Network Formulation and Load-Flow Analysis

Introduction

Load-flow, or power-flow, analysis determines the steady-state voltages, power transfers, losses and equipment loading of an electrical network. It is fundamental to system planning, operational security, voltage-control studies, contingency assessment and the integration of renewable generation.

The analysis begins with a network model, normally represented by the bus-admittance matrix, Y_bus. Nonlinear algebraic equations then relate bus voltages to specified real and reactive powers. Gauss–Seidel, Newton–Raphson and fast-decoupled algorithms solve these equations numerically.

All numerical quantities in this chapter are in per unit unless otherwise stated. The network is assumed to operate in balanced sinusoidal steady state, allowing a positive-sequence single-phase equivalent.

Learning Objectives

After studying this chapter, students should be able to:

• construct Y_bus by inspection and systematic branch stamping;
• represent transmission lines and off-nominal complex transformer taps;
• obtain and construct the bus-impedance matrix Z_bus;
• classify slack, PV and PQ buses;
• formulate the nonlinear power-flow equations;
• apply Gauss–Seidel, Newton–Raphson and fast-decoupled methods;
• enforce generator reactive-power limits correctly;
• calculate branch currents, losses and equipment loading; and
• interpret engineering-software results and diagnose convergence problems.

1. Network Representation

A bus is an electrically common node at which generators, loads, shunts or branches are connected. Numbering the buses from 1 to n, Kirchhoff’s current law gives

or, in matrix form,

Here, V_i is the complex bus voltage in pu or volts, I_i is the net current injected into the network in pu or amperes, and Y_ik is an admittance in pu or siemens. Positive current and power denote injection into the network.

Figure: A bus-branch model from which the network admittance matrix is assembled.

1.1 Primitive branch model

For a transmission line between buses i and j, let z_ij=r_ij+jx_ij be its series impedance and

its series admittance. A π-equivalent model places half of the total charging admittance, usually jb_c/2, at each end. For a line without a transformer, its contributions are

A shunt admittance connected directly at bus i is added only to Y_ii. Mutual coupling requires a more general primitive-admittance formulation and cannot be represented by independent series stamps alone.

1.2 Complex transformer tap stamping

Consider a branch whose off-nominal tap is on the bus-i side. Define

where \tau is the tap magnitude and \phi is the phase shift. With series admittance y and total charging admittance jb_c, the exact stamp under this convention is

The conjugate t^* is essential. A phase-shifting transformer generally makes Y_ij\ne Y_ji, although the physical branch remains reciprocal when transformer equations and reference directions are treated consistently.

Numerical stamp. Let y=10-j20, b_c=0.04 and t=1.05\angle10^\circ. The branch contributions are approximately

These values should be added to any contributions already present in the four matrix locations.

2. Constructing the Bus-Admittance Matrix

For a network without mutual coupling, inspection gives

The second expression applies directly only to branches with unity taps. Complex-tap branches must use the preceding stamp.

Worked Example 1: Three-bus Y_bus

A three-bus network has series impedances z₁₂=j0.20, z₁₃=j0.25 and z₂₃=j0.40. Neglect line charging.

The branch admittances are

Therefore,

Each row sums to zero because there are no shunts to ground. This row-sum check is useful, but it does not apply when shunts or certain transformer models are present.

3. Bus-Impedance Matrix

The bus-impedance matrix is defined by

A reference bus must be removed before inverting an otherwise ungrounded network matrix. Elements of Z_bus have useful interpretations: Z_ii is the driving-point impedance at bus i, while Z_ij is a transfer impedance. The matrix is central to fault analysis and network-equivalent calculations.

Direct inversion is suitable for small systems. For large or incrementally assembled networks, the Zbus building algorithm is more efficient conceptually. Let \mathbf{Z} denote the existing matrix.

The denominators must be nonzero. These formulas assume passive branch additions and consistent bus ordering.

Worked Example 2: Incremental Zbus construction

First connect bus 1 to the reference through j0.2:

Next connect new bus 2 to bus 1 through j0.3:

Finally, add a link from bus 2 to the reference through j0.5. Since z+Z₂₂=j1.0,

This example includes both radial bus additions and a link addition; it is not merely a direct inversion.

4. Bus Classification and Power Specifications

At each bus there are four principal variables: voltage magnitude |V|, voltage angle \delta, real-power injection P and reactive-power injection Q. Only two are normally specified.

Loads are negative injections under the adopted convention. Thus a load consuming P_L+jQ_L is entered as P=-P_L and Q=-Q_L.

A PV bus can maintain its specified voltage only while generator reactive power remains within limits. If Q_G violates a limit, it is fixed at that limit and the bus is temporarily treated as a PQ bus. After convergence, software may test whether restoration to PV status is feasible.

