Balancing the Load . p d f

Abstract

The way electricity is produced in the UK is currently changing, with renewables being increasingly utilised and the concept of community energy schemes growing in popularity and feasibility. It therefore makes sense that the way electricity demand and supply are balanced should be reviewed. The large scale and reactive nature of current load balancing techniques are detrimental to the efficient of the system, which raises the question of whether it is possible to use smaller, community based schemes to reduce the load imbalances at source.

This paper intends to establish if there is any feasibility to using a de-centralised system that increases the load factor of domestic electrical demand at a local level, and therefore reducing the quantity of load balancing required. The proposed system will attempt to improve the load factor and reduce the peak loading of groups of up to 100 houses by synchronising the electrical load signatures of suitable domestic appliances. The opportunity to synchronise will be created by allowing the use of a short delay on appliance operation, up to 15% of the appliance cycle length. For this research, the appliances deemed suitable were washing machines and dishwashers.

In this paper, a model was developed to test the proposed system in a variety of scenarios in order to gain an understanding of the effectiveness of the concept. Factors tested included the number of houses and maximum length of delay period. The effects of profile variation was also investigated.

Results of these tests showed that the dishwasher was the most effective appliance and that peak load reduction of 20% could be achieved on appliance loads when maximising the delay period to 15%. The reliability of positive results improved with the maximisation of the delay period and by increasing the number of houses in the system to 100.

Overall, the paper concluded that the concept of appliance load profile synchronisation, for the purpose peak electrical load reduction, is valid and warrants further research.

Acknowledgements

Firstly, I would like to thank Dr Nick McCullen, my academic supervisor, for his guidance and advice throughout the year. I would also like to thank Jack Kelly, a researcher into energy meter disaggregation, for the work he undertook and data he collected, which has enabled me to undertake my research. Finally, thanks go to my parents, my friends and my girlfriend for their support throughout my time working on this dissertation..

Introduction

Electricity is fundamental to the development and growth of most countries, not least in the UK. It is used in almost every aspect of society and it is therefore essential that the electric supply is constant and reliable. Electrical load balancing is a vital aspect of maintaining a reliable electricity supply with the effects of mishandling load balancing resulting in power cuts, and the financial and social costs associated with this[1].

Currently, electrical load balancing is predominately achieved by matching electricity generation with electrical demand at all times. In practise, this means predicting the upcoming electrical demand of the consumers and ensuring the correct amount of electricity generation is on supply at the right time. This is an ongoing process that all large scale electrical networks must undertake to ensure reliability of supply. In the UK, the balancing of the entire national electricity network is a service undertaken by National Grid[2].

There are several issues with this centralised, top down approach to electrical load balancing, especially as renewable sources become an increasingly large proportion of the energy supply. The passive approach of only controlling the generation of electricity causes the need for significant back-up generation capability and spinning reserve, power stations generating but not connected, to react to changes in electricity demand. Secondly the centralised nature of the system means that the imbalances are measured and corrected at national scale where they have been compounded to significant levels[3].

More recently, National Grid have started to implementing methods of controlling the demand of electricity as well as controlling the generation. Current schemes used by the National Grid are discussed in the literature review section of this paper, however again they are notably top down approaches. This control of electricity demand is referred to as demand side management (DSM).

This raises the question as to whether a de-centralised, bottom-up load balancing system could be employed to fully or partially balance energy demand and generation in order to reduce the amount of reserve generation that is on standby at all times. Either demand side management or generation side principles could be employed at local levels. There is currently a lot of research into the use of micro-renewables and energy storage systems to balance load however there have been less research into dynamic demand side management systems to actively balance load. This paper will therefore focus on investigating a de-centralised load balancing system that uses demand side management.

The use of DSM systems can also have other benefits including social, financial and environmental benefits, as discussed by Ofgem in their review of non-traditional business models in the energy sector[4]. One such benefit of some DSM systems is the reduction in peak loading which can help increase the spare capacity in an electrical network[5]. This is discussed further in section 2.4 of the literature review.

This study investigates the potential for a small scale peak load reduction system that would help alleviate load balancing requirements by synchronising the use of domestic appliances. As an initial study, this paper aims to provide a proof of concept as the basis for further research.

Literature Review

Introduction

This literature review will focus on building an understand of the methods and schemes currently used to balance electrical supply and demand in the UK, and what research has been undertaken on DSM to date.

Current UK load balancing

Up to 53GW of electricity is consumed across the UK at any one time and the responsibility of balancing the generation of power with the level of consumption on a day to day basis is undertaken by National Grid[6]. National Grid constantly monitors the electrical supply and uses a number of different services/controls to ensure the network remains balanced[7].

