A fast and efficient back propagation algorithm to forecast active and reactive power drawn by various capacity Induction Motors

Abstract

Power system operators/planners are always face problem regarding reactive power compensation. Reactive power plays an important role in maintaining voltage stability and system reliability. In this paper, a new algorithm based on back propagation neural network is used by using suitable number of layers and various constants is presented, for forecasting the active and reactive power consumed by various capacities Induction Motor. Firstly, Database of active power (P) and reactive power (Q) for different voltages and frequencies are generated through real time experiment on various capacities Induction Motor. Then, Back propagation Neural Network (BPNN) is designed to predict the P and Q drawn by in induction motor for different voltages and frequency condition. Back Propagation technique is used for training. These trained BPNN models are used to predict P & Q for many unseen operating conditions and the results are found to be coming fast and very accurate.

Review of Big Data Analytics for Smart Electrical Energy Systems

“…The feasibility of ANN-based approaches for load forecasting has been validated via a few industrial applications. For example, in [36,37], an ANN-based approach has been used for reactive power prediction. The BDA-based method developed in [8,9,38] has been widely used for short-term load forecasting, with the mean forecast errors around 2% to 5%.…”

Accuracy of ANN based methodology for load composition forecasting at bulk supply buses

“…The AI-based total demand forecasting approach includes a training process and a validation process. As the approach for both P and Q forecasting has been introduced in [1,3,4], in this paper, only data collection, ANN parameter configuration and training algorithm selection are discussed in details.…”

“…In [1,3,4], demands of previous days are chosen as the input for ANN training, which induces obvious forecasting errors when the dates transfer from weekdays to weekends and vice versa. To reduce such errors, one of the ANN inputs for training in this study is selected as the normalized historical average daily demand curve (NHADDC) of either the weekday or the weekend, dependent on the type of the forecasting day.…”

“…The disaggregation approach can be summarized as follows: i) generate random voltages and weighting factors for training, and derive all combinations of voltages and weighting factors; ii) use the result from i) and Equation (2), (4) and (5) to define the input (P&Q&V) and the target (weighting factors) for ANN; iii) train ANN and save trained ANN; iv) remove voltage in ii), repeat iii); v) repeat i) to generate data for validation; vi) validation using saved network in both iii) and iv) and compare the two cases via confidence level analysis; vii) if results for both cases in vi) are comparable, which means the disaggregation can be implemented without voltage, then connect disaggregation box with forecasting boxes and analyse joint confidence level. Further details will be described in the subsections.…”

“…Significant effort has been put in the past into both total demand forecasting and load disaggregation. Artificial intelligence (AI) tools such as artificial neural networks (ANN) and (adaptive neural) fuzzy inference systems (FIS) have been widely applied to both real power forecasting [1,2] and reactive power forecasting [3,4] and have demonstrated high accuracy. As far as load disaggregation is concerned, intrusive load monitoring [5] and non-intrusive load monitoring [6,7] were the most widely used approaches.…”