---
title: IEEE 68-Bus Network Benchmark
url: https://www.emergentmind.com/topics/ieee-68-bus-network
type: topic
---

# IEEE 68-Bus Network Benchmark

The IEEE 68-Bus Network, identified in the cited literature as the IEEE 68-bus New England/New York interconnection system or the IEEE 68-bus New England–New York benchmark, is a transmission-network test system used primarily for dynamic security, control, and stability studies rather than for static power-flow benchmarking alone. In the cited works, it is described as a system with **16 generator buses** and **52 load buses**, and, in one cascading-failure study, as a **5-area** network with **87 lines** [1702.03477], [2205.00103]. Across these studies, the benchmark functions as a common experimental substrate for stochastic load-side frequency regulation, passivity-based decentralized stability analysis, fast cascading-failure simulation, and PMU-based wide-area voltage control, while detailed bus, line, and machine data are often delegated to external benchmark sources or the Power System Toolbox rather than reproduced in full [1702.03477], [1809.09894], [2205.00103], [2102.05156].

## 1. Benchmark identity and reported descriptors

The papers provide a consistent high-level identification of the network, but not a single self-contained canonical specification. The most explicit descriptors reported in the cited literature are summarized below.

| Descriptor | Reported value | Source context |
|---|---|---|
| System name | IEEE 68-bus New England/New York interconnection | Stochastic load-side frequency control |
| System name | IEEE 68-bus NE-NY benchmark | Cascading-failure simulation |
| Buses | 16 generator buses, 52 load buses | Frequency-control study |
| Areas | 5 | Cascading-failure study |
| Lines | 87 | Cascading-failure study |
| Generators | 16 | Cascading-failure study |

The benchmark is therefore identifiable at the level of network size and intended use, but the cited papers repeatedly stop short of full numerical reconstruction. The stochastic frequency-control study states that the “relevant data” are taken from the **Power System Toolbox** data files and does not reproduce the full bus, line, or machine table [1702.03477]. The cascading-failure study likewise states that a detailed description is available in an external source and does not provide complete bus data, line parameter tables, transformer data, generator dynamic parameters, or the exact initiating contingencies for each Monte Carlo trial [2205.00103]. The PMU-based voltage-control paper explicitly gives only limited setup detail for the 68-bus case and does not specify generator/load counts, inter-area structure, exact FACTS parameters, or a 68-bus-specific missing-PMU or topology-change experiment [2102.05156].

This documented incompleteness is significant. It means that the IEEE 68-bus network, as it appears in these studies, is better understood as a **benchmark family anchored by a common NE-NY test case** than as a fully specified dataset within any one of the cited papers. A plausible implication is that reproducibility depends not only on the benchmark name but also on the surrounding modeling stack: device models, controller assumptions, relay settings, and data-source conventions.

## 2. Modeling representations used on the IEEE 68-bus network

The cited literature instantiates the IEEE 68-bus network through several distinct model classes, each tailored to a different research question.

In the stochastic load-side frequency-control study, the bus-level dynamics are written as
\[
\dot \omega_j=-\frac{1}{M_j} (\hat d_j +d_j-P_j^m+P_j^{out}-P_j^{in}),\qquad \forall j\in {\cal G}
\]
\[
0=\hat d_j +d_j-P_j^m+P_j^{out}-P_j^{in},\qquad \forall j\in {\cal L}
\]
\[
\dot P_{ij}=W_{ij}(\omega_i-\omega_j),\qquad \forall (i,j)\in {\cal E},
\]
with
\[
W_{ij}:=3\frac{|V_i||V_j|}{X_{ij}\cos(\theta_i^0-\theta_j^0)}.
\]
After eliminating the load-bus algebraic equation, the implemented 68-bus case study uses a reduced deterministic model in \((\omega_G,P)\), then augments it with multiplicative stochasticity in selected line weights [1702.03477].

In the cascading-failure study, the network is represented by nonlinear DAEs with discrete relay and breaker states,
\[
\dot{x} = f(x,V,z),\qquad
0 = I(x,V,z) - Y_N(z)V,\qquad
0 \succ h(x,V,z),
\]
where \(x\) is the dynamic state vector, \(V\) the real and imaginary parts of bus-voltage phasors, and \(z\) the discrete protection-state vector [2205.00103]. This is the most detailed of the cited formulations and supports relay-driven topology changes, islanding, and device-level protections.

