WoodburyEngine.java

/**
 * Copyright (c) 2020, RTE (http://www.rte-france.com)
 * This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/.
 * SPDX-License-Identifier: MPL-2.0
 */
package com.powsybl.openloadflow.dc.fastdc;

import com.powsybl.commons.PowsyblException;
import com.powsybl.commons.report.ReportNode;
import com.powsybl.math.matrix.DenseMatrix;
import com.powsybl.math.matrix.LUDecomposition;
import com.powsybl.openloadflow.dc.DcLoadFlowContext;
import com.powsybl.openloadflow.dc.DcLoadFlowParameters;
import com.powsybl.openloadflow.dc.equations.AbstractClosedBranchDcFlowEquationTerm;
import com.powsybl.openloadflow.dc.equations.ClosedBranchSide1DcFlowEquationTerm;
import com.powsybl.openloadflow.dc.equations.DcEquationSystemCreationParameters;
import com.powsybl.openloadflow.dc.equations.DcEquationType;
import com.powsybl.openloadflow.equations.Equation;
import com.powsybl.openloadflow.network.*;
import com.powsybl.openloadflow.network.action.AbstractLfBranchAction;
import com.powsybl.openloadflow.network.action.AbstractLfTapChangerAction;
import com.powsybl.openloadflow.network.action.LfAction;

import java.util.*;

import static com.powsybl.openloadflow.dc.DcLoadFlowEngine.distributeSlack;
import static com.powsybl.openloadflow.dc.DcLoadFlowEngine.solve;

/**
 * @author Geoffroy Jamgotchian {@literal <geoffroy.jamgotchian at rte-france.com>}
 * @author Ga��l Macherel {@literal <gael.macherel@artelys.com>}
 * @author Pierre Arvy {@literal <pierre.arvy@artelys.com}
 */
public class WoodburyEngine {

    private final DcEquationSystemCreationParameters creationParameters;

    private final List<ComputedContingencyElement> contingencyElements;

    private final DenseMatrix contingenciesStates;

    private final List<AbstractComputedElement> actionElements;

    private final DenseMatrix actionsStates;

    public WoodburyEngine(DcEquationSystemCreationParameters creationParameters, List<ComputedContingencyElement> contingencyElements,
                          DenseMatrix contingenciesStates) {
        this.creationParameters = Objects.requireNonNull(creationParameters);
        this.contingencyElements = Objects.requireNonNull(contingencyElements);
        this.contingenciesStates = Objects.requireNonNull(contingenciesStates);
        this.actionElements = Collections.emptyList();
        this.actionsStates = DenseMatrix.EMPTY;
    }

    public WoodburyEngine(DcEquationSystemCreationParameters creationParameters, List<ComputedContingencyElement> contingencyElements,
                          DenseMatrix contingenciesStates, List<AbstractComputedElement> actionElements, DenseMatrix actionsStates) {
        this.creationParameters = Objects.requireNonNull(creationParameters);
        this.contingencyElements = Objects.requireNonNull(contingencyElements);
        this.contingenciesStates = Objects.requireNonNull(contingenciesStates);
        this.actionElements = Objects.requireNonNull(actionElements);
        this.actionsStates = Objects.requireNonNull(actionsStates);
    }

    /**
     * A simplified version of DcLoadFlowEngine that supports on the fly bus and branch disabling.
     * Note that it does not update the state vector and the network at the end (because we don't need it to just evaluate a few equations).
     */
    public static double[] runDcLoadFlowWithModifiedTargetVector(DcLoadFlowContext loadFlowContext, DisabledNetwork disabledNetwork, ReportNode reportNode) {
        return runDcLoadFlowWithModifiedTargetVector(loadFlowContext, disabledNetwork, reportNode, Collections.emptyList());
    }

    /**
     * A simplified version of DcLoadFlowEngine that supports on the fly bus and branch disabling, and pst actions.
     * Note that it does not update the state vector and the network at the end (because we don't need it to just evaluate a few equations).
     */
    public static double[] runDcLoadFlowWithModifiedTargetVector(DcLoadFlowContext loadFlowContext, DisabledNetwork disabledNetwork, ReportNode reportNode, List<LfAction> lfActions) {
        Collection<LfBus> remainingBuses;
        if (disabledNetwork.getBuses().isEmpty()) {
            remainingBuses = loadFlowContext.getNetwork().getBuses();
        } else {
            remainingBuses = new LinkedHashSet<>(loadFlowContext.getNetwork().getBuses());
            remainingBuses.removeAll(disabledNetwork.getBuses());
        }