Load balancing is undertaken using several different methods, at several different time scales. It is important to recognise that load balancing is not just a near immediate demand response issue. Load balancing is a long, medium and short term problem tha t requires constant action at each timescale. Predicting the UK’s energy demand in decades to come is essential in understanding what new power sources need to be planned, designed and constructed. This will increasingly become a larger problem as more renewable energy source are implemented into the energy network and supplies become more unreliable and less flexible[8].

In the long term National Grid use historic data and event analysis, to forecast electrical demand for coming days, weeks and months. This allows them to schedule additional power sources if and when required[2]. In the short term, a number of different methods are utilised and a summary of these is shown in Table 1.

Table 1: A summary of short term load balancing schemes operated by the National Grid[9]

Description of balancing method Technique

Load Balancing Response

Description of

10-30 This reacts directly to Power stations operating off grid are

response

seconds

the electrical

connected to increase generation. Some frequency of the grid. large electricity consumers, such as aluminium smelters[5], are also turned off as a response to the frequency.

Fast reserve

2 minutes

This activates on

Large electrical consumers who are

receipt of an

contracted to the scheme, typically

electronic instruction

energy intensive industrial processes,

from National Grid

are paid to shut down.

Short Term

20 minutes

This activates on

Backup generators are started and large

Operating

receipt of an

electrical consumers, who are again

Reserve (STOR)

electronic instruction

contracted, shutdown energy intensive

from National Grid

processes.

Figure 1 shows a timeline with the most dominated forms of load balancing actions undertaken in the UK. It provides an overview of the timescales involves and the highlights that the National Grid are only a residual balancer operating in the day-to-day timescale.

Year(s) Month

Minutes or Ahead

Day

Within

Hours

Ahead

Ahead

Day

Ahead

less

Frequency construction of

Design and

Maintenance

Scheduling of

Short Term

Response power stations

Return to service

Scheduling of

Fast

decisions

power stations

Reserve

Figure 1: Timeline of the load balancing actions used in the UK. Shown against typical time required for planning and impletmentation[9]

Demand Side Balancing Services

The term ‘demand side’ refers to services that operate by altering or affecting the demand of the electrical supply. This is opposed to ‘load side’ services that are essentially affecting the quantity of electricity generated[8].

The one limitation with current load balancing systems and services in the UK is described by the review paper named “Demand side management: Demand response, intelligent energy systems, and smart loads”, where it states “The classical modus operandi of electric energy systems is unidirectional and top-down oriented.”[10]

This suggests that there may be scope to tackle electrical load imbalances at source rather than at a national scale where the effects have been compounded.

One of d emand side management’s main advantages is that it is often less expensive to control demand to an extent rather than build new power stations or energy storage devices in the pursuit of a balanced energy network[10].

In the paper, Palensky and Dietrich also note a number of problems associated with the use of some demand side management (DSM) techniques, such as the risk of damaging the quality of processes involved in the system. For example the curtailing of particular building comfort systems, heating, ventilation or lighting, simply for the benefit of reducing load for a short period.

An addition knock on effect is identified in the form of overcompensation after the peak reduction period has been completed, to the extent that a greater peak is created afterwards. Since most appliances and building service processes are programmed to operate in the most efficient way, the effect of delaying a device process for a period may prevent optimum operation and cause the device to work more intensely, after the period has finished, to return to its design parameters. Not only could this cause greater peaking but also increase the overall energy consumption of the process as it is pushed from its optimised cycle.

The paper “The concept of demand-side management for electric utilities” categorises the various ways in which DSM can operate and are shown graphically in Figure 2 in the form of load profile alterations[10][11].

Figure 2: Demand side management electrical load balancing categories shown in the form of load profile graphs with electrical consumption shown against time[11].

 Peak clipping – Reducing loading at peak periods. This effect can be generated by turning off non essential devices and processes at peak loading times.  Valley filling – Opposite of peak clipping, valley filling means increasing load at energy demand low

points during the day. This can be achieved by turning appliances on during the relevant periods.  Load shifting – This method redistributes electricity demand, specifically moving energy load at peak times to demand periods of low demand. It therefore has the combined effect of peak clipping and valley

filling.  Strategic conservation – This essentially refers to programs that not only help to reduce peak demand but also overall demand. It has the effect of smoothing out peaks. It is manifested in improved efficiency

of industrial processes or electrical devices.  Strategic load growth – This refers to the changing of load profiles in order to assist in Valley filling. A

major example would be that of switching from ‘other fuel’ based heating to electrical heating. This would potentially increase electrical loading during low electrical demand periods, thus smoothly out load profiles.

 Flexible load shape – This refer to load profiles which consist of load generated from flexible sources, where there is no detrimental effect caused by the device operating in any manner.