5. Nonlinear Power-Flow Equations

Complex power injected at bus i is

Let Y_ik=G_ik+jB_ik and V_i=|V_i|e^j\delta_i. Substitution gives

These equations are nonlinear because voltage magnitudes and trigonometric functions of voltage angles are unknown.

6. Gauss–Seidel Method

For PQ bus i, rearranging the nodal equation gives

The newest available voltages are used immediately. An acceleration factor \alpha may be applied through

usually with 1<\alpha<2, although an unsuitable value can cause divergence.

For a PV bus, compute reactive power from the current voltage estimate, enforce limits, update the voltage as if it were PQ, and then restore its specified magnitude while retaining the updated angle.

Worked Example 3: Correct two-bus solution

Bus 1 is a slack bus with V₁=1\angle0^\circ. Bus 2 supplies a load S_L=1+j0.5 through a lossless line z=j0.2, so y=-j5. At bus 2, P₂=-1 and Q₂=-0.5. Starting with V₂⁽⁰⁾=1+j0,

The next update is

Continuing the iteration gives the high-voltage solution

A low-voltage mathematical solution, approximately 0.1683-j0.2000, also exists. Normal operation selects the high-voltage branch, but the second solution illustrates the nonlinear nature of power-flow equations.

The current directed from bus 2 toward bus 1 is

Current is not numerically equal to complex power. Verification gives

The sending-end power is approximately S₁₂=1+j0.8417. The line has no real loss because its resistance is zero, but it absorbs reactive power

7. Newton–Raphson Method

Newton–Raphson linearizes the mismatch equations about the current estimate. Define

For non-slack angle variables and PQ-bus magnitude variables,

The four blocks are partial derivatives: H=\partial P/\partial\delta, N=\partial P/\partial|V|, M=\partial Q/\partial\delta and L=\partial Q/\partial|V|. The linear system is solved rather than explicitly inverting the Jacobian.

For i\ne k, representative Jacobian elements are

where \delta_ik=\delta_i-\delta_k. Diagonal elements are obtained by differentiating the corresponding bus equations or using established identities.

Newton–Raphson applied to the two-bus example

For the lossless two-bus system,

At the flat start, the Jacobian is diagonal with entries 5 and 5. Thus

The first updated magnitude is therefore |V₂|=0.9, not 0.8246.

Newton–Raphson has quadratic convergence near a nonsingular solution and is the standard general-purpose method for large transmission systems. Each iteration is computationally heavier than Gauss–Seidel, but many fewer iterations are normally required.

8. Fast-Decoupled Load Flow

Transmission systems commonly satisfy X\gg R, voltage-angle differences are modest, and voltage magnitudes remain near 1 pu. Under these conditions, real power is strongly coupled to angle and reactive power to voltage magnitude. Neglecting weak Jacobian couplings gives

Here \mathbf{B}' and \mathbf{B}'' are constant susceptance-based matrices formed according to the selected implementation. Their factorizations can be reused, reducing computation and memory.

The method is less reliable in low-voltage distribution networks, networks with high R/X ratios, heavily loaded systems, or systems containing many phase-shifting and power-electronic devices.

9. Convergence and Numerical Diagnostics

A common convergence criterion is

where \varepsilon is a specified mismatch tolerance in pu. Voltage changes may be monitored as an additional—not substitute—criterion.

Frequent causes of failure include:

• incorrect bus or branch data;
• disconnected islands without a slack bus;
• zero or extremely small impedances;
• poor initial voltages;
• infeasible real- or reactive-power schedules;
• inconsistent transformer-tap conventions;
• ill-conditioning near voltage collapse; and
• repeated PV–PQ switching at reactive limits.

A converged numerical result is not automatically a feasible operating state. Voltage limits, thermal ratings, generator capability and contingency requirements must also be checked.

10. Branch Flows, Losses and Equipment Loading

For a line without an off-nominal transformer, current from bus i to bus j is

The corresponding complex power is

Branch loss is

For a series impedance r+jx without shunt charging,

Real equipment loading should be compared with the applicable rating. Current loading is 100|I|/I_rated percent, whereas MVA loading is 100|S|/S_rated percent. The two should not be interchanged when voltage differs significantly from rated value.

A voltage profile plots |V_i| against bus or electrical distance. Low voltages often indicate insufficient reactive support, excessive transfer, unsuitable tap settings or high feeder impedance. High voltages may arise from light loading, line charging or excessive capacitive compensation.

