The technique selects one or more areas with the properties similar to those of an isolated area. It develops a relationship between the load-dependent factors for the selected area(s) with those for an isolated area. However, there is no conceptual difficulty in using the method to predict the loads of any isolated area.
RESULT (MAXIMUM DEMAND) OBTAINED FOR
LIST SYMBOLS
LIST OF TABLES
I NUMBER OF
CHAPTER -1
INTRODUCTION
INTRODUCTION
From recent years the situation has changed and load forecasting is getting more attention than before. Load forecasting is the prediction of the electrical load on a given system at a future time. Different approaches of load forecasting, types of load forecasting and forecasting models are presented in chapter-2.
CHAPTER - 2
LOAD FORECASTING TECHNIQUES
FORECASTINGMODELS
A model is a representation of reality; not any representation but a representation prepared in function of the decisions to be taken. The symbols used on the map constitute the information marked only as a function of the decisions to be made: optimizing the transmission of energy. Therefore, the preparation of a model requires the selection of significant elements of reality and their ordering according to the range of decisions to be made.
The underlying basic trend is derived from historical demand data using the time series regression analysis. The forecast of future electricity demand and projections of the various explanatory variables must then be available. In this type of model, the change in the underlying trend is qualitative – based on the judgment and.
The first consists of the regression analysis of the historical demand for electricity and GNP, industrial production, population data, etc. A similar procedure is possible when a clear trend of the elasticity coefficient can be deduced from the historical data. Therefore, they are recommended only if major changes in electricity use patterns, electricity tariffs, structure of the economy, etc. are not expected, which would modify the parameters of the regression equation of the elasticity coefficient.
Furthermore, the success of correlation methods depends largely on the ability to accurately predict future GDP, industrial production, and population.
CHAPTER- 3
Remember that an isolated area is a place where electricity consumers cannot be connected to the main grid for technical or economic reasons. Some isolated areas may be industrially or commercially important and may have developed their own electricity systems. This chapter presents a methodology [27, 28] to predict the loads of an isolated area where the load history is not available. If this is the case, the information does not reflect the realistic demand for electricity.
The methodology developed by AHoque [27, 28] is based on the identification of the factors on which electrical load growth depends. Area/areas with the known load-determining variables similar to those of the isolated area are selected to evaluate the contribution of load-determining variables. A suitable area in which the load-determining variable is similar to that of the typical.
Then the load of the typical isolated area is derived from the load of the selected area. The total electrical load demand, L(t) in an isolated area is the sum of the above four types of loads. Therefore, the load of an isolated area can be expressed as. 3.6), expresses that the load is a function of six time-dependent variables.
Since the contributions of different areas in the evaluation of the burden of an isolated area may in fact be different, the values of Pj's may therefore differ.
CHAPTER-4
DATA FOR LOAD FORECASTING
Deciding on the appropriate variable to predict depends largely on the needs of the planner. The contribution that the forecast will make to the planner to determine the time period that each value of the variable should cover, the level of detail required, the frequency with which it is requested, the required level of precision, the appropriate segmentation of the variable and the value of the forecast. Because of the importance of each of these aspects in defining appropriate predictor variables, all of these aspects are discussed very briefly.
The frequency with which data is collected is related to the time period covered by each value of the variable. A careful conservation of the data from Table 4.1 shows that there is closeness between the factors of some areas, but the peak demand is very different. Note that all the PBSs grouped in Table 4.2 belong to almost similar Adult Literacy Rates.
From Table 4.3 and Table it is observed that the value of the weighting factor related to adult literacy rate is the highest, while that related to percapita income is the lowest. However, it observed from Table 4.2 that the maximum demand of Jessore-2 is the highest among other PBSs but the literacy rate is the lowest. It is also observed from Table 4.1 that maximum demand Jessore-l is higher than Jessore-2 but its literacy rate is lower.
From table 4.5 it can be observed that the actual requirements (maximum and average) vary greatly from the calculated requirements for this particular case due to some undesirable negative effects on Population, Road Length (Communication) and Distance from Load to Water.
