Literature Review Of Load Shedding Methods Computer Science Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

In chapter 1, a brief discussion about active distribution networks was presented. The importance of operation of islanded distribution networks was discussed. This chapter is intended to give the reader a better understanding of the load shedding methods currently applied and proposed over the years. However, it is assumed that the reader is familiar with basic power system engineering.

In section 2.2, the area of probability of islanding and the need for load shedding is discussed. To achieve this objective, existing load shedding methods are reviewed to understand their working principle, requirements, advantages and limitations. The main categories identified are the following (i) Manual / SCADA Load Shedding (ii) Load Shedding using thresholds, (iii) Adaptive Load Shedding, (iv) Intelligent Load Shedding and (v) Load Shedding Based on Static Optimisation which are described and discussed in sections 2.2 through 2.7 respectively. Finally a summary is given in section 2.8 from which a new load shedding method for an islanded distribution network able to address the limitations of existing methods will be proposed.

probability of islanding

There is by now a considerable amount of literature on load shedding. That most of the literature however deals with large interconnected systems. For smaller systems when a loss of mains / grid event occurs the islanded distribution network have different operating characteristics and restrictions that require different load shedding guidelines. These are due to the low inertia of the distributed generators, the limited spinning reserve and limited communication systems [0] - [0]. Load shedding is a practice used power system and serves as a function to try to arrest any frequency or voltage drop when a fault isolating part of the distribution network occurs.

Faults in power systems are inevitable, for various reasons such as adverse weather conditions, ageing and failure of equipment, accident, and animal contact. In general, faults happen when an abnormal physical contact occurs between lines or on lines to earth that create a short-circuit path. If the system is not well protected, the high fault current due to the short-circuit path can cause damage to the equipment in the system. Faults also affect the reliability and quality of the power supply, leading to power interruption by frequency and voltage collapse and voltage sag events. Regardless of the interruption period, the losses are often enormous both to the customers and power utility companies.

There are two types of fault, determined by the physical nature of the short-circuit path: temporary or permanent. Common causes of temporary faults on overhead lines are lightning strike resulting in a flashover of the insulator; bird or animal contact; and momentary contact due to wind or trees. Faults caused by these events exist for a very short period of time. On the other hand, a permanent fault remains in the system until the short-circuit path is removed. Common causes of permanent faults in power system networks are cable insulation failure, objects falling on the overhead lines, dirt on insulators and lines falling to the ground.

When faults occur, a protection device operates to isolate the faulty line from the rest of the system (loss of mains / grid). The generators designated to provide voltage and frequency control will respond to control the island voltage and frequency. In order to achieve smooth transition to island operation, the generators must firstly ride through the fault or failure and secondly act to balance the active and reactive power in the islanded network. With a carefully designed load shedding method the operation of the islanded distribution network might be possible. It is important however that the design of the load shedding method is designed on the understanding of the characteristics of the system involved, including system topology and dynamic characteristics of its generation and load. A poorly design method may be ineffective and eventually lead to total customer interruption.

Over the years, however, utility experience and extensive studies on a number of systems have resulted in different methods guidelines. In the following section, principles and guidelines for load shedding methods are reviewed.

manual / SCada Load Shedding

Manual or operator initiated load shedding [0] is not a reliable method to be used to avoid frequency deviation. However it can be used by some utilities to manually shed load or open ties (interconnectors) with adjacent areas at a frequencies below automatic underfrequency thresholds. This type of action might be necessary to prevent any further frequency deviation and to recover the frequency back to the nominal value. This load shedding scheme cannot be used for the islanded distribution network as it will be very slow as the frequency and voltage in the network will collapse within few seconds making it impossible for the operator to decide the correct "defence" action required for safe operation.

automatic Load Shedding using thresholds

An automatic load shedding for transmission system using different schemes such as underfrequency, undervoltage and combinations of the two can be employed to avoid frequency or voltage collapse during a significant imbalance between generation and load. These types of load shedding methods are very dependant on off line studies of the system's dynamic performance and only consider the greatest probable imbalance between generation and load. These methods have to be coordinated with the protections of the generating units, shunt capacitors and other automatic actions that occur in the system during frequency and voltage variations.