11. Applied Case: Software-Assisted Voltage Improvement

Background. This is an illustrative case involving a six-bus industrial supply network. The receiving section includes induction-motor demand and a 20 MVA transformer. The planning limit is 0.95–1.05 pu, and continuous branch loading must remain below 100% of rating.

Problem. A Newton–Raphson study converges, but the lowest bus voltage is 0.923 pu and the transformer carries 106% of its continuous MVA rating. Total real-power loss is 2.1 MW.

Analysis. Engineers first validate base quantities, transformer tap orientation, generator limits and load signs. The software’s branch report shows that the transformer overload is caused by both real transfer and reactive demand. A sensitivity study indicates that local reactive compensation raises receiving-end voltage and reduces upstream current.

Approach. A switched capacitor bank is placed near the motor load, and the transformer tap is adjusted within its permitted range. The power flow is rerun with reactive limits enforced. An N−1 contingency study is then performed; the base-case solution alone is insufficient for approval.

Outcome. The revised base case satisfies voltage and continuous-loading limits. The contingency analysis identifies a separate post-outage overload, leading to an operational transfer limit rather than unjustified reliance on the base-case result.

Learning points. Software convergence is only the beginning of engineering interpretation. Model validation, limit enforcement, branch-direction checks and contingency assessment are necessary before decisions are made.

12. Software Practice, Governance and Contemporary Developments

Professional packages exploit sparse matrices, reordered factorizations, automatic controls and large equipment databases. A defensible workflow should:

1. select and document MVA and voltage bases;
2. validate topology, impedances, taps, phase shifts and shunts;
3. assign one valid reference per energized island;
4. confirm load and generation sign conventions;
5. solve with limits and controls initially simplified if necessary;
6. restore controls systematically and investigate warnings;
7. compare power balance, voltage limits, losses and ratings; and
8. archive assumptions, software version and study cases.

State-estimation data, digital twins and machine-learning tools increasingly assist model correction, contingency screening and initial-state prediction. AI-generated estimates must not replace physical equations, protection constraints or independent validation. Training data can omit rare stressed conditions, making purely data-driven recommendations unsafe near operating boundaries.

Utilities must also meet grid-code voltage requirements, generator capability limits and reliability criteria. Deliberately suppressing nonconvergent cases, manipulating ratings or presenting unverified outputs as secure operation is an ethical and governance failure. Cybersecurity is relevant because altered topology, tap or measurement data can produce plausible but dangerous results.

Chapter Summary

The bus-admittance matrix converts branch data into nodal current–voltage equations. Ordinary lines use diagonal and negative off-diagonal stamps, whereas off-nominal complex transformer taps require conjugate-aware contributions. The bus-impedance matrix can be obtained by reduced-matrix inversion or built incrementally through radial and link additions.

Slack, PV and PQ classifications identify which two bus variables are specified. The resulting nonlinear power equations may be solved by Gauss–Seidel, Newton–Raphson or fast-decoupled methods. Newton–Raphson is generally the most robust, while the fast-decoupled method is efficient when transmission-network assumptions hold.

Engineering interpretation requires correct conversion of voltages to branch currents and complex powers. Convergence must be supplemented by checks of power balance, reactive limits, voltage profile, losses, equipment loading and contingencies. Reliable software studies depend on validated models, documented assumptions and responsible professional judgment.

Key Terms

Bus: An electrically common node in a power network.

Bus-admittance matrix: Matrix relating bus-current injections to bus voltages.

Bus-impedance matrix: Inverse of the reduced bus-admittance matrix.

Driving-point impedance: Diagonal Zbus element relating a bus voltage to injection at the same bus.

Transfer impedance: Off-diagonal Zbus element relating voltage at one bus to injection at another.

Slack bus: Reference bus with specified voltage magnitude and angle.

PV bus: Generator bus with specified real power and voltage magnitude.

PQ bus: Bus with specified real and reactive power.

Power mismatch: Difference between specified and calculated bus power.

Jacobian: Matrix of power-equation partial derivatives used in Newton–Raphson iteration.

Flat start: Initial condition with non-slack voltages usually set to 1∠0° pu.

Reactive-power limit: Minimum or maximum reactive output permitted for a generator.

Complex tap ratio: Transformer ratio containing both magnitude and phase shift.

Voltage profile: Distribution of bus-voltage magnitudes across a network.

Fast-decoupled load flow: Approximate method exploiting active-power/angle and reactive-power/voltage coupling.

Branch loss: Sum of complex powers entering both ends of a branch.

Equipment loading: Current or apparent power expressed relative to an applicable rating.

Voltage collapse: Loss of a feasible or stable high-voltage operating solution.