        DcLoadFlowParameters parameters = loadFlowContext.getParameters();
        if (parameters.isDistributedSlack()) {
            distributeSlack(loadFlowContext.getNetwork(), remainingBuses, parameters.getBalanceType(), parameters.getNetworkParameters().isUseActiveLimits());
        }

        // we need to copy the target array because:
        //  - in case of disabled buses or branches some elements could be overwritten to zero
        //  - JacobianMatrix.solveTransposed take as an input the second member and reuse the array
        //    to fill with the solution
        // so we need to copy to later the target as it is and reusable for next run
        var targetVectorArray = loadFlowContext.getTargetVector().getArray().clone();

        if (!disabledNetwork.getBuses().isEmpty()) {
            // set buses injections and transformers to 0
            disabledNetwork.getBuses().stream()
                    .flatMap(lfBus -> loadFlowContext.getEquationSystem().getEquation(lfBus.getNum(), DcEquationType.BUS_TARGET_P).stream())
                    .map(Equation::getColumn)
                    .forEach(column -> targetVectorArray[column] = 0);
        }

        if (!disabledNetwork.getBranches().isEmpty()) {
            // set transformer phase shift to 0
            disabledNetwork.getBranches().stream()
                    .flatMap(lfBranch -> loadFlowContext.getEquationSystem().getEquation(lfBranch.getNum(), DcEquationType.BRANCH_TARGET_ALPHA1).stream())
                    .map(Equation::getColumn)
                    .forEach(column -> targetVectorArray[column] = 0);
        }

        if (!lfActions.isEmpty()) {
            // set transformer phase shift to new shifting value
            lfActions.stream()
                    .filter(AbstractLfTapChangerAction.class::isInstance)
                    .map(action -> ((AbstractLfTapChangerAction<?>) action).getChange())
                    .filter(Objects::nonNull)
                    .forEach(tapPositionChange -> {
                        LfBranch lfBranch = tapPositionChange.getBranch();
                        loadFlowContext.getEquationSystem().getEquation(lfBranch.getNum(), DcEquationType.BRANCH_TARGET_ALPHA1).ifPresent(
                                dcVariableTypeDcEquationTypeEquation -> {
                                    int column = dcVariableTypeDcEquationTypeEquation.getColumn();
                                    targetVectorArray[column] = tapPositionChange.getNewPiModel().getA1();
                                }
                        );
                    });

            // set transformer phase shift to 0 for disabled phase tap changers by actions
            lfActions.stream()
                    .filter(AbstractLfBranchAction.class::isInstance)
                    .map(lfAction -> ((AbstractLfBranchAction<?>) lfAction).getDisabledBranch())
                    .filter(Objects::nonNull)
                    .flatMap(lfBranch -> loadFlowContext.getEquationSystem().getEquation(lfBranch.getNum(), DcEquationType.BRANCH_TARGET_ALPHA1).stream())
                    .map(Equation::getColumn)
                    .forEach(column -> targetVectorArray[column] = 0);
        }

        boolean succeeded = solve(targetVectorArray, loadFlowContext.getJacobianMatrix(), reportNode);
        if (!succeeded) {
            throw new PowsyblException("DC solver failed");
        }

        return targetVectorArray; // now contains dx
    }

    private double calculatePower(LfBranch lfBranch) {
        return calculatePower(lfBranch.getPiModel());
    }

    private double calculatePower(PiModel piModel) {
        return AbstractClosedBranchDcFlowEquationTerm.computePower(creationParameters.isUseTransformerRatio(), creationParameters.getDcApproximationType(), piModel);
    }

