0e-5);
    Assert.assertEquals(0.22694, physicalCovariances.getEntry(7, 7), 1.0e-5);
    Assert.assertEquals(0.13106, physicalCovariances.getEntry(8, 8), 1.0e-5);
    ByteArrayOutputStream bos = new ByteArrayOutputStream();
    ObjectOutputStream oos = new ObjectOutputStream(bos);
    oos.writeObject(moved.getEstimatedEarthFrame().getTransformProvider());
    Assert.assertTrue(bos.size() > 155000);
    Assert.assertTrue(bos.size() < 160000);
    ByteArrayInputStream bis = new ByteArrayInputStream(bos.toByteArray());
    ObjectInputStream ois = new ObjectInputStream(bis);
    EstimatedEarthFrameProvider deserialized = (EstimatedEarthFrameProvider) ois.readObject();
    Assert.assertEquals(moved.getPrimeMeridianOffsetDriver().getValue(), deserialized.getPrimeMeridianOffsetDriver().getValue(), 1.0e-15);
    Assert.assertEquals(moved.getPrimeMeridianDriftDriver().getValue(), deserialized.getPrimeMeridianDriftDriver().getValue(), 1.0e-15);
    Assert.assertEquals(moved.getPolarOffsetXDriver().getValue(), deserialized.getPolarOffsetXDriver().getValue(), 1.0e-15);
    Assert.assertEquals(moved.getPolarDriftXDriver().getValue(), deserialized.getPolarDriftXDriver().getValue(), 1.0e-15);
    Assert.assertEquals(moved.getPolarOffsetYDriver().getValue(), deserialized.getPolarOffsetYDriver().getValue(), 1.0e-15);
    Assert.assertEquals(moved.getPolarDriftYDriver().getValue(), deserialized.getPolarDriftYDriver().getValue(), 1.0e-15);
}