In the passivity-based stability paper, the network is recast in the **system reference frame** as a static MIMO map,
\[
\begin{bmatrix} I_a \\ I_b \end{bmatrix}
=
\begin{bmatrix} G_n  & -B_n \\ B_n & G_n \end{bmatrix}
\begin{bmatrix} V_a \\ V_b \end{bmatrix}
=
H_{2n}
\begin{bmatrix} V_a \\ V_b \end{bmatrix},
\]
while each bus is modeled as a \(2\)-input/\(2\)-output dynamical subsystem with \(u_i=[-I_{a,i},-I_{b,i}]\) and \(y_i=[V_{a,i},V_{b,i}]\) [1809.09894]. This formulation is expressly designed to preserve passivity even when network conductance is nonzero.

In the PMU-based wide-area voltage-control study, the relevant dynamics are those of dynamic load buses, which are modeled by
\[
\dot{\theta}_{k} = \frac{1}{\tau_{\theta_k}}(P_k-P_k^s),\qquad
\dot{V}_k = \frac{1}{\tau_{V_k}}(Q_k-Q_k^s),
\]
then linearized into a vector Ornstein–Uhlenbeck process
\[
d\mathbf{x}=A\mathbf{x}\,dt+H\,d\boldsymbol{\xi},
\qquad
\mathbf{x}= \begin{bmatrix} \boldsymbol{\theta}\\ \mathbf{V} \end{bmatrix}.
\]
The estimated state matrix \(\hat A\) yields scaled Jacobian blocks such as \(\hat{J}_{\mathbf{Q}\mathbf{V}}\), which are then used for online control synthesis [2102.05156].

Taken together, these formulations show that the IEEE 68-bus network is not tied to a unique mathematical abstraction. It supports reduced electromechanical models, stochastic parametric models, full relay-driven DAE simulations, passivity-based input-output representations, and ambient-data-driven sensitivity estimation. This suggests that the benchmark’s practical value lies in its ability to bridge traditionally separate literatures: frequency control, oscillatory stability, cascading failures, and wide-area voltage control.

## 3. Stochastic load-side frequency regulation and fragility

One cited paper uses the IEEE 68-bus network as its principal numerical platform for analyzing the fragility of decentralized load-side frequency control under renewable-induced stochastic parametric uncertainty [1702.03477]. In that study, the system is explicitly described as containing **induction motor loads**, **constant-power loads**, and **controllable loads**. More specifically, it reports **29 induction motor loads that are frequency sensitive**, **35 controllable loads**, and the remaining loads as **uncontrollable frequency-insensitive loads** [1702.03477].

The implemented control law is a simplified quadratic-disutility formulation. Starting from
\[
c_j(d_j)=\frac{d_j^2}{2\alpha_j},
\]
with saturation neglected, the decentralized load law becomes
\[
d_j=\alpha_j\omega_j.
\]
The interpretation reported in the paper is that higher cost means less willingness or effectiveness of controllable loads to change, and larger cost values reduce the tolerable uncertainty [1702.03477].

The uncertainty model is not additive power noise. Instead, renewable penetration is represented as **multiplicative parametric uncertainty in transmission coupling coefficients** through stochastic bus-voltage products on selected links:
\[
|V_i||V_j|= 1+\sigma d\xi_{k}, \qquad \forall (i,j)\in{\cal S},
\]
so that
\[
W_{ij}=W_{ij}^0+\sigma W_{ij}^0 d\xi_k,
\qquad
W_{ij}^0:=3\frac{1}{X_{ij}\cos (\theta_i^0-\theta_j^0)}.
\]
For the 68-bus implementation, renewable replacement is placed at buses **54, 55, 56, 60, 63, 64, and 65**, with connecting buses **6, 10, 19, 25, 32, 36, and 52**, giving **\(s=7\)** uncertain links or locations [1702.03477].

