Development of regional load management system based on rural, semi urban and urban loads-a critical analysis

Received Aug 2, 2021 Revised Nov 13, 2021 Accepted Dec 17, 2021 The sharp rise in population during the last half century has created immense pressure on the resources required for generation of energy essential to lead a comfortable and healthy lifestyle. The drive towards 100% electrification in developing countries like India has also contributed to this increase in demand. Till recently, fossil fuel was use to supply the bulk of this power. Now, the world is moving more and more towards renewable energy. This paper presents a model where several regions are combined together based on the demand profile of the regions segregated as urban, semi urban and rural along with the flexibility to schedule loads on the basis of availability of renewable energy sources within the area of the regions. The main focus is on detailed neural-networking based load forecasting and developing a load management system to manage load based on availability of distributed generation capacity and available tariff system. A solution is proposed in this paper based on a new approach to answer load management on the basis of region, population demographics and per capita energy consumption. A considerable amount of improvement to manage demand is intended to be attained and has been demonstrated in this research work.


INTRODUCTION
The electricity demand is predicted to increase substantially in the future. The major contributors would be the rising population, improvement in lifestyle, rural electrification and growing demand for air conditioning and digital devices in the semi-urban and urban areas. The growth of electricity demand is projected to be 2.1% per year globally, of which a major portion will be contributed by the developing countries. The increased generation to meet the demand will result in higher CO2 emissions and faster depletion of fossil fuels. These factors combined with the sustainable development goals are pushing renewable energy to the forefront, playing a major role in meeting the electricity demand growth [1], [2]. In such a scenario, the proper utilisation of the available resources and providing efficient service to the consumers can be done by effective load management systems [3]. Load management approach can either be targeted towards supply side management or demand side management as explored by Deka et al. [4] and Tan et al. [5]. Compared to the supply side management, demand side management has been a more explored area. The simplest approach towards demand side management considers the objective of peak load and consumer cost reduction proposed by Teive [6] but it is not the most efficient as it does not consider the integration of renewable sources or appliance usage pattern of consumers thus reducing consumer comfort. The application of computational intelligence is finding increased application in many fields [7]- [11]. It has

PROPOSED LOAD MANAGEMENT STRATEGY
In this section we first present the load switching and rescheduling models of the regional loads and then propose a method to determine the percentage of loads that can be supplied by renewable energy. This process will contribute towards renewable energy growth planning based on the composition and growth of load demand in the regions supplied by different utilities.

Model of switching schedule of loads
The area supplied by each utility can be divided into three regions Urban, Semi-Urban and Rural based on the population. The loads of every region can be further subdivided into residential, commercial, industrial, agricultural and public lighting. It is understandable that each area will have different concentration of loads based on the income and lifestyle of the people and accordingly the percentage of total load as well as loads of individual category that can be supplied with the renewable energy will vary. The loads of each type again comprise of non-shift able and flexible loads. In our work we have taken Residential and commercial loads of an urban and a rural area respectively. To provide an effective model for load to renewable energy mapping first it is required to develop a switching profile of each load and a rescheduling strategy to reduce the peak load and stress on generation facilities. In order to develop a switching profile fuzzy logic has been used in this work. It has been seen that a reasonable modelling of consumption pattern can be done by computational intelligence using individual appliances as basic building blocks [37]. An upper limit on the number of appliances that can be switched on at the same time is provided by a prefixed hourly maximum power consumption value. This value has been determined by forecasting with ANN. ANN was selected for it's efficiency in multivariate forecasting [38], [39]. The fuzzy rule base of the system has been developed after extensive study of the data provided by surveys over multiple cities and towns spread over different regions [40], [41].
The number of urban residential and commercial appliance ownerships are considerably higher compared to the rural areas. The usage patterns of the appliances were used to develop the Fuzzy Rule base an example of which is shown in Table 1. The time of the day has been divided into five membership functions namely early morning (EM), morning (M), afternoon (A), evening (E), late evening (LE). Also, the priority of the appliances and their probability to be switched on has been divided into 5 membership functions namely very high (VH), high (H), medium (M), low (L), very low (VL). Trapezoidal membership functions have been taken for both the input parameters whereas the output parameter has been modelled by triangular membership functions. The entire day has been divided into T time periods for each of which every appliance is assigned a switching value through the fuzzy logic program such as: Where Obj is the objective function and κ is the penalty function incorporated into the objective function to ensure that the appliances complete their runtimes. This has been used as a soft penalty function to allow a small amount of flexibility in the appliance runtimes. This can be justified with the fact that rt is the average runtime of the l appliance which means some appliances belonging to the l set will have less runtimes than the average while some will have more. The best population among multiple runs is selected based on the Obj function value. The area and load division structure are shown in Figure 1(a) and the structure of the fuzzy system is shown in Figure 1