Underfrequency load Shedding

The underfrequency load shedding scheme as explained in the following papers [0] - [0] uses relays detecting the system's frequency. These are designed to operate on the instantaneous frequency value where they trip when the frequency drops below the set point of the relay. The shedding is accomplished in the systems distribution or transmission stations where major load feeders can be controlled by tripping of the circuit breakers (CB) automatically. Different settings can be applied in these load shedding schemes.

Multiple stages can be used in the scheme [0]. The substation loads can prioritised and grouped according to the importance of the load. The relays can be set to control one or more groups of loads and when there is a frequency drop these can be disconnected sequentially where the group with the highest probability being disconnected the last. Each group disconnected should contribute to the system rate of change of frequency decline. If the load to be disconnected is small compared to the overall imbalance then the contribution will be insignificant and would cause further problems to the systems frequency decline.

Another setting usual for this type of scheme is the time delay [0]. The time may can be required and used usually to avoid any frequency transient dips that could arise in the system. The time delay also avoids unnecessary load shedding by allowing the load / frequency controls in the system to respond to the frequency deviation. However load shedding performed with long time delays should be set appropriately as it will make the system more vulnerable to system stability if eventually load shedding is required. This method will work adequate in a situation where the system frequency decline is slow.

For example, as discussed in [0], in the UK as stated in the NationalGrids' GridCode each transmission area has to disconnect a defined percentage of the peak Demand that each Network Operator whose system is connected to the GB Transmission System shall disconnect by low frequency relays at a range of frequencies. The defined frequencies and the amount of loads are given in Table 1 -1.

Table 1‑1: Load Shedding Scheme employed in the UK

Frequency (Hz)

% Demand disconnection for each

Network Operator in Transmission Area































Total % Demand




The percentages in Table 1 -1 are cumulative such that, for example, should the frequency fall to 48.6 Hz in the NGET Transmission Area, 27.5% of the total Demand connected to the GB Transmission System in the NGET Transmission Area shall be disconnected by the action of low frequency relays.

A significant drawback of this method is that the system's frequency must be already be low before the relay can operate which can delay the load shedding action and the frequency recovery of the system. Additionally these types of schemes usually shed more than the required amount of load.

Undervoltage load Shedding

Undrevoltage load shedding method has been successfully deployed in transmission systems to protect them from voltage collapse [0] - [0]. System studies are required to determine which systems are potential candidates for suitable the undervoltage load shedding method. This method is most useful in slow decaying systems where the undervoltage load shedding relay time relays can coordinated accordingly and operate to alleviate the system from overload conditions and low voltages.

Voltage collapse can be studied using steady state simulations for the identified areas using a power flow analysis. System planning engineers conduct numerous studies using P-V and Q-V as well as other analytical methods to determine the amount of load required to be shed to preserve voltage stability under different disturbances.

Dynamic simulations can then determine the speed of the collapse and load shedding settings.

An example as discussed in [0] in the US in the Puget Sound area, which is prone to voltage collapse has been studied. The voltage trip thresholds were determined from the results of steady state simulations of worst contingencies. The time delays for the relays were coordinated to address control actions of the automatic capacitor switching, generator limits, on load tap changing transformer using dynamic simulations.

Table 1‑2: Load Shedding Scheme employed in the US

Voltage (pu)

Time delay (s)

% Demand disconnection for

Network Operator in Transmission Area










When the monitored bus voltages fall to 0.90 pu or lower for a minimum of 3.5 s then 5% of the load is disconnected. Additionally another 5% of load disconnection should occur when the voltage falls to 0.92 pu or lower for 5.0 s.

There limitation associated with proper application of undervoltage load shedding is the location of its where the relaying may be appropriately applied. If it is placed on a distribution line the effects of auto tap changers mask a system overload condition from the relay, or alternatively a line switching operation or the startup of a large industrial plant on one feeder could fool the relay. The relay would not be appropriate at locations directly adjacent to generation powerful enough to control bus voltages even during severe overloads. The relay is best applied to locations with fairly stiff voltages under all normal conditions, so a low voltage condition will reliably indicate a severe overload condition, as may be assumed to be the case at large substations associated with bulk power transmission lines and therefore this method cannot be effectively applied in islanded distribution networks where DG unit power and load demand varies.

combination load Shedding

In order to increase the security of the above discussed methods for underfrequency load shedding the relay could be set up to supervise the voltage, the current or the rate of change of frequency. According to their combined settings, the relay could either be blocked or initiate tripping of the CB to avoid any misoperations.