After affine stochastic reformulation and coordinate transformation, the crucial multiplicative-noise subsystem becomes
\[
dx = {\cal A} x\,dt + \sum_{k=1}^{s} \sigma B_k C_k x\, d\xi_k.
\]
The mean-square stability condition is then written in robust-control form, and the critical uncertainty level is reported as
\[
\sigma_*=\frac{1}{\rho(\hat G)},
\qquad
\sigma_*^2=\frac{1}{\rho(\hat G)^2}.
\]
For the 68-bus case, the central quantitative result is that the critical variance is **very small**, on the order of \(10^{-3}\), with maximum reported value
\[
\sigma_*^2 = 1.9\times 10^{-3}
\]
at
\[
\alpha = 0.5.
\]
The paper further states that the critical variance decreases as the controllable-load cost increases and decreases with increasing renewable penetration [1702.03477].

The reported instability mechanism is operationally important. When
\[
\sigma^2 > \sigma_*^2,
\]
the system becomes **mean-square unstable**, with unbounded frequency growth or oscillation even while bus voltages remain within **0.95–1.05 p.u.** in the illustrated unstable case [1702.03477]. A step change in power is applied after **\(t=10\) s**, and frequency at **generator bus 53** is plotted. The paper states that, under this stochasticity, the decentralized controller is ineffective: frequencies oscillate, leave the operating range, and continue oscillating.

The same study also specifies the renewable-replacement parameter modifications used in the 68-bus experiments. Original generator inertia values lie between **1 and 5**, while renewable-bus inertia is set to **0.5**; original generator damping lies between **0 and 5**, while renewable-bus damping is set to **6**. The reported trends remain consistent for renewable-bus inertia in the range **0.5–1** and damping in the range **5–6** [1702.03477].

These results make the IEEE 68-bus network a benchmark for a specific control-theoretic lesson: **acceptable voltage magnitudes do not guarantee acceptable stochastic frequency behavior** under decentralized load-side regulation. The paper’s emphasis is therefore not on the benchmark’s static structure, but on the interaction between renewable-induced parametric variability and decentralized primary-frequency control.

## 4. Passivity-based decentralized stability and controller redesign

Another cited study uses the IEEE 68-bus network as one of its two principal test cases for a **system-reference-frame, passivity-based stability analysis and control framework** [1809.09894]. The network is referred to there as the **IEEE New York / New England 68-bus interconnection system**, and simulations are carried out in the **Power System Toolbox (PST)** [1809.09894].

The generator buses are modeled with the **fourth-order synchronous machine model** augmented by **turbine governors**, **exciters**, and **power system stabilizers (PSSs)**, using models available in the PST manual. For each generator bus, the authors solve a power flow, linearize the local dynamics about the equilibrium, and test passivity using either LMIs or frequency-domain positivity of \(G_i(j\omega)+G_i^T(-j\omega)\). A study-specific detail is that the frame angle \(\delta_i\) is obtained from each generator’s q-axis transient emf \(E'_{q,i}\) rather than the q-axis bus voltage \(V_{q,i}\) [1809.09894].

The theoretical motivation is that, in the system reference frame, the network remains passive even with losses:
\[
u^\textrm{T} y
=
[V_a^\textrm{T}\ \ V_b^\textrm{T}]
\begin{bmatrix} I_a \\ I_b \end{bmatrix}
=
V_a^\textrm{T} G_n V_a + V_b^\textrm{T} G_n V_b \geq 0,
\]
because \(G_n\) is positive semidefinite [1809.09894]. This removes the need for the common lossless-network simplification in decentralized passivity arguments.

The 68-bus experiments consider four controller scenarios:  
1. **No turbine governor / no exciter / no PSS**  
2. **Turbine governor / no exciter / no PSS**  
3. **Turbine governor / exciter / no PSS**  
4. **Turbine governor / exciter / PSS** [1809.09894]

The reported outcomes are sharply differentiated. In cases **(i)** and **(ii)**, generator buses are **non-passive**, and after a sudden load change **the power system collapses**. In case **(iii)**, adding excitation control significantly damps the generators; the system remains **stable but oscillatory**. Specifically, generators at buses **53, 59, 61, and 64** become passive, while the remaining generator buses remain non-passive. In case **(iv)**, adding **PSSs** further improves passivity and robustness; all generator buses become passive **except 58, 62, 63, and 65** [1809.09894].

The disturbance scenarios are also explicitly quantified. A sudden change of **1 pu** is applied at load buses **1, 9, and 18**, corresponding to a **total change of 300 MW**. The paper also reports that the IEEE 68-bus system has a **total load of 18.33 GW**. Frequency deviation and voltage deviation are then plotted at **bus 27** [1809.09894]. In a second study, after exciter redesign, a larger disturbance is applied: a sudden change of **3 pu** at load buses **1, 9, 18, 20, 37, and 42**, corresponding to **1800 MW**, which the paper states is **10% of the grid-connected load** [1809.09894].