Classification of regions
The load profiles that were developed using fuzzy logic are utilised to classify different regions based on the correlation between the output of the fuzzy logic program and the actual load demand input or the region. The synthetic load profile L and the actual load demand profile L are first normalized using the formula: where: L , , =fuzzy logic output of h hour of the i load of Y type of X region N=total number of loads of Y type in X region L =maximum value of the input load profile in time period T Then the correlation between them is obtained using the following formula: L ∈ XYifcor L , , L , ≥ 0.8 (12) where: ∀X ∈ [Urban, Semi Urban, Rural] ∀Y ∈ [Residential, Commercial, Industrial, Agricultural, PublicWorks]

Rescheduling of loads for better load factor
After formation of the switching profiles each appliance of Y type in X region is obtained as a , , a , … . , a , ∈ R × where M = ∑ ∑ ∑ a and N = H where X =total number of regions, Y =total number of consumer types and A = total number of appliances in under each consumer type of each region. Each appliance will have a different set of starting and stopping time t , t ∈ R ( × ) between which they can be rescheduled. This start and stop times have been decided by taking into consideration the different user data obtained from surveys and data repositories of utility services. PSO has been used to perform the rescheduling to minimize the objective function developed as: Where H denotes the maximum time slots the day has been divided into, M denotes the total number of appliances under consideration defined earlier, ϕ ∈ [0,1] denotes the switching status of the load satisfying the following conditions: P ( ) = the power consumption demand of ith load on the runtime slot corresponding to the h time slot of the day and avg. Load = ∑ ∑ P ( ) . The function of the load scheduling program is to bring the load demand factor as close to 1 as possible by filling in the valleys and reducing the peaks in the daily demand curve. This serves to ensure that the load consumption is fully optimised so that when the loads are selected for renewable or grid supply there is no wastage of resources due to inefficient load consumption pattern. The demand to be supplied at each time slot can be defined by: Where NSL (h) = j non-schedulable load at h time slot, CL (h) = k connected load at h time slot which was either already present in the time slot of concern or disconnected from another time slot and connected to the present time slot and DL (h) = l disconnected load which has been disconnected from the time slot in concern and shifted to another time slot.

Tagging the scheduled loads to determine which loads can be supplied by renewable energy
The loads are tagged in two parts-those which can be supplied by renewable energy and those who can be supplied by grid. The division of loads is done by using a fuzzy rule base, an example of which is given in Table 2. The renewable energy available in the region is classified into three categories: H, M and L. The power consumption of each load has been classified into five categories: VH, H, M, L and VL and the time duration of continuous consumption for each load starting from the hour under consideration is divided into five categories: VH, H, M, L and VL. On every hour the different loads are tagged to be supplied by either renewable energy or directly from the grid based on the following: where: S , =switch status of i load of Y type in the X region in (h + rt) hour 1 =possibility to be supplied by renewable sources 2 =supplied by grid λ =available power from renewable source energy of i type in X region N =total number of renewable energy sources in the Y region rt =runtime of i load of Y type in X region P =power rating of the i load of Y type in the X region W , =tagged status of i load of Y type in the X region at h hour The loads that are tagged each hour for being supplied by a certain source continues to be supplied from that as source for the consecutive hours in which it is in continuous operation. While classifying the different loads at those hours the load that is already tagged due to continuous operation from a previous hour is skipped but provided with the highest priority to be provided from the supply it is already tagged with.

Redistributing the regional loads in case of source failure
Sometimes a source may fail to deliver in a region at any hour such that ∑ P(act) , < ∑ P(pred) , where: ∑ P(act) , =actual available power from renewable sources of the X region in the h hour. ∑ P(pred) , =predicted available power from renewable sources of the X region in the h hour. In such a case it is necessary to reorganise the loads projected to be supplied by renewable energy. The loads are redistributed according to their priority and state of operation such that L , > L , ifrt > 0, rt = 0 and PR = PR ,