One combination load shedding scheme is to use an underfrequency load shedding relay with voltage supervision. Basically the operation procedure of load shedding is blocked from operating unless the voltage is below a given threshold. The underfrequency relay will be able to trip the CB as long as the bus voltage it is monitoring is lower than a set point.

Another combination is to use current supervision instead of the voltage. The purpose of the current supervision is to select which feeders to trip. This can achieved by monitoring which feeders are loaded above a certain point and then the relay will initiate the load shedding signal.

An alternative is to use the rate of change of frequency for supervision [0] - [0]. During a disturbance the supervision of the rate of change of frequency can block the tripping for very fast frequency changes but would allow for typical frequency decay rates. Also instead of measuring the instantaneous rate of change of frequency supervision is to use the frequency change trend. In other words by monitoring the average rate of frequency change will provide a more secure decision for tripping during disturbances. The load shedding decision of the scheme is made by monitoring the frequency change over a specified amount of time usually few hundred milli seconds. Therefore making the operation of the relay slower than the ones employing the rate of change of frequency.

automatic ADAPTIve Load Shedding

Adaptive control involves updating the amount of load to shed used by the method to cope with the fact that the conditions such as the power imbalance between generation and load of the system are time-varying or uncertain. It is important in these circumstances to minimise consumer disruption through proper design of the load shedding arrangements. An adaptive load shedding, is based on the relays reacting to a disturbance either by being instructed the amount to shed or by having certain defined criteria based on the rate of change of frequency.

Anderson and Mirteydar in [0] present an adaptive methodology for setting of underfrequency relays that is based on the initial rate of change of frequency at the relay. The frequency performance of the islanded is represented by a linear system frequency response as shown in Figure 1 -1 and presented in more detail in the literature in [0].

Figure 1‑1: Simplified frequency response with disturbance input



= inertia constant (s)


= fraction of total power generated by HP turbine


= reheat time constant (s)


= mechanical power gain factor


= droop characteristic (pu)


= damping factor

Clearly the only observed quantity that gives any clue as to the size of the disturbance is the initial slope of frequency decline.

The use of the initial slope to estimate the magnitude of the disturbance requires that every substation in the island will observe slightly different slopes and will therefore shed load based on different estimates of the disturbance. However on average the system as a whole will shed approximately the correct amount of load.

To set the parameters for the relays as explained they are based on a simulation of the frequency response for the system. In the example given (H = 3.5 s, FH = 0.3, TR = 8.0 s, Km = 0.85, R = 0.06 and D = 1) the evaluation of the frequency and its slope against different amounts of disturbances are given in Table 1 -3.

Table 1‑3: Initial Slope and Maximum Deviation vs Upset (frequency nominal 60 Hz)








































The lowest frequency permitted in the system is 57 Hz from the nominal 60 Hz. Therefore when a magnitude greater than -0.0521 pu/s is observed load shedding must be triggered.

This method relies on the fact that the amount of load shedding is a function of only the inertia constant and the observed slope. The inertia constant is the rotating kinetic energy of all units in the island divided by the total connected volt ampere rating of the units. This parameter has to be estimated. Therefore, the initial slope is the only unknown. The load shedding amount is computed in per unit, which makes it easy to apply to every load and to every load shedding relay.

A positive is that communication is not required between relays and the boundaries of the island are not required to be known. However the drawbacks are that if it is applied for the islanding application of islanded networks this might not be possible as the method needs good estimates of the inertia of the system D, R, TR, Km and FH. This can significantly change with the varying DG units and loads in the distribution network.

Another adaptive load shedding method presented by Terzija in [0] uses similarly as the previous method a variation of the typical swing equation. Due to the dynamic responses of turbines, governors, other control actions, spinning reserve, loads are not taken in account in the calculation of the required amount of load to be shed as given in .

