Ann-Trained Using Bat Algorithm for Modeling University-Based Energy Consumption on Short Term Basis

Adequate planning and right decision making in the energy sector lies on accurate forecasts of the load demand. In this paper, Artificial Neural Network (ANN) trained via Bat algorithm was employed for short term load projection of University of Ibadan, Nigeria. Daily load demand of the study area was obtained from the log records. The neural network was built, trained with historical data gotten from the premier University in Nigeria and then used to predict 24 hour’s load demand from Dec., 1 st to Dec., 7 th , 2016. The experimental results indicated that the proposed method achieved a Mean Absolute Percentage Error (MAPE) of 6.60% and a Mean Percentage Error (MPE) of 4.17%. This research finds application in scheduling of power demand in power system.

south-south Nigeria, ANN was trained with back propagation so as to reduce the error-based function. The authors submitted that interval type 2 fuzzy logic-ANN was able to handle the uncertainties and flexibly tuned the interval type 2 fuzzy logic parameters which resulted in better forecasting of the study case with minimized error in forecast. Having examined the earlier works reported with the use of ANN on Nigerian grid, this works therefore aim to extend the effort of earlier works by using Bat Algorithm to train the hidden layer of ANN. The rest of this paper is organized thus; material and methods were discussed in section 2, result and discussion was presented in section 3 while section 4 presented the conclusion.

Material and Methods
This section presents the material and methods adopted in the course of this work.

Data Collection
A non-weather data was used for this work, a daily energy consumption of the University of Ibadan as contained in the daily hourly load reading sheets for a week was collected as shown in Table 1.

Artificial Neutral Network Structure
ANN is a computational model that is structured to function in similitude of biological neural networks. ANN is endowed with capability to learn from training data set and this feature of ANN is termed adaptive learning. After learning by training which can be supervised or unsupervised, it creates its own structure by way of selforganization. Three stages are involved in ANN: training, testing and validation. In this present work, the whole data set was divided into two sets: training set and testing set. Training set consists of 80% of whole data and testing set contains the rest data. The training set was used to make a model which, therefore, predicts the load in the future. The number of hidden neuron was automated with the aid of Bat Algorithm (BA). The parameter sets of BA used in this work is as shown in Table 2 and the summary of ANN structure used in this work is as shown in Table 3.   Variable used as input hour of the day, day of the week, and daily load demand 3 No of hidden neutrons Automated using bat algorithm 4 No of iterations 100 5 Training algorithm Bat Algorithm 6 Hidden layer transfer function Sigmond 7 Output layer transfer function Sigmond

Performance Metrics
The prediction accuracy is a good approach to measure the accuracy of artificial neutral network. In this work, mean percentage error (MPE) and mean absolute percentage error (MAPE) was used to assess the accuracy of the forecast. The least value of MPE and MAPE is an indication of optimum performance of the proposed techniques. Mean Percentage Error; = ∑ × 100% (1)

Mean Absolute Percentage Error
where, t x = real load at time ( t ), f x = forecasted load at( t )time.

Result and Discussion
This section presented the outcome of this research which aimed to project the hourly load demand of a typical university environment. The study case used is the premier University; University of Ibadan, Ibadan, Oyo State Nigeria. Figure 1 to Figure 7 depicted the actual and forecasted load in Megawatts (MW) for each day in the month of December, 2016 were plotted against the time in hours. Figures 1 to 7 show that ANN model developed produced an output that is much closer to the actual output. It was observed that the peak load values for Tuesday, Wednesday, Thursday and Friday were high due to the considerable operations of residential and commercial loads while the peak load value for Monday is lower due to its proximity to the weekends. Saturday and Sunday being the weekends have the lowest peak value due to negligible operation of commercial and industrial loads within the university environment.   Table 4, it can be seen that the ANN showed higher forecasting error in the days when people have specific start-up activities such as Tuesday and variant activities such as during Sundays. This is probably because of pick-up loads associated with such days. Also, from Table 4, the average MPE and MAPE for this model was found to be 4.17% and 6.60% respectively; this implies a high degree of forecasting accuracy for this model in spite of the simplicity of its input variables and the low volume of data used. Notwithstanding the scale sensitivity of MAPE which would often time take extreme values for low volume data, the average MAPE is still small.

Conclusion
This paper presented application of ANN trained with Bat algorithm to forecast university-based load demand on short term. The forecast results showed that ANN is well suited for this kind of load forecast. Its forecasting reliabilities were evaluated by computing the mean absolute error between the exact and predicted values, the output curves obtained from ANN model developed showed the forecasted load traced out almost same path with the actual consumed in each day of the week. This work when integrated into the power system will go a long way in helping the operator to know the strategies to be adopted once the exact projections of the load expected in future time are known.