RESULTS AND DISCUSSION
Based on the mathematical discussion given in section 2 an algorithm 1 was developed to form a reference template to compare and distinguish between different regions and load categories, reschedule them to decrease the load factor and finally categorize them into the loads than can be supplied by renewable and those which need to be supplied by grid. The output of the program for providing the template for comparison is given in Figure 2. The seasonal variation of the daily load demand output for Urban residential area unit is shown in Figure 2(a) and that of Urban commercial unit area is shown in Figure 2 The data obtained from the fuzzy tool whose outputs have been shown above were used to find correlation with the actual energy consumption data obtained from different surveys and utility reports [40], [41]. The output of the correlation method provided good results in identifying both the region and type of consumer based on the normalized load demand values and the load demand profile. The result of the comparison has been shown below in Table 3. The urban residential demand curve scores above 0.8 when correlated with the urban residential data provided by the fuzzy tool whereas we get much lower correlation values when the actual urban residential demand data is correlated with the other outputs of the fuzzy tool. The same result is observed for the other regional data also. Thus, it can distinctly identify the load demand profile based on the region and the consumer category. After the region and consumer category was identified from the composite actual energy consumption data, the appliances most prevalent among the identified consumer category of the identified region type were selected from the list of appliances developed from the data obtained from online survey and physical survey. The number of these appliances were set according to the population of the region obtained from government websites and the ownership percentage of each appliance among the population. Then PSO was applied to reschedule the appliances to bring the load factor as close as possible to 1. For developing an effective rescheduling model, the loads of the different consumer categories of a region were classified into shiftable loads and non-shiftable loads. The start and stop times between which each load can be rescheduled was fixed considering the region type, consumer type and load type as these factors are very important in scheduling loads. A load of same type may have different operating schedules based on the region and the consumer who is using it. The output after scheduling the loads for combined area loads and rural residential loads is shown in Figure 3(a) and Figure 3  After rescheduling each load of every hour was passed through the fuzzy tool developed for tagging the loads to be either supplied by the renewable sources available or from the grid. The pseudocode of the developed fuzzy tool is given in Algorithm 1. The outputs providing the percentage and actual values of the rural residential hourly loads available for being supplied by renewbale in summer is shown in Figure 4(a) and Figure 4(b) respectively. The seasonal variation can be compared by observing the outputs providing the rural residential hourly loads available for being supplied by renewbale in winter shown in Figure 5(a) and Figure 5(b) respectively. This system provides us with a good insight into the characteristics of the load demand both region wise and consumer category wise. The major contribution of this work is providing a detailed model in which the loads are mapped to their probable supplies based on the region wise classification of urban, semi-urban and rural. The model is capable of identifying the type of region and consumer category to be addressed from the load demand profile of the region.  The percentage of loads that are tagged for being supplied with renewable energy is found to be largely varying both region wise and consumer category wise. The percentage share of the loads of each region that can be supplied from renewable sources is shown in Table 4. The load demand of the regions and the available renewable energy vary with changes in climatic factors. The demand supply variations have been shown in Table 5 referencing the values of which we can see a higher percentage of residential loads are selected for renewable supply compared to commercial loads. The reason being that residential units consume less power, are less critical and have more scheduling flexibility. For similar reasons, rural percentage of both the category of loads are higher than urban and semi urban. the values of urban and semi urban regions appear close to each other as the types of appliances used in these regions are nearly the same. The lower values obtained during winter can be attributed to the following two reasons. The first is lesser renewable availability due to decreased solar intensity and shorter days. The second is the reduced use of space cooling appliances resulting in increased percentage share of heavier and more crtitical loads, which are less likely to be selected for renewable supply by the fuzzy tool, in the total load demand of the region. So, it is evident that the developed tool can distinctly distinguish between the different load profiles.
Also, the performance of the proposed method to suggest a renewable integration approach with distinctly different percentages for the different regions and consumer categories is also satisfactory. Summarising, it can be said that the proposed method is capable of providing a region wise load management solution distinctly capturing the diversity of the load demand characteristics of different regions and consumer categories.

CONCLUSION
The paper describes a new approach towards load management and renewable integration method based on regional approach considering the load demand characteristics of different consumer categories. The region classification tool and load selection for renewable supply has been developed using fuzzy logic and the rescheduling of loads using different parameters varying regionally for a better load factor has been done using PSO. The proposed solution assures to contain the entire range of regions and loads as obtained from data available from online repositories and surveys of different organisations. It is shown by the test results that the performance of the proposed method in recognising the type of region and consumer category from the load demand profile is satisfactory. Summarising, it can be said that the proposed method is capable of providing a region wise load management solution distinctly capturing the diversity of the load demand characteristics of different regions and consumer categories.