Where H is the inertia constants and assumed to be known in advance to the disturbance. The adaptive approach is based on real time estimation of fc (frequency of equivalent inertial centre) which is proposed to be calculated centrally by measuring the local frequencies at each generator.

The proposed method assumes that the time constants in the power system are large and with modern communication this method would be possible for big power systems. However in distribution networks communication is believed not to change drastically in the near future making this application difficult to implement. This is because the estimation and control information are evaluated after the disturbance occurred.

Van Cutsem and Otomega proposed a method in [0] which relies on a set of load shedding controllers distributed over the region susceptible to voltage instability. Each controller monitors the bus voltage and act on a set of loads located at that bus. Each controller acts when its monitored voltage falls below some threshold and trips at different time according the severity of the drop. The action can be repeated until the voltage is above the threshold voltage.

The principle of operation of the controller is described as follows.

The delay Ï„ depends on the time evolution of V as follows.

A block of load is shed at a time t0 + Ï„ such that:

where C is a constant to be adjusted. This control law yields an inverse-time characteristic: the deeper the voltage drops, the less time it takes to reach the value C and, hence, the faster the shedding. The larger C, the more time it takes for the integral to reach this value and hence, the slower the action.

Furthermore, the delay is lower bounded:

to prevent the controller from reacting on a nearby fault. Indeed, in normal situations time must be left for the protections to clear the fault and the voltage to recover to normal values.

Similarly, the amount ΔPsh of power shed at time t0 + τ depends on the time evolution of V through

where K is another constant to be adjusted, and ΔVav is the average voltage drop over the [t0, t0 + τ]interval, i.e.,

Moreover, the whole system will tend to shed first where voltages drop the most. This location changes with the disturbance. Hence, the proposed scheme automatically adjusts the shedding location to the disturbance it faces. Note that the above features are achieved without resorting to a dedicated communication network. The controllers do not exchange information, but are rather informed of their respective actions through the power system itself.

The drawback for this method for distribution network is that the tuning which consists of choosing the best values for Vth, C and K. A C and K combination suitable can be identified by minimising the total load shedding over all disturbance scenarios. Clearly this method would shed more loads for some scenarios. An additional concern is that the dynamic performance of the DG units and loads is not taken in account when performing load shedding if applied to the islanded distribution network and by trying to shed in steps the frequency drop in the network might drop significantly.

automatic Intelligent Load Shedding

Applications of intelligent load shedding in power system engineering (e.g. genetic algorithms, artificial neural networks, MonteCarlo etc.) have been demonstrated in [0] - [0]. The characteristics which are inherent to intelligent methods, such as the ability to learn and generalization make it feasible for applications such as load shedding.

You et al. in [0] discuss of a method that uses the rate of change of frequency to load shed. The method uses the same approach to calculate the required amount of load as in [0] and at the same time, the conventional load shedding method with undefrequency thresholds is incorporated to form a new two level load shedding method.

The conventional load shedding method has longer time delays and lower frequency thresholds which can be used to prevent unnecessary load shedding in response to small disturbances. If the disturbance is large, the second layer will be activated and a block signal to the first layer is enabled. The second layer based on the rate of change of frequency load shedding will shed more load quickly at the early stage of the disturbance.

Similarly as to paper [0], this method will have the same limitations when applied to the islanded distribution network. In the paper [0] which follows this study, the explanation of the selection of the settings for the relays is discussed. Agent technology is to try to assure that the method will withstand all possible disturbances. Traditionally after a major disturbance, the system is revisited and settings of devices and control actions are changed so that the system will withstand the same disturbance in the future. This however due to analysis of the system significant time and cost will be required.

For the autonomous and adaptive learning capability for the agents, the reinforcement learning technique is used. Reinforcement learning is learning by interaction.

The agent tries actions on its environment and then, the tendencies of taking particular actions are reinforced by receiving scalar evaluations of its actions. Thus determining the amount of load to be shed required to avoid collapse.