Domestic DSM systems

This section will focus on existing research of DSM systems applied to the domestic sector, where load imbalances can be corrected at source. If peaks generated by everyday domestic appliances, such as kettles and ovens, could

be “offset” or coordinated at the generation source then the peaks created within single dwellings may be significantly reduced. For example if two appliances with heating elements synchronised their modulated heating cycles so that as one was operating the other was in its ‘off’ section of its cycle then the combined consumption at any time would not exceed the consumption of just a single appliance. This peak reduction would then be passed on to the electrical network and prevent that particular peak event from being imposed on central UK load balancing services. Several research papers have investigated different approaches to this.

Appliance re-programming

A research paper by Newborough and Augood named “Demand-side management opportunities for the UK domestic sector” investigates the possibilities involved with reprogramming domestic appliances such as ovens and dishwashers to have more “load conscious” cycles/programs[3]. It identifies that often the cause of electrical load peaks is the use of heating elements to heat water or food rapidly. It analyses the peak consumption of many of the most common household appliances and determines where cycling could be extended over a marginally longer period in order to reduce the peak load.

For example, appliances such as kettles are designed to heat water as rapidly as possible for the user. This results in sharp peaks that may only last a few minutes. This is also applies to the heating elements in ovens which create sharp peaks on a cycle as they heat up the air within the oven. The fact that ovens operate by cycling their heating elements on and off rather than by maintaining the elements consistently on means that there is potential for offsetting the cycle peaks using other appliances that operate on a cycle by alternating the “on” periods of each appliances.

Newborough and Augood note that in general the load generated by the heating requirements within an appliance are ten times higher than the appliances other uses. Significantly, this may suggest that by operating the heating cycle of two appliances alternately, the peak load could be almost halved.

It states that the loads generated from non-heating appliances, such as TV or radios are often relatively low and therefore any reprogramming would not result in worthwhile peak load reductions. Lighting is also difficult to modulate due to its essential nature. It therefore concentrates on appliances with heating elements since that’s where the largest improvements could be made[3].

Critically, although the philosophy behind the research is sound, the major issue with the scheme is that appliances have a set operation programs and altering these post production is not a viable option. However, as noted, it would be useful for manufacturers to adopt some of the load conscious programs outlined in the paper.

The paper also identifies quite simply that one of the easiest methods of reducing peak loading would be to target ovens and hobs, as the appliance causing the largest peak loads, and switch to a gas fuelled version.

It summarises by also suggesting that more load conscious tariff systems could be adopted by energy suppliers whereby customers are penalised for using large amount of electricity during peak periods. This would be similar to the current ‘Economy 7’ tariff, however with shorter and more numerous peak periods.

Load factors

From a demand side management perspective the measurement for success in improve load balancing is to be producing a higher load factor for the profile in question. A load factor is essentially the ratio between the mean electrical consumption and the peak electrical load consumption. For any system the ideal load factor would be 100% (1.0), indicating that peak consumption was equal to the average consumption. It was found that domestic load factors ranged from 6-43%, typically lying between 8-15%[3]. This demonstrates quite clearly that each domestic property individually performs very poorly and that there is significant room for improvement. It must, however, be noted that the load factor of an individual property does not represent the profile of multiple properties, as the varying nature of resident’s habits (for example, eating times) means that simultaneous load generation is unlikely, therefore creating smoother profile is created.

Dynamic Demand Response

The paper “Potential for domestic dynamic demand-side management in the UK” by Infield and Short focuses on current research in dynamic demand response control whereby domestic appliances are used to respond directly to the frequency of the network[12].

The frequency of the UK electrical supply is designed to operate at 50hz. By the nature of the constantly changing electrical demand and supply across the entire country this frequency fluctuates. When demand increases faster than the electricity supplied, the frequency will drop and, vice versa, when supply is greater than demand then the frequency will increase. The consequences of the frequency increasing or decreasing excessively are severe, with the possibility of generators failing abruptly. It is clearly therefore critical that the frequency is monitored and supply or demand adjusted to ensure equilibrium. Frequency is also therefore the perfect property to monitor to ensure a reliable energy supply.

Currently the frequency is monitored by National grid whose frequency response service ensures the frequency is kept with legal operational limits (49.2-50.2hz)[12].

In the paper, Infield and Short, analyse the benefits of a system that employs fridges and freezers in the UK to provide electrical frequency control in the first instance. It exploits the thermal storage properties of fridges and freezers to allow them to operate their compressors based on the frequency of the energy network at any point in time rather than based on a fixed cycle. In other words, due to fridges and freezers being thermally insulated, the compressor inside can be turned off for a length of time without a detrimental effect on its contents. By turning each fridge into an independent frequency regulator the research shows that savings can be made on the current balancing system used for immediate response.

The paper uses a national scale model with 40 million dynamic demand units, to estimate the effectiveness of the system and concludes that it would successful assist the frequency response of the UK electric supply[12]. It continues by suggesting that the system would also be financially viable, although it should be noted that the economic assessment is very brief with little detail.