    /**
     * Returns the value of the right-hand side member, associated with the linear system to be solved in order to compute
     * the flow transfer factors.
     */
    private double getAlphaRhsValue(DenseMatrix states, ClosedBranchSide1DcFlowEquationTerm p1, int columnState, AbstractComputedElement computedElement) {
        double newAlpha = 0;
        // enabling of transformers is not yet supported
        if (computedElement instanceof ComputedTapPositionChangeElement computedTapPositionChangeElement) {
            TapPositionChange tapPositionChange = computedTapPositionChangeElement.getTapPositionChange();
            PiModel newPiModel = tapPositionChange.getNewPiModel();
            newAlpha = newPiModel.getA1();
        }
        return states.get(p1.getPh1Var().getRow(), columnState) - states.get(p1.getPh2Var().getRow(), columnState) + newAlpha;
    }

    /**
     * Returns a value on the diagonal of the alpha matrix linked to the switch in impedance caused by an action
     */
    private double computeDeltaXForDiagonalValues(LfBranch lfBranch, AbstractComputedElement element) {
        double oldPower = 0;
        double newPower = 0;
        // if tap position change, the power transiting on the branch might have changed
        if (element instanceof ComputedTapPositionChangeElement tapChangeElement) {
            TapPositionChange tapPositionChange = tapChangeElement.getTapPositionChange();
            newPower = calculatePower(tapPositionChange.getNewPiModel());
            oldPower = calculatePower(lfBranch);
        } else if (element instanceof ComputedSwitchBranchElement switchElement) {
            // if enabled, power is now transiting on the branch
            if (switchElement.isEnabled()) {
                newPower = calculatePower(lfBranch);
                // if disabled, no power is transiting on the branch anymore
            } else {
                oldPower = calculatePower(lfBranch);
            }
        } else {
            throw new IllegalStateException("Unexpected computed element type");
        }
        return 1d / (oldPower - newPower);
    }

    /**
     * Returns the value of the matrix associated with the linear system to be solved in order to compute the flow transfer factors.
     */
    private double getAlphaMatrixValue(LfBranch lfBranch, ClosedBranchSide1DcFlowEquationTerm p1, AbstractComputedElement element, boolean onDiagonal) {
        if (element instanceof ComputedContingencyElement) {
            double deltaX = onDiagonal ? 1d / calculatePower(lfBranch) : 0d;
            return deltaX - (contingenciesStates.get(p1.getPh1Var().getRow(), element.getComputedElementIndex())
                    - contingenciesStates.get(p1.getPh2Var().getRow(), element.getComputedElementIndex()));
        } else {
            double deltaX = onDiagonal ? computeDeltaXForDiagonalValues(lfBranch, element) : 0;
            return deltaX - (actionsStates.get(p1.getPh1Var().getRow(), element.getComputedElementIndex())
                    - actionsStates.get(p1.getPh2Var().getRow(), element.getComputedElementIndex()));
        }
    }

    /**
     * Compute the flow transfer factors needed to calculate the post-contingency state values.
     */
    private void setAlphas(DenseMatrix states, int columnState) {
        if (contingencyElements.size() + actionElements.size() == 1) {
            AbstractComputedElement element = actionElements.isEmpty() ? contingencyElements.iterator().next()
                    : actionElements.iterator().next();
            LfBranch lfBranch = element.getLfBranch();
            ClosedBranchSide1DcFlowEquationTerm p1 = element.getLfBranchEquation();

            // we solve a*alpha = b
            double a = getAlphaMatrixValue(lfBranch, p1, element, true);
            double b = getAlphaRhsValue(states, p1, columnState, element);
            element.setAlphaForWoodburyComputation(b / a);
        } else {
            // set local indexes of computed elements to use them in small matrix computation
            AbstractComputedElement.setLocalIndexes(contingencyElements);
            AbstractComputedElement.setLocalIndexes(actionElements);
            int size = contingencyElements.size() + actionElements.size();
            DenseMatrix rhs = new DenseMatrix(size, 1);
            DenseMatrix matrix = new DenseMatrix(size, size);

            for (ComputedContingencyElement contingencyElement : contingencyElements) {
                int i = contingencyElement.getLocalIndex();
                LfBranch lfBranch = contingencyElement.getLfBranch();
                ClosedBranchSide1DcFlowEquationTerm p1 = contingencyElement.getLfBranchEquation();
                rhs.set(i, 0, getAlphaRhsValue(states, p1, columnState, contingencyElement));