The redesign mechanism centers on a modified exciter transfer function,
\[
E_{f,i}=\frac{K_a}{1+sT_a} \frac{1+sT_c}{1+sT_b} E_{g,i},
\]
where \((1+sT_c)/(1+sT_b)\) is an added **phase lag compensator** [1809.09894]. For **generator bus 53**, passivity is reported to be violated over approximately
\[
\omega \in [0.3, 3]\ \text{rad/s},
\]
and the lag-compensator tuning is shown to restore positivity of the relevant eigenvalue locus [1809.09894]. The paper provides parameter values of the modified exciters for generator buses **53–68**, with buses **64–68** marked as cases where no modification was applied.

The study also links passivity to classical eigenvalue analysis. With a simple exciter that still violates passivity, the linearized IEEE 68-bus system is reported to be **small-signal unstable**, with at least one eigenvalue in the **right half plane**. With the modified exciter including a lag compensator, all eigenvalues move to the **left half plane** and damping improves [1809.09894]. At the same time, the paper explicitly notes that passivity is a **sufficient**, not necessary, decentralized condition: the system may remain stable even when some buses are non-passive.

This use of the IEEE 68-bus network is notable because it treats the benchmark as a test of **local controller redesign under lossy-network coupling**, not simply as a modal-analysis benchmark. The network becomes a medium for evaluating whether local passivation correlates with improved wide-area robustness.

## 5. Cascading-failure simulation and oscillatory-instability benchmarking

The IEEE 68-bus NE-NY benchmark is the key oscillatory-instability case in a study on fast cascading-failure simulation using a predictor-corrector variant of the implicit Backward Euler Method, denoted **BEM-PC** [2205.00103]. Unlike the IEEE 118-bus and Polish systems in the same paper, which were artificially modified by assigning negative damping to some machines, the 68-bus benchmark is selected because it **naturally exhibits interarea oscillatory modes** [2205.00103].

The dynamic model used for the 68-bus case includes a **4th-order synchronous generator model** with states \(E_q'\), \(E_d'\), \(\delta\), and \(\Delta \omega\), together with a **first-order governor**, **static exciter**, **static constant-power loads**, and **dynamic loads modeled as synchronous condensers** [2205.00103]. Protection logic includes undervoltage load shedding, overcurrent relays, generator out-of-step protections, tripping of generators with non-oscillatory instability, and a **special protection scheme (SPS)** for oscillatory instability. For the 68-bus system specifically, the overcurrent relay averaging window is set to
\[
T_w^{OC} = 4 \text{ s},
\]
rather than the shorter value used in the other systems, because of the benchmark’s **low-frequency interarea modes** [2205.00103].

The simulation campaign comprises **500 Monte Carlo simulations**, each initiated by **two random initial line outages** [2205.00103]. This makes the 68-bus case the paper’s main benchmark for the numerical pathology called **hyperstability**, in which plain BEM can converge to a post-disturbance unstable equilibrium instead of revealing unstable oscillations. The predictor-corrector remedy proceeds by running BEM, extracting a reduced linearized system matrix
\[
A = P_{11} + P_{12}P_{22}^{-1}P_{21},
\]
performing eigendecomposition of \(A\), and using the right eigenvector of an unstable oscillatory mode to identify the participating generators [2205.00103].

The modal results reported for the 68-bus benchmark are precise. The least damped predisturbance mode is
\[
-0.0804 \pm 2.4474j,
\]
while unstable cascade cases yield an unstable mode estimated as
\[
0.0061 \pm 2.3740j.
\]
The corresponding speed mode shape shows that generators **G14–G16 oscillate against the rest of the generators** [2205.00103]. Once this unstable mode is detected, the predefined corrective action is to **trip line 41–42** **5 seconds after the latest event** [2205.00103].