The major disadvantage to this type of system is the consumption spike that lags behind after a response event has initiated. Infield and Short acknowledge this limitation and it is a well reported problem with many frequency response DSM techniques[13][9].

Additional benefits of DSM

Reducing the imbalances of the UK electrical network provides financial benefits to the National Grid, through

a reduction in the scale of services they would need to provide. A proportion of these savings, could then be passed on to consumers. Currently, however, only 1% of consumers ’ energy bills are related to the cost of load balancing services[2]. This is a relative small sum and uptake of a DSM system may prove slow, or even no existent, if financial benefit for the National Grid was the only motivation.

There are, however, a large number of other benefits to the use of DSM for distribution network operators (DNOs). DNOs are companies that build, own and maintain the UKs distribution level electrical networks. Their role, amongst other things, is to ensure that electrical networks have the capacity to supply the properties and infrastructure connected to their network. This means that as new developments and infrastructure are constructed they must recalculate the electrical demand that will be generated and determine if reinforcement is required to the existing network. This is potentially where the use of DSM system could reduce or eliminate the need of reinforcement. This principle was reiterated by Western Power Distribution during a talk at a Community Energy conference in January 2015[13].

During the design of electrical networks it is the maximum peak load that defines the sizes of network apparatus. The maximum peak load is calculated by summing the maximum load of all properties to be connected to the network and then adjusting using a coincidence factor. The coincidence factor is a reducing factor that accounts for the extremely low probability of all properties requiring their maximum possible load simultaneously[14]. The use of DSM systems that reduce peak loading could effectively reduce the coincidence factor, and increase the load factor, of the properties within the system network. This is similar to the effect of load factors increasing as more houses are included in the consumption profile. Figure 3 shows the relationship between the coincidence factors used in electricity network design and the number of customers connected to the downstream network. It demonstrates that as the end usage point is neared the network coincidence factor increases. This suggests that the zone benefitting the most from reduce peak coincidence factor, and subsequently DSM systems, would be below 100 properties.

Figure 3: Showing the relationship between coincidence factor and number of properties connected to the downstream network. [5]

By reducing the coincidence factor, spare capacity could be created within the network boundaries. This could have large benefits where new housing or commercial developments wish to be connected to the electricity network, by potentially preventing the need to reinforce the existing network infrastructure.

This is a concept that G.Strbac reiterates in “Demand side management: Benefits and challenges” by saying that the use of DSM systems to reduce the pressure on electrical network nearing capacity has not been widely considered, in spite of the high costs of network reinforcement that it could prevent[5]. Adding further weight to the concept are the Electricity Networks Strategy Group (ENSG) who again support the investigation into new systems and technologies to assist in network capacity reinforcement and recognise that there are increasingly more opportunities for emerging solutions[16].

In addition, as the use of electric powered vehicles become more prominent there are concerns regarding the capacity of electrical network infrastructure[15]. The use of DSM systems may have an opportunity to form part of the solution to this concern.

There are however issues with the use of DSM in network capacity roles and Strbac notes that some DSM systems can have an actively negative effect on coincidence factors of electricity networks[5]. In systems that control the operation of appliances to correct load imbalances, this is because after many appliances have been turned off to reduce demand, they often simultaneously turn back on, to fulfil their primary function, and thus create another large peak in demand.

Literature Review Conclusion

The literature review has shown that although there is significant research being undertaken on domestic load balancing methods, little research has been carried out on de-centralised systems of appliances that interact with one another with the purpose of balancing load locally. Focuses have been on either centralised control of DSM or on the interaction and operation of individual devices, such as frequency response. Therefore it would seem reasonable to investigate a peak electrical load reduction system that allows appliances to interact and coordinate their operation and cycling.

From the research reviewed it can be interpreted that significant peak load reduction need not require complete appliance rescheduling but in fact just a synchronisation of appliance load signatures. The modularity of energy consumption that is typical of many domestic appliances is a characteristic of load profiles that has not been utilised by DSM techniques to date.

Acting in a de-centralised and localised manner also means that there is potential to reduce the design coincidence factor of electricity networks, thus adding an additional, and perhaps more financially robust, benefit.

Research

Study Objectives and Proposed System

This research will analyse the effectiveness of a demand side management system that uses appliance synchronisation to reduce peak electrical load a domestic context. The effectiveness will be assessed based on the peak load reduction and load factor increase achieved by the system in several simulated scenarios. The results will be compared with simulated scenarios where no peak reduction system is employed.

The system will utilised a short delay, in the form of a percentage of the appliance cycle length, to provide a range of start times for the appliance. Within this range the most optimum start time will be calculated and the appropriate delay put onto the appliance being synchronised. By undertaking this procedure on each appliance within the system the peak load and load factor will, in theory, decrease and increase respectively.