The paper does not discuss whether the technique is applied online or offline through simulation. Clearly for the online this would not be ideal as it will take a lot of number of failures until the agents are properly set for that particular disturbance. For the offline simulation a concern is that for islanded distribution networks the topology, DG unit power and load demand will change thus making the decision of the action of the agents is difficult to train. Another concern is communication between agents. Fast communication would be required for coordinated decisions.

Another approach to load shedding is the use of fuzzy expert system and is described in [0]. In this paper Sallam and Khafaga described a method to control the voltage instability by load shedding using fuzzy technique as fuzzy controller.

The operation of the method relies on the experts' knowledge which is expressed by language containing ambiguous or fuzzy description. The aim of this study is to design and analyse a fuzzy controller for the study to control against load and voltage instability by calculating the optimum load shedding as output.

Similarly in [0] the authors propose genetic algorithms for the optimum selection of load shedding. These techniques search and optimise the amount of load shedding using objectives and constrains required for a practical load shedding method.

Also in [0] the authors introduce another technique using the artificial neural networks is presented. To prepare the training data set for the artificial neural network, transient stability analysis of the power system is required and to find the minimum load shedding for various scenarios. By selecting the total power generation, total load demand and frequency decay rate as the input neurons for the method, the minimum of load shedding is determined to maintain the stability of the power system.

In paper [0] Thalassinakis and Dialynas introduce a computational method using MonteCarlo simulation approach for the calculation of the settings of the underfrequency load shedding relays is discussed. The frequency performance as previously discussed in section 1.5 is used here as well. The strategy for the relay settings will be determined against amount of load to shed, time delay, rate of change of frequency and underfrequency level. A new strategy is developed by changing these settings. The MonteCarlo then computes the system through reliability indices of generating units, the system frequency and load shedding indices.

load shedding based on static optimisation

The first theory of applying load shedding using an on line dynamic simulation of the power system network was introduced by La Scala et al. [0]. Followed by an improvement of the method combing a control action to ensure angle and voltage stability enhancement in [0]. The first paper that introduced the same concept applied for large power systems to the smaller distribution network is described in [0] by Nelson and Aponte. A more recent study using similar technique is also presented in [0].

The paper presented in [0] describes the philosophy and the implementation of a preventive load shedding control algorithm for the application in dynamic security assessment. The methodology is based on nonlinear programming techniques, for assessing control actions to guarantee the dynamic security of power systems. The basic idea is that the online dynamic preventive control can be seen as a static optimisation problem with minimising function and equality and inequality constrains. The equality constrains consist in the discretisation at each time step of the differential algebraic set of equations representing the power system. The inequality constrains define a domain where the system trajectories should be contained in order to satisfy the requirements for the system performance stability and steady state voltage dips.

In [0] the formulation includes corrective actions based on load shedding. The proposed method assumes that the analysis is performed to detect particular disturbances threatening the dynamic security of the system. The analysis is based suing the n-1 rule which is performed in advanced and applying the results immediately after the detected contingency. Each analysis has its associated strategies consisting with the corresponding amount of load to be shed at a fixed number of controlled nodes. The optimisation however is evaluated based on the steady state values of angle, voltage and active power (generator and load). Load shedding based on static optimisation performs load flow to calculate the initial P, V for all the nodes in the system. Then the method performs a transient simulation assessment to ensure the system is stable against angle and voltage. Followed by an approach to the minimisation of a function in presence of equality and inequality constrains consist in incorporating the inequalities in the cost function by adopting the penalty factor method and treating the whole problem as a minimisation in presence of the sole equality constrains by the use of Lagrange multipliers.

This method has been used for synchronous generators in transmission systems. However in distribution networks because of the diversity of the generators and their ride through capability this approach could result in conditions where optimised solutions do not meet the requirements as shown in Figure 1 -2 [0].

Figure 1‑2: Ride through capability of Generating Unit, DC Converter or Power Park Modules.

Explanation of graph required.

Each Generating Unit, DC Converter or Power Park Module shall remain transiently stable and connected to the system without tripping. However for small generating units connected in the distribution network their transient behaviour could be as shown in Figure 1 -2 'b' and 'c' where local protection and circuit breaker operation of generators or sensitive equipment will be disconnected after such a response. Similar to the voltage is for the frequency range.