Hyperstability is not reported as a rare corner case. Table VII in the paper gives **82 cases** with hyperstability detected and **418 cases** without hyperstability, so **16.4%** of the 500 runs encounter the issue [2205.00103]. Despite this, the corrected BEM-PC method closely matches the trapezoidal-method benchmark. For the 68-bus case, the paper reports mean end-of-cascade error **0.132** in bus states, **0.032** in machine states, and **0.138** in line states. The path-agreement measure
\[
R = \frac{1}{|C|} \sum_{i=1}^{|C|} \frac{|A_i \cap B_i|}{|A_i \cup B_i|}
\]
has mean **0.997**, minimum **0.428**, maximum **1**, and median **1** [2205.00103]. Runtime ratio, defined as TM relative to BEM-PC, has mean **19.687**, minimum **0.235**, maximum **120.737**, and median **15.582**, yielding an average speedup of about **20×** [2205.00103].

The same paper also uses the benchmark to compare model fidelity. For the 68-bus system, the **4th-order model** has most poorly damped mode
\[
-0.0804 \pm 2.4474j,
\]
whereas the **classical model** has most poorly damped mode
\[
-0.0718 \pm 5.0125j,
\]
with materially different mode shapes: in the 4th-order model, **G14–G16 oscillate against generators in NETS and NYPS**, while in the classical model **G15 oscillates against G14 and G16** [2205.00103]. Table XI further reports nonzero final-demand-loss error in **79** NE-NY cases for the TM classical model and in **297** cases for AC-QSS; nonzero line-outage error occurs in **62** classical-model cases and **322** AC-QSS cases [2205.00103].

This body of results positions the IEEE 68-bus network as a benchmark not only for oscillatory instability itself, but also for the fidelity of fast dynamic cascade simulators and the adequacy of reduced-order generator models.

## 6. PMU-based wide-area voltage control

A further cited study uses the IEEE 68-bus network as the **larger-scale validation case** for an **online model-free wide-area voltage control (WAVC)** method based on PMU data and FACTS devices [2102.05156]. The paper explicitly contrasts the 68-bus role with that of the IEEE 39-bus system: the 39-bus case carries most of the detailed stress tests, while the 68-bus case is used more narrowly as a **large-scale feasibility demonstration** [2102.05156].

The control architecture assumes PMU measurements at the relevant dynamic load buses and uses ambient data to estimate sensitivity matrices online. For the 68-bus study, the paper states that **300 s PMU measurements** are collected to estimate the sensitivity matrices in \(J\), and the algorithm remarks specify a sampling frequency of **60 Hz** [2102.05156]. The core estimation step is
\[
\hat{A}=\frac{1}{\Delta t}\log\!\left(\hat{G}(\Delta t)\hat{C}^{-1}\right),
\]
with
\[
\hat{A} =
\begin{bmatrix}
\hat{T}_{\theta}^{-1}\hat{J}_{\mathbf{P}\boldsymbol{\theta}} &
\hat{T}_{\theta}^{-1}\hat{J}_{\mathbf{P}\mathbf{V}} \\
\hat{T}_{V}^{-1}\hat{J}_{\mathbf{Q}\boldsymbol{\theta}} &
\hat{T}_{V}^{-1}\hat{J}_{\mathbf{Q}\mathbf{V}}
\end{bmatrix},
\]
so that the sensitivity information needed for control is extracted from PMU covariance and lag-correlation statistics [2102.05156].

The 68-bus figure reported in the paper focuses specifically on the accuracy of the estimated
\[
J_{\mathbf{Q}\mathbf{V}}.
\]
The text states that comparison between the true and estimated \(J_{\mathbf{Q}\mathbf{V}}\) shows **good accuracy of the estimation** [2102.05156]. Control is then implemented through FACTS devices interpreted as SVCs at **five voltage-controlled buses**:
\[
\{20,25,29,41,42\}.
\]
Voltage response is illustrated at **voltage-uncontrolled Bus 21**, and the paper states that the proposed algorithm can effectively restore voltages using the estimated matrix \(\hat{J}_{\mathbf{Q}\mathbf{V}}\), with response **very close** to that obtained using the true \(J_{\mathbf{Q}\mathbf{V}}\) [2102.05156].