The system involves many variables that will be discussed and analysed throughout the methodology and discussion sections.

Data Selection Load Profile Data

Accurate and high resolution data at both dwelling and appliance level was required to carry out the research. The data needed to be of high resolution to allow the proposed system to synchronise load profiles as accurately as possible and over potentially short periods of time.

Data from an open source was chosen to be used as it was, at the time of writing, the highest resolution open source data available of the UK domestic sector. The data was complied by Jack Kelly and William Knottenbelt and contained readings of every household appliance at a 6-second resolution over an approximately 1-2 year period. Data from two houses was selected for use and a summary of the study meta-data can be found in Table

2 below.

Dwelling information House 1

House 2

Type of dwelling

End of terrace

End of terrace

Year of construction

Number of occupants 4

Occupant description 2 adults, 2 young children (one born

2 adults, 1 working out all day and the

during metering), 1 dog

other sometimes at home.

First Measurement

2012-11-09

2013-02-17

Last Measurement

2015-01-05

2013-10-10

Total Duration

Mean dropout rate

Number of

52 appliances and the overall

18 appliances and the overall

appliances metered

dwelling consumption

dwelling consumption

Table 2: Summary of open source meta data [17]

It is essential to understand the demographical characteristics of the profile data, as shown in Table 2, so that the probability data could be matched from similar properties to ensure accuracy of the modelling results. Due to the limited data available, it was not possible to match every characteristic. Number of occupants and type of occupant were considered the predominately important parameters, because appliances are typically driven directly by human behaviour, and therefore the probability data obtained matched these characteristics.

The data itself was structured in CSV files containing two columns, the first representing the time of each reading, and the second representing the electrical consumption at that instance. This was shown in the unit watts[17].

Probability Data

In addition, further data was obtained from a study undertaken by the University of Strathclyde that showed the probability of appliances being used at particular times of the day, as well as the average number of times an appliance was used per day. The data had been collected from a housing estate that contained a mixed demographic. Although the building type was not consistent with the load profile data it was deemed acceptable for use in the model because the demographic match was good, with the data from the 2 adult with 2 children section being used. This means that the accuracy of interactions and potential for synchronisation is as true to reality as possible with the data available[18].

Validation of data

The modelling script requires that all meter reading data be consistently at 6-second resolution. This means that if readings are missing from the appliance data then the error is carried onto each subsequent reading. For example, a 12 second gap between readings would cause each subsequent data point to represent a reading 6 seconds earlier than occurred in reality. Therefore, each piece of meter reading data is checked and validated before being included into the synchronisation model.

A validation script was written using MathWorks MATLAB[19] to perform this operation. When gaps in the data were discovered, the validation script inserted additional readings to maintain the 6-second resolution. The new values added were of equal value to the reading prior to the data gap. Since there is no way to know what the readings should have been as this was deemed the most suitable approach to repairing the data gaps.

Investigation into suitable appliances

In order for the proposed synchronisation system to work effectively it is first essential to understand the types of domestic appliances and the properties of their typical load signatures. This will help gain an initial understanding into which domestic appliances are more or less suitable for use in the scheme.

A number of appliance properties and factors were considered when determining the suitable appliances. These were as follows:

1. Level of peak loading reduction potential

2. User interaction level

3. Length of operating cycle

4. Likelihood and scale of load profile variation

5. Percentage ownership of appliance in UK

A shortlist of likely suitable appliances was made in order to begin the assessment. The appliances deemed to have merit in investigating were as follows:

Washing Machine, Dishwasher, Tumble dryer, Electric Oven, Microwave, Refrigerator, Kettle and Toaster.

All audio-visual appliances were disregarded due to their naturally high user interaction level, owing to their primary purpose of entertainment. Most audio-visual equipment also has relatively low peak electrical consumption in comparison to kitchen appliances. For example a typical laptop charging has a peak power consumption of 85W [20], compared to a dishwasher using 2000W at peak, meaning that its potential for peak load reduction is minimal regardless of the associate disruption that would be caused to consumers. This is supported by findings in the literature review [3].

Analysis

Peak loading reduction potential of any appliance is simply the load factor of the individual appliance under its typical operating cycle. It essentially describes the ‘peakiness’ of the profile and judges if there is significant Peak loading reduction potential of any appliance is simply the load factor of the individual appliance under its typical operating cycle. It essentially describes the ‘peakiness’ of the profile and judges if there is significant

For appliances of high peak load reduction potential the load factor should be as low as possible, as this represents the largest difference between peak and average loading and therefore provided the most potential for improvement to be made by aligning the load profile appropriately.

Figure 4 shows three example load signatures of each appliance being investigated. These provided the basis for the results shown in Table 3.

