Abstract:
The daily and weekly energy consumption patterns at the Transnet Port Terminal (TPT) in East London varies stochastically. This is as a result of the transient weather patterns that exist at the harbor. It has therefore become imperative to wisely manage this load in order to save electricity costs and for future infrastructure development. Hence the ongoing supply of electricity to port consumers requires an accurate and adequate short-term load forecast (STLF) for quality, quantity, and efficient management.
Many researchers have recently proposed Artificial Neural Networks for short-term load prediction. However, most of the studies have not considered the quickly changing weather patterns that exist at the port. Therefore, the objective of this study is to establish a supervised short-term load prediction using ANN models, and to verify the effectiveness of such predictions by using the real load data from the TPT. The suggested system architecture uses open- loop training with real load and weather information, and then a closed-loop network is used to produce a prediction with the predicted load as its feedback data.
Data collection points were set up in the ring network of the port by installing new power measuring meters, and weather data obtained from local meteorology offices in order to build a suitable alternative of localised data management (data base) for saving all data gathered. Hence, profiling of the load in the TPT was done and load forecasting was carried out, leading to improved load management strategies for the harbor terminal. ANN short-term load prediction (STLP) models were developed utilising its own performance to improve precision by essentially implementing a load feedback loop that is less reliant on external data. To ensure that the timeseries data recorded at the port were well modeled, the Nonlinear autoregressive exogenous model (NARX) for load prediction were developed using mean squared error (MSE) as a performance metric.
Furthermore, to show the efficacy of the proposed model for STLP, the adaptive neuro-fuzzy inference system (ANFIS) was used with the same data for short-term predictions. The minimum mean squared errors obtained for both NARX and ANFIS models were 0.0010939 and 0.0032 respectively, indicating that the NARX model is more accurate during the forecast of departmental loads. The results of the predictions using the hourly timeseries indicated a close match between the forecasted and actual load demand at the port terminal. The effects of the load forecast could be used as a guide for implementing management plans for internal load, such as the generation of urgent electricity and the programme of implementation for demand-side management policies.