CHAPTER-5
BACKPROPAGATION TRAINED NEURAL NETWORK
When the network receives an input, the update of activation values propagates forward from the input layer of the processing units, through each inner layer, often called hidden layer, to the output layer of processing units. When signal patterns are applied to the input layer of the network, it propagates upwards towards the output layer through the connections (commonly called weights) of the intermediate layer, known as hidden layer. Output from the hidden layer is then propagated through the weights between hidden and output layers and an output pattern is generated.
However, each unit in the intermediate layer receives only a fraction of the total error signal, which is roughly based on the unit's relative contribution to the original output. After training, when presented with an arbitrary input pattern, the units of the hidden layers of the network will respond with an active output very close to the target value. This tendency can be extrapolated to the hypothesis that all hidden units are somehow associated with specific features of the input pattern as a result of training.
The error that is minimized by GDR is the sum of the squared errors of all output units. But in case of very complex map, the size of the hidden layer may be large for the convergence of the network. If a local minimum is reached, the grid error may still be unacceptably high.
Choosing the value of the learning rate parameter, 11, has a significant effect on the learning rate.
CHAPTER-6
COMPARISION WITH LINEAR REGRESSION ANALYSIS
This is not surprising because the coefficients obtained from LRA have no physical meaning. Moreover, the coefficients, even given their negative sign, gave inaccurate results when used to predict other requirements for known values (Table 4.5). On the other hand, the trained NN, when used to predict the demands from an area with a similar climate, has given a close agreement with the actual values (Table 6.4).
Instead, the NN method should be used to predict loads of an isolated area such as Sandwip. 01 predicted load of an isolated area, calculated by NN method 02 Expected load of an isolated area, calculated by LRA.
CHAPTER -7
CONCLUSION AND RECOMENDATIONS
OBSERVATION AND DISCUSSION
The weighting factors evaluated according to equation 3.8 of the isolated area load prediction technique are shown in Table 4.3. For this evaluation, data from 6 PBSs of almost similar adult literacy rates were considered. From Table 4.3 it is observed that the weighting factors related to adult literacy rate are the highest while those related to percapita income are the lowest mentioned earlier.
But from the same Table 4.3 it is also seen that weighting factors in relation to population, road length e.g. For this reason we decided to predict the load of the typical island by using neural network method.
CONCLUSION
Therefore, the objective of this research has been to develop a load forecasting technique suitable for an isolated area.' The main objective of this research has been achieved by presenting a technique of forecasting the load of an isolated area where either the supply of electricity as an energy source has not yet started or the history of the development of the load is not available.
RECOMMENDATIONS FOR FURTHER RESEARCH
Ahsan and Eheteshamul Hog, “Bibliography of Load Forecasting,” Electrical and Electronic Research Bulletin, BUET., Vol. 7] Ibrahim Moghram and Saifur Rahman, “Analysis and Evaluation of Five Short-Term Load Forecasting Techniques,” IEEE Transactions on Power Systems, Vol. Miri, “On-line weather-sensitive and industrial group bus load prediction for microcomputer-based applications,” IEEE transaction on, pp.
Scheer and BA Amith, "Integrating Load Management Programs into Utility Operations and Planning with a Load Reduction Forecasting System", IEEE Transaction, Vol. Malhami, "Statistical Synthesis of Physically Based Load Models with Applications to Cold Load Pick-up", IEEE Transactions, Vol. 28] Hoque, A., Ahsan, Q. and Beattie, W.e., "Load Forecasting Technology for Isolated Area", Proceedings of ISEDEM' 93, Third International Symposium on Power Distribution and Power Management, Marina Mandarin Singapore, edited by C.S.
ANNEXURE -1
OBJECTIVES AND PRIORITY SELECTION CRITERIA OF RURAL ELECTRIFICATION
PROGRAM
PRIORITY SELECTION CRITERIA
In order to ensure that the final feasibility study on area selection would be done in an objective manner, the consultations developed criteria for priority base selection and a weighting factor was assigned to each criterion. The following table AI.I reflects the percentage of influence each criterion exerted on the final selection.
ANNEXURE -2
COMPUTER PRINT OUTS OF LRA
ANNEXURE -3
PROGRAMME FOR BACKPROPAGATION