Therefore the load flow with corrective control for angle and voltage stability approach for the load shedding optimisation is not appropriate for distribution networks.

In [0] and [0] describe of a method implemented in distribution networks where not only the amount of load shedding is optimised but also the time for the disconnection. The current trend is to apply the corrective measures as soon as possible or delayed for the sake of event discrimination. The study and results however show that when the corrective action is applied at the optimal time increased damping and enhanced response are observed.


The use of load shedding as a tool to keep the network stable has been constantly evolving, and different approaches have been formulated. Relaying schemes like underfrequency and ROCOF [0] - [0] are some examples of the mechanisms implemented to trigger a load shedding event. Typical load shedding schemes based on predefined threshold set points is quick, simple and reliable measure against system disturbance. When the frequency of the system reaches a specified threshold value, a time delay is inserted prior to the shedding action in order to avoid overshedding and assist the coordination of the next stage of load shedding action. This technique however when adopted for the islanded operation of small distribution networks would have several disadvantages. Too few frequency levels could lead to overshedding, but on the other hand, time delays between stages could add up and may not allow for enough load to be shed in time to re-establish nominal frequency.

The implementation of ROCOF techniques mitigates some of these problems. The ROCOF value calculation is an immediate indicator of the power imbalance; but for the distribution network the variation of the DG units operation would make this measurement unreliable. Also the average ROCOF calculation may take too long and eventually make the load shedding method slow in operation. Even if accurate measure of the islanded distribution network ROCOF value, the relays will be difficult to set properly, unless imbalance can be predicted. If these parameters can be predicted, the ROCOF relays may improve the load shedding method.

Newer adaptive techniques have come to light to fulfil more complicated and challenging modern demands of power systems. These techniques are also constantly changing to accommodate newer technologies and operating philosophies. In the case of adaptive load shedding methods to avoid system frequency and voltage collapse is favourable to the load shedding with predefined set point. However these methods would require distribution network communication infrastructure to improve significantly in order for these methods to be easily implemented.

The optimal load shedding problem formulation has also been evolving. In this approach, the purpose was to find the minimum amount of load to shed while satisfying load flow equations and static constrains like line flows, voltage, angular limits and shedding constrains. The approach also takes into consideration the machine governor, the frequency and voltage characteristics of loads. The desired outcome is to obtain the dynamic response of each load shedding operation and to detect the best load shedding option to minimise an objective function based on system quantities. The steady state of the trajectories are squared to estimate the "cost" of an iteration and to obtain the next load shedding values that should further reduce the goal function. This is done until convergence is achieved.

In addition intelligent load shedding method that use power system engineer knowledge have been introduced in power systems for the load shedding action. An identified problem for the above methods when the islanded distribution network is changing with unknown scenarios is a concern. Additional training is required for correct decisions of the relays to maintain stability.

A problem identified for distribution systems that many proposed methods don't discuss is the communication infrastructure.

Figure 1 -3 illustrates the approximate time frame associated with different protection and control actions for the power system network and more specifically for an active distribution network. It also shows the relationship between the duration of the phenomena and the time frame of the protection and control actions most likely used to limit their consequences. The time scale is in logarithmic, and the underline on the action shows typical operating range. Additionally another important factor to be compared is the costs, especially for the communication requirements for each action.

Also indicated in the figure is the important area identified of operation for the proposed load shedding method. The current generation of SCADA systems operates in the range of 2 seconds to 2 minutes. Since fast communication systems are expensive and scarcely available in distribution networks, the proposed method uses dynamic state information to allow each load shedding relay to perform in decentralized manner. Details of the proposed method are described in the following section.

Figure 1‑3: Communication Economics for Load Shedding in Distribution Network

For these reasons when designing a new load shedding method for a distribution network have to be taken into account as excessive load shedding will result in overfrequency conditions, or unnecessary loss of service continuity and revenue and too little shedding in failing to arrest the frequency decline, which may in turn lead to further loss of generation on underfrequency or even system voltage and frequency collapse. Also the use of existing communication system should be used to ensure the distribution network stability when operating in islanding.

As a result a new load shedding method with different architecture and concept is required to allow islanded distribution network operation.