Review Questions

Short Answer Questions

1. What information is contained in diagonal and off-diagonal elements of Y_bus?
2. Why must the reference bus be removed before inverting an ungrounded network admittance matrix?
3. Which variables are specified at a PV bus?
4. Why is a converged power flow not necessarily an acceptable operating condition?
5. State two assumptions supporting fast-decoupled load flow.

Long Answer Questions

1. Derive the real- and reactive-power injection equations from \mathbf{I}=\mathbf{Y}_bus\mathbf{V}.
2. Explain how line charging, shunts and a complex transformer tap are stamped into Y_bus.
3. Describe the Zbus building algorithm for radial and link additions.
4. Develop the Newton–Raphson load-flow formulation and explain the dimensions of its Jacobian blocks.
5. Compare Gauss–Seidel, Newton–Raphson and fast-decoupled methods in terms of convergence, storage and applicability.

Analytical / Application Questions

1. Construct Y_bus for a three-bus network containing two ordinary lines, one phase-shifting transformer and a shunt capacitor.
2. For the two-bus system in Worked Example 3, verify the high- and low-voltage solutions and calculate sending-end power and line reactive absorption.
3. A software study converges with one bus at 0.91 pu and a transformer at 104% loading. Develop a technically justified investigation and mitigation plan.

Multiple Choice Questions with Answers

Question 1

For a network without mutual coupling, which statement correctly describes a diagonal element of Y_bus?

A. It is always the negative of the bus voltage.

B. It is the sum of admittance contributions incident on the bus, including applicable shunts.

C. It is the sum of branch impedances connected to the bus.

D. It equals the corresponding diagonal element of Z_bus.

Correct Answer: B. It is the sum of admittance contributions incident on the bus, including applicable shunts.

Question 2

A transformer has tap t on the bus-i side and series admittance y. Neglecting charging, what is its contribution to Y_ij?

A. -y/t^*

B. -yt^*

C. y/|t|²

D. -y/t

Correct Answer: A. -y/t^*

Question 3

Which pair of variables is specified at a conventional PV bus?

A. Real power and reactive power

B. Voltage magnitude and voltage angle

C. Real power and voltage magnitude

D. Reactive power and voltage angle

Correct Answer: C. Real power and voltage magnitude

Question 4

In Worked Example 3, what is the first Gauss–Seidel update from the flat start?

A. 0.9000-j0.2000 pu

B. 0.8317-j0.2000 pu

C. 0.9000+j0.2000 pu

D. 0.8246-j0.1000 pu

Correct Answer: A. 0.9000-j0.2000 pu

Question 5

What is the high-voltage solution for bus 2 in the corrected two-bus example?

A. 0.2614\angle-49.91^\circ pu

B. 0.8554\angle-13.52^\circ pu

C. 1.0000\angle0^\circ pu

D. 0.9000\angle-20.00^\circ pu

Correct Answer: B. 0.8554\angle-13.52^\circ pu

Question 6

At the high-voltage solution of Worked Example 3, which statement is correct for current directed from bus 2 to bus 1?

A. I₂₁=1+j0.5 pu

B. I₂₁\approx-1+j0.8417 pu and -V₂I₂₁^*=1+j0.5 pu

C. I₂₁\approx0.8317-j0.2000 pu

D. I₂₁\approx-1-j0.8417 pu and V₂I₂₁^*=1+j0.5 pu

Correct Answer: B. I₂₁\approx-1+j0.8417 pu and -V₂I₂₁^*=1+j0.5 pu

Question 7

For the two-bus example, what voltage-magnitude estimate results from the first Newton–Raphson update from a flat start?

A. 0.8246 pu

B. 0.8554 pu

C. 0.9000 pu

D. 1.1000 pu

Correct Answer: C. 0.9000 pu

Question 8

A new bus p is connected to an existing bus k through impedance z. Which Zbus relation is correct?

A. Z_pp=z-Z_kk

B. Z_pp=1/(z+Z_kk)

C. Z_pp=Z_kk+z

D. Z_pp=Z_kkz

Correct Answer: C. Z_pp=Z_kk+z

Question 9

Under which condition is fast-decoupled load flow generally most appropriate?

A. A radial network with a high resistance-to-reactance ratio

B. A transmission network with X\gg R and voltage magnitudes near 1 pu

C. A disconnected network containing no reference bus

D. A network operating at the exact voltage-collapse point

Correct Answer: B. A transmission network with X\gg R and voltage magnitudes near 1 pu

Question 10

For a series branch with current magnitude |I| and impedance r+jx, neglecting shunt charging, what is its real-power loss?

A. |I|r

B. |I|²x

C. |I|²r

D. r/|I|²

Correct Answer: C. |I|²r