Figure 4: Example electrical load signatures for shortlisted appliances. Three examples are shown to demonstrate variation

User interaction level is the level of direct and focussed attention the user has with the appliance. For example

a TV is used with near constant user focus whereas a fridge has almost zero user interaction. Under this parameter, suitable devices would need to operate with very little or no direct user focus, to avoid user disruption or inconvenience.

Length of operating cycle is simply the length of time that the appliance takes to complete its operation. For example a kettle boils for around 3-5 minutes dependant on the quantity of water it holds. The length of any appliance operating cycle does not directly affect its suitability for the proposed system since peak consumption is not dependant on cycle length, however short operating cycles potentially reduce the delay length opportunity as users would reasonably expect these to complete their cycle faster.

Likelihood and scale of profile variation is the chance of an appliances typical load profile variation due to external conditions. These conditions could include temperature, illumination or user input. For example a washing machine set to cycle at 30°C will have a marginally shorter heating period if the mains water input is 6°C rather than 4°C.

UK ownership percentage of each appliance is also an important factor to consider since regardless of how appropriate an appliance may be to operate with a peak load reduction system if the appliance is not widely owned then it inclusion in the system may prove negligible.

These five factors form the basis for assessing if it is suitable to include an appliance in the proposed synchronisation based peak load reduction system.

Results

Table 3 summarises the results of the appliance suitability study. Some results have been obtained directly from the load profile data used for this study and some from two other sources, a survey undertaken by BRE on behalf of the Department of Energy and Climate Change in 2011[21] and from ‘Statistical Review of UK Residential Sector Electrical Loads’ by Tsagarakis, Collin and Kiprakis [22].

Table 3: Summary of key domestic appliance characteristics

Appliance name Ownership

Likelihood and Overall in UK (%)

Level of Peak

User

Length of

Scale of profile Suitability

variation (H/M/L)

(H/M/L) Washing Machine

(Load factor)

(H/M/L)

Dish washer

41 H L

95-105

Tumble Dryer

67 No data available

User

controlled

Electric Oven

69 No data available

H User

From Table 3 it can be seen that the most suitable appliances are the washing machine and dishwasher, with the refrigerator, kettle and toaster also offering potential. There are other appliances that may be suitable however since they are not as common in the UK domestic sector they have not been considered in this study.

Based on the findings the washing machine and dishwasher appliances are the most suitable and will be used in the proposed system model. This is because they show high levels of potential for peak load reduction with low levels of profile variation. They also have low user interaction levels and relatively long operating cycle lengths. Despite the comparatively low ownership of the dishwasher, it has been deemed suitable due to its other well performing factors.

Methodology

In order to investigate if the proposed system could effectively reduce peak electrical loading a model was produced that could simulate the effect of the system in a number of different scenarios, with the two selected appliances. The results could then be compared the results to the same scenarios without the use of the system. The modelling of such a system was undertaken using the MathWorks MATLAB[19] software package which enabled flexibility of programming for such a bespoke model.

This study has been broken down into two parts in order to simplify the modelling process. The first stage will focus on establishing the core synchronisation script that will calculate the optimum delay time of the appliance. This will then be analysed and tested using the load profiles of a single appliance and a single dwelling.

The second stage of modelling will aim to simulate and test the system applied to multiple appliances within a network of varying size. In this stage several different tests will be undertaken to assess the successfulness of the system.

Measuring the effectiveness of the system

Two performance indicators will be used to measure the effectiveness of the system, the percentage peak load reduction and the percentage load factor increase across a 24 hour day. The percentages showing the ratio between results with and without the synchronisation script.

It was decided that both indicators should be used as they each represent the successfulness of the system concerning each of the two major benefits. The percentage peak load reduction indicates the suitability with relation to increasing spare capacity in a network, and the percentage load factor increase indicates the benefit to load balancing.

Stage 1 – Single house and single appliance Model Coding

The first step in building a working model was to import the load profile data for a single dwelling and for the appliance to be analysed. Hourly probability data of the selected appliance was then also imported from the University of Strathclyde study.

Next, a weighted random start time with the data’s 24 hour period was generated. This represented the time at which the appliance was turned on and was based on the probability data. By using a weighted start time the appliance load signature would be more likely to be inserted over periods of the day that the appliance would be used in reality, and therefore interact with the patterns of electricity consumption used at that time of the day. The start time was allocated by adding the equivalent number of elements, in 6 second intervals, to the start of the appliance load profile. Addition zero elements were added to the end of the array to ensure arrays were of consist length to ensure correct MATLAB syntax, as shown in Figure 5.

ApplianceIndividual = [zeros(1,starttimeMatrix(i)),

Appliance zeros(1,addedtimeMatrix(i))];

Figure 5: Extract of model showing start time allocation

The model was then programmed to calculate the peak load and load factor of the combined load profile to represent the business-as-usual scenario, and therefore providing the comparison results.