The underlying control synthesis is built from the inverse sensitivity matrix
\[
S=J^{-1}
=
\begin{bmatrix}
S_{\boldsymbol{\theta}\mathbf{P}} & S_{\boldsymbol{\theta}\mathbf{Q}} \\
S_{\mathbf{V}\mathbf{P}} & S_{\mathbf{V}\mathbf{Q}}
\end{bmatrix},
\]
together with the controlled/uncontrolled partition and the online optimization
\[
\min_{\Delta \mathbf{V}_c(t_{i+1})} \|\Delta\mathbf{V}_u(t_{i+1})\|_{\infty}
\]
subject to voltage and reactive-power constraints on the controlled buses [2102.05156]. Because SVCs are shunt reactive devices, the formulation sets \(\Delta \mathbf{P}_c=\mathbf{0}\).

Equally important are the stated non-results. The paper does **not** report, for the 68-bus case, a missing-PMU study, a measurement-noise study, a topology-change study, a performance-index table, or a direct model-free versus model-based comparison after topology change; those detailed robustness analyses are reported only on the 39-bus system [2102.05156]. Thus, for the IEEE 68-bus network, the paper’s contribution is best read as evidence of **scalability and feasibility**, not as an exhaustive characterization of robustness under all sensing and topology contingencies.

## 7. Scope, extrapolation, and unresolved specification

The cited literature establishes the IEEE 68-bus network as a versatile research benchmark, but it also delineates clear boundaries on what is directly specified and what remains external.

First, no single cited paper provides a complete standalone numerical definition of the benchmark. The stochastic frequency-control paper does not enumerate the full line list, line reactances \(X_{ij}\), nominal angles \(\theta_i^0\), complete generator inertias and damping constants bus by bus, exact controllable-load allocation by bus, the exact incidence matrices \(E_G,E_L\), or the full deterministic operating point \(u^*\) [1702.03477]. The cascading-failure paper does not provide complete bus data, full line parameter tables, transformer data, exciter/governor parameter tables, or all relay thresholds for the 68-bus case [2205.00103]. The PMU-based voltage-control paper explicitly leaves unspecified many canonical benchmark properties and several 68-bus-specific controller details [2102.05156]. This suggests that, in practice, serious use of the IEEE 68-bus network requires coupling the published methodology to an external benchmark dataset or toolbox implementation.

Second, some methodologies discussed alongside IEEE-style systems are not directly validated on the IEEE 68-bus network at all. The paper on **optimal slack-bus selection** does **not** explicitly study the IEEE 68-bus network; it reports results for **IEEE-57**, **IEEE-118**, **Pegase-89**, and **Pegase-1354**, and states that any application to IEEE 68 must be treated as an **extrapolation of the method**, not as a reported result [1707.02845]. Its main analytical criterion is to choose the unique slack bus
\[
g=\arg\min_g\left(-\boldsymbol{\Omega}_g^\top\mathbf P\right),
\]
where \(\boldsymbol{\Omega}_g\) is the vector of resistance distances from generator \(g\), computed from a weighted Laplacian built from the lossless AC operating point [1707.02845]. The same paper argues that, to order \(\mathcal O(\gamma^2)\), a **single slack bus is generically optimal**, while distributed slack becomes relevant only through higher-order corrections or near-degeneracy among candidate generators [1707.02845]. For IEEE 68, however, the paper provides **no actual numerical values, no figures, no tables, and no explicit bus ranking** [1707.02845].

Third, the benchmark’s reported uses emphasize different subsystems. In the stochastic frequency-control study, the decisive issue is multiplicative uncertainty in line-coupling coefficients induced by renewable replacement [1702.03477]. In the passivity-based paper, the focus is generator-bus passivation in a lossy system-reference-frame formulation [1809.09894]. In the cascading-failure study, the key feature is the network’s naturally occurring interarea oscillatory instability under relay-driven contingencies [2205.00103]. In the PMU-based voltage-control paper, the benchmark is mainly a large-scale test of online sensitivity estimation and SVC reference updates [2102.05156]. A plausible implication is that the IEEE 68-bus network is less a single “problem instance” than a common dynamic scaffold on which different methodological communities test distinct hypotheses.

Within the scope of the cited literature, the IEEE 68-bus network is therefore best characterized as a **multi-purpose NE-NY dynamic benchmark**: large enough to exhibit interarea modes and wide-area control structure, yet compact enough to support detailed DAE simulation, stochastic stability analysis, passivity-based redesign, and measurement-driven control synthesis. Its enduring research role arises not from one fixed published parameter table in these papers, but from its repeated reuse as a technically rich common reference system.

Source: https://www.emergentmind.com/topics/ieee-68-bus-network