                // loop on contingencies to fill top-left quadrant of the matrix
                for (ComputedContingencyElement contingencyElement2 : contingencyElements) {
                    int j = contingencyElement2.getLocalIndex();
                    double value = getAlphaMatrixValue(lfBranch, p1, contingencyElement2, i == j);
                    matrix.set(i, j, value);
                }

                // loop on actions to fill top-right quadrant of the matrix
                for (AbstractComputedElement actionElement : actionElements) {
                    int j = contingencyElements.size() + actionElement.getLocalIndex();
                    double value = getAlphaMatrixValue(lfBranch, p1, actionElement, false);
                    matrix.set(i, j, value);
                }
            }

            for (AbstractComputedElement actionElement : actionElements) {
                int i = contingencyElements.size() + actionElement.getLocalIndex();
                LfBranch lfBranch = actionElement.getLfBranch();
                ClosedBranchSide1DcFlowEquationTerm p1 = actionElement.getLfBranchEquation();
                rhs.set(i, 0, getAlphaRhsValue(states, p1, columnState, actionElement));

                // loop on contingencies to fill bottom-left quadrant of the matrix
                for (ComputedContingencyElement contingencyElement : contingencyElements) {
                    int j = contingencyElement.getLocalIndex();
                    double value = getAlphaMatrixValue(lfBranch, p1, contingencyElement, false);
                    matrix.set(i, j, value);
                }

                // loop on actions to fill bottom-right quadrant of the matrix
                for (AbstractComputedElement actionElement2 : actionElements) {
                    int j = contingencyElements.size() + actionElement2.getLocalIndex();
                    double value = getAlphaMatrixValue(lfBranch, p1, actionElement2, i == j);
                    matrix.set(i, j, value);
                }
            }
            try (LUDecomposition lu = matrix.decomposeLU()) {
                lu.solve(rhs); // rhs now contains state matrix
            }
            contingencyElements.forEach(element -> element.setAlphaForWoodburyComputation(rhs.get(element.getLocalIndex(), 0)));
            actionElements.forEach(element -> element.setAlphaForWoodburyComputation(rhs.get(contingencyElements.size() + element.getLocalIndex(), 0)));
        }
    }

    /**
     * Calculate post-contingency states values by modifying pre-contingency states values, using some flow transfer factors (alphas).
     */
    public void toPostContingencyStates(DenseMatrix preContingencyStates) {
        Objects.requireNonNull(preContingencyStates);

        for (int columnIndex = 0; columnIndex < preContingencyStates.getColumnCount(); columnIndex++) {
            setAlphas(preContingencyStates, columnIndex);
            for (int rowIndex = 0; rowIndex < preContingencyStates.getRowCount(); rowIndex++) {
                double postContingencyValue = preContingencyStates.get(rowIndex, columnIndex);
                for (ComputedContingencyElement contingencyElement : contingencyElements) {
                    postContingencyValue += contingencyElement.getAlphaForWoodburyComputation()
                            * contingenciesStates.get(rowIndex, contingencyElement.getComputedElementIndex());
                }
                preContingencyStates.set(rowIndex, columnIndex, postContingencyValue);
            }
        }
    }

    /**
     * Calculate post-contingency and post-actions states values by modifying pre-contingency states values, using some flow transfer factors (alphas).
     */
    public void toPostContingencyAndOperatorStrategyStates(double[] preContingencyStates) {
        Objects.requireNonNull(preContingencyStates);
        setAlphas(new DenseMatrix(preContingencyStates.length, 1, preContingencyStates), 0);
        for (int rowIndex = 0; rowIndex < preContingencyStates.length; rowIndex++) {
            double postContingencyAndOperatorStrategyValue = preContingencyStates[rowIndex];
            for (ComputedContingencyElement contingencyElement : contingencyElements) {
                postContingencyAndOperatorStrategyValue += contingencyElement.getAlphaForWoodburyComputation()
                        * contingenciesStates.get(rowIndex, contingencyElement.getComputedElementIndex());
            }
            for (AbstractComputedElement actionElement : actionElements) {
                postContingencyAndOperatorStrategyValue += actionElement.getAlphaForWoodburyComputation()
                        * actionsStates.get(rowIndex, actionElement.getComputedElementIndex());
            }
            preContingencyStates[rowIndex] = postContingencyAndOperatorStrategyValue;
        }
    }
}