It was decided that the best way to determine the optimum offset time was to simply analyse each possible offset time and select the best performing permutation. The load profile data was at 6-second resolution meaning each offset time at a 6-second interval within the maximum delay length was analysed. Figure 6 shows the insertion of new zero elements to model the delay of the appliance.

for j = 0:delaymax

trans = ApplianceMatrix (i,:);

sync = [zeros(1,j),trans];

Figure 6: Extract of model showing load profile array adjustment

The load profile data was stored as two independent matrix, as shown in Figure 6 as ApplianceMatrix, which contained each occurrence of the appliance use in each row and the electrical consumption at each interval in each column. Therefore to offset the time to each interval a zero element was inserted at the start of the array, essentially shifting the data along by 6 seconds.

The script then calculated the performance indicators for that offset time, shown in Figure 7, before writing the result into an array for comparison once all other time intervals within the maximum delay time had been tested.

averageload_sync = sum(SyncMatrix,2)/numel(SyncMatrix); peakload_sync = max(SyncMatrix); loadfactor_sync = averageload_sync/peakload_sync;

perf_vec_lf (j+1) = loadfactor_sync; perf_vec_pk (j+1) = peakload_sync;

Figure 7: Extract of model showing performance indicator calculation and storage

Once completed the model selected the optimum offset time by comparing the performance indicators written into the ‘perf_vec_lf’ and ‘perf_vec_pk’ arrays, as shown in Figure 7. As two performance indicators were used it was decided that the optimum offset time was the one with the highest load factor increase that also didn’t not result in an overall increase in peak load. This concept is analysed further in the discussion section.

The results of a single simulation were highly variable and depended on the dwelling load profile at the selected start time. This was to be expect so in order to understand the aggregate performance of the system each simulation scenario was run 300 times, with the performance indicator results for the optimum delay time of each simulation stored in a CSV file. This data could then be statistically analysed to understand the variation of results and therefore provide a degree of certainly of the level of peak load reduction that could be achieved with each appliance.

Stage Two – Single Appliance Type in Multiple Houses Model Coding

After stage one of the modelling it was decided that the assumption that the household dwelling load profile would be known for a full 24 hours was too implausible. Although the habitual behaviours of many households would allow a self-learning program to predict the total house load profile with reasonable accuracy, the precision required for the synchronisation script to operate productively would be unlikely to be reliably achieved.

Despite this, the stage one modelling created a successful profile synchronisation script which was therefore carried over the stage two modelling.

The decision was made to rework the model to only synchronise each appliance load signature against the demand generated by other appliances in the system. The system would detect and track the progress of appliances within the network, thus being able to accurately predict the load created by each appliance for the duration of appliance cycle beyond the present moment.

This means that, for example, when a dishwasher is started at 3pm the system know that in 1 hour 40 mins, at 4.40pm, there will be a large electrical load, lasting 20 minutes, due to the dishwasher drying its contents. Therefore when another dishwasher within the system is started at 3.10pm the system can see that there will be an overlap of 10 minutes where both dishwashers will be in the drying segment of their cycles. The system can This means that, for example, when a dishwasher is started at 3pm the system know that in 1 hour 40 mins, at 4.40pm, there will be a large electrical load, lasting 20 minutes, due to the dishwasher drying its contents. Therefore when another dishwasher within the system is started at 3.10pm the system can see that there will be an overlap of 10 minutes where both dishwashers will be in the drying segment of their cycles. The system can

Figure 8: Showing an example of a successful synchronisation using 25 dishwashers

Multiple houses were added to the scripting to imitate the use of multiple devices within a 24-hour day. The complexity of programming a multi-appliance simulation was increased however this would more accurately represent the results that could be achieve in real situations by accounting for all appliances to be included in the proposed system.

Another difference from the stage one modelling was that data was handled using matrices instead of arrays. This meant that data such as the weighted random start times could be stored as a single variable where each row of data corresponding to an instance of an appliance operation. This allowed the system to use far fewer variables and therefore run at a faster speed.

Testing Summary

The second stage of modelling involved a number of tests to determine the effectiveness of the system. The tests are designed to examine a number of factors that the system would be exposed to under real life conditions. These factors include varying number of houses, and therefore appliances, within the system network, variations on the maximum delay time and varying the appliance load profile data used. Additionally it was considered important to analyse the effect of the system where the load profile produced after the optimisation occurred varied compared to the profile used in the synchronisation script. Each of these tests were performed with Washing Machines and Dishwashers individually as well as a combined scenario.

The model simulates the scenario in each test 300 times, in three sets, in order to gain an average readings of peak load and load factor reduction as a percentage of the business as usual scenario. The scenarios are undertaken a large number of times in order to produce an accurate picture of the variance, range and averages in the performance indicators, peak load and load factor reduction. Based on the available computing power the tests were undertaken in sets of 100 to reduce the time each run took to complete.

Test One – Number of Houses within the system

This test was designed to gain an understand as to how many houses, and therefore appliances, were required to

be part of the system in order to produce effect peak load reduction. The simulations were run using the following number of houses, 5, 10, 25 and 100. This was based on the findings of the literature review, that coincidence factors had more potential to be reduced for networks with fewer than 100 houses. It was assumed that each house contained one appliance that would operate once over a 24 hour period, with a weighted start time, and with each operation producing the same load profiles. For this test, the maximum delay period was set at 10% of the cycle length.

Test Two – Appliance load profile variation

This test is designed to ensure that results obtained through test one are not simply due to the shape of the particular appliance load profile used and therefore to show that small variations in load profiles do not significantly affect the systems results. There are many manufacturers of appliances and therefore it was important to determine if the results are a property of all washing machines or just of the washing machine load profile used in test one. Load profiles used in this test were from the house two data set.

Test Three – Delay period variation

In this test the maximum delay period was varied to analyse the effect that it has on the performance indicator results. The maximum delay period was tested at 5%, 10% and 15%. These were undertaken assuming 25 houses and used the profile data from the house one data set.

Test Four – Load profile inaccuracy test

In reality the system would use a stored load signature of each appliance in order to carry out the synchronisation. The actual load profile produced by the appliance during operation would likely have small variations caused by external conditions. Therefore, test four is designed to determine what effect the load profile inaccuracy may have on the system performance. To implement this test, after the model has determined the optimum delay time, a new appliance load profile is used in place of the profile used to calculate the delay time. The model then continues as before and calculates the performance indicators.

It is assumed that 25 houses are included in the system and the maximum delay period is set at 10%. The two profiles were both selected from house one so that the consumption deviations would be small and therefore represent the deviation to be expected on a day-to-day basis.

Results

Test One – Number of Houses within the system

Test One was undertaken on both the washing machine and the dishwasher appliances separately. The aim was to understand what scale the system worked most productively at when using each of the appliances. They were each tested when using 5, 10, 25 and 100 houses and the simulation run 300 times for each number of houses. The results of the tests have been summarised in Table 4 and Table 5 below. It is important to note that the values shown are based only on the load generated by the appliance in being tested and not a whole house load.

Table 4: Results of the Washing Machine under test 1 conditions

100 Houses Washing Machine

Peak load

Load

Peak load

Load

Peak load

Load Peak load

factor reduction

Average 10.2 7.9 7.1 5.6 7.5 6.1 4.6 3.2 Maximum

Negative Results

Standard Deviation

value Upper 99%

Confidence Interval Lower 99%

Confidence Interval

Dokumen yang terkait

ANTARA IDEALISME DAN KENYATAAN: KEBIJAKAN PENDIDIKAN TIONGHOA PERANAKAN DI SURABAYA PADA MASA PENDUDUKAN JEPANG TAHUN 1942-1945 Between Idealism and Reality: Education Policy of Chinese in Surabaya in the Japanese Era at 1942-1945)

1 29 9

EVALUASI PENGELOLAAN LIMBAH PADAT MELALUI ANALISIS SWOT (Studi Pengelolaan Limbah Padat Di Kabupaten Jember) An Evaluation on Management of Solid Waste, Based on the Results of SWOT analysis ( A Study on the Management of Solid Waste at Jember Regency)

4 28 1

FAKTOR-FAKTOR YANG BERPENGARUH TERHADAP PENDAPATAN TENAGA KERJA PENGRAJIN ALUMUNIUM DI DESA SUCI KECAMATAN PANTI KABUPATEN JEMBER The factors that influence the alumunium artisans labor income in the suci village of panti subdistrict district jember

0 24 6

Implementasi Prinsip-Prinsip Good Corporate Governance pada PT. Mitra Tani Dua Tujuh (The Implementation of the Principles of Good Coporate Governance in Mitra Tani Dua Tujuh_

0 45 8

Improving the Eighth Year Students' Tense Achievement and Active Participation by Giving Positive Reinforcement at SMPN 1 Silo in the 2013/2014 Academic Year

7 202 3

Improving the VIII-B Students' listening comprehension ability through note taking and partial dictation techniques at SMPN 3 Jember in the 2006/2007 Academic Year -

0 63 87

Teaching speaking through the role play (an experiment study at the second grade of MTS al-Sa'adah Pd. Aren)

6 122 55

The Effectiveness of Computer-Assisted Language Learning in Teaching Past Tense to the Tenth Grade Students of SMAN 5 Tangerang Selatan

4 116 138

The correlation between listening skill and pronunciation accuracy : a case study in the firt year of smk vocation higt school pupita bangsa ciputat school year 2005-2006

9 128 37

Designing the Process Design Process 001

1 44 9