Short-Term Load Demand Forecasting Using Artificial Neural Network

Currently, traditional electricity consumers are now shifting to a new role of prosumers since more integration of renewable energy to demand side. Accurate short-term load demand forecasting is significant to safe, stable, and reliable operation of a renewable energy-dominated power system. In this paper, a short-term load forecasting model based on a bidirectional long short-term memory network (Bi-LSTM) using kernel transfer operator is proposed to achieve short-term load demand forecasting. To consider the influence of seasonality, holiday effects and weather on load demand forecasting, and simultaneously to improve the accuracy and performance of the forecasting model, this paper implements the dimensionality reduction of the input data by introducing an improved kernel transfer operator based on the Perron-Frobenius method. On this basis, a Bayesian Bi-LSTM for short-term load demand forecasting is formulated to obtain the probability prediction interval of load demand. To verify the validity of the proposed method, the actual historical load demand in a certain region in China was used for model training and verification. The forecasting results are compared with several conventional load demand methods using probability prediction technics. Simulation analysis shows that the proposed method can effectively reflect the short-term uncertainty of load demand, and is superior to the conventional methods in terms of forecasting accuracy and computational performance.

Electricity demand forecasting has significant impact on planning and operation of a power system. Parameters that affect short and long-term load forecasting are the temperature, calendar day, geographical variations, gross national product, socio-demographic trends, energy efficiency etc. The weather conditions seriously affect load demand on short term. This paper focuses on a comparison of different practical methodologies for short-term load forecast and their application and implementation on real load demand and temperature data for Republic of North Macedonia. The paper provides detailed comparison of several practical algorithms for short-term load forecast: polynomial and sinuses regression, machine learning and artificial neural networks. The power load is considered from the aspect of two variables – temperature and calendar date. A case study is presented and results are discussed and analysed. Finally, conclusion and recommendations are presented.

Electrical load forecasting is one of the important parts for smart grid system. The reliable prediction of the load demand contributes to the efficient and economical operations and planning. The artificial neural network is used extensively in load demand forecasting. The nonlinear nature of the electrical load demand conforms to the ability of the artificial neural network in calculating the nonlinear relationship of inputs and outputs. Among many models of neural networks, radial basis neural networks yield superior performance in small error and fast simulation time. However, it is challenge to design the radial basis neural networks. The excessive numbers of hidden neurons lead to lacking of generalization or so called overfitting problems. This paper proposes an approach to design the radial basis neural networks that use as least numbers of hidden neurons as possible. The error criterion is optimized based on modified genetic algorithm as the numbers of hidden neurons are incrementally increased. Simulation results of short term load forecasting are calculated in Matlab, and compared to the orthogonal least square error method. The proposed approach gives better results with the same numbers of hidden neurons.

A load frequency controller (LFC) is a crucial part in the distribution of a power system network (PSN) to restore its frequency response when the load demand is changed rapidly. In this paper, an artificial neural network (ANN) technique is utilised to design the optimal LFC. However, the training of the optimal ANN model for a multi-area PSN is a major challenge due to its variations in the load demand. To address this challenge, a particle swarm optimization is used to distribute the nodes of a hidden layer and to optimise the initial neurons of the ANN model, resulting in obtaining the lower mean square error of the ANN model. Hence, the mean square error and the number of epochs of the ANN model are minimised to about 9.3886 × 10−8 and 25, respectively. To assess this proposal, a MATLAB/Simulink model of the PSN is developed for the single-area PSN and multi-area PSN. The results show that the LFC based on the optimal ANN is more effective for adjusting the frequency level and improves the power delivery of the multi-area PSN comparison with the single-area PSN. Moreover, it is the most reliable for avoiding the fault condition whilst achieving the lowest time multiplied absolute error about 3.45 s when compared with the conventional ANN and PID methods.

This paper presents the optimal operation of a building energy management system (BEMS), with combined heat and power (CHP) generation, thermal energy storage (TES), and battery energy storage (BES), subject to load demand uncertainty. The main objective is to reduce the total operating cost (TOC) and total CO2 emission (TCOE). First, we develop two models of load demand forecasting, one for weekday and the other for weekend, using artificial neural networks, long short-term memory, and convolutional neural networks. Then, we incorporate the predicted load demand and load demand uncertainty for planning the energy dispatch of the BEMS. TES aims to store the thermal energy waste from the power generation of CHP and discharge the thermal energy to the absorption chiller to supply the cooling load. BES and spinning reserve (SR) play an important role in handling the uncertainty of the load demand. The operation of BEMS, subject to the load demand uncertainty, is formulated as a linear program. We can efficiently solve the linear program and provide an optimal solution that satisfies the dispatch constraints. Thereafter, we determine the optimal size of BES, based on economics and environmental optimal operation. The proposed BEMS is compared to the previous BEMS, without BES and SR. Furthermore, we propose the multi-objective optimal operation, where the normalization for TOC and TCOE is introduced, and the multi-objective function is defined as a linear combination of normalized TOC and TCOE. The numerical results reveal the trade-off relationship between TOC and TCOE. In particular, when TCOE is minimum, TOC becomes maximum. On the other hand, when TOC is minimum, TCOE becomes maximum. The relationship provides a method to select the operating point, as well as analyze the power flow for the multi-objective optimal operation.

With increasing research on the field of alternative energy resources for sustainable development, more renewable energy sources are getting penetrated to the distributed network to form "microgrid". This leads to more complexity in the distribution network in terms of energy management and realtime control. The objective of this paper is to design a system that forecasts the short (daily), medium (seasonal) and long term (yearly) load demand and the availability of energy resources at the microgrids. For scheduling the storage and transaction of electrical energy between neighboring microgrids, an energy management system (EMS) is designed. The EMS makes use of the forecasted data and real time data all together for managing an array of interconnected microgrids. The seasonal and yearly forecaster for a geographical boundary helps in maintenance scheduling and long term infrastructure development plan respectively. Recently, Artificial Neural Network (ANN) is found as a promising tool for statistical forecasting in real time applications. Hence, this paper makes use of ANN feature to forecast both load and availability of energy resources at microgrids in different scenarios like daily, seasonal, and yearly. The layered ANNs architecture is developed and trained with Levenberg-Marqurardt Back Propagation Algorithm. The entire design is simulated in MATLAB/Simulink. The proposed concept can be used in today's real time energy infrastructure to prevent future energy crises with improved reliability and smooth coordination among microgrids located at different areas.

Buildings are one of the major sources of greenhouse gas emissions and electricity consumption in urban areas all around the world. The load demand of large buildings is highly uncertain due to large penetration of solar PV. As a result, it leads to serious power system stability and quality issues for network operators and energy managers. Therefore, accurately forecast the load demand of buildings is utmost important to design a better energy management system and reduce greenhouse gas emission. However, variable PV output power, random nature of weather, diversity and complexity of buildings are big hurdles to accurately predict the load demand. In this paper, a deterministic hybrid intelligent forecast framework is proposed based on a combination of cascade-forward back-propagation network (CFBPN) in ensemble network for accurate load demand forecast. The wavelet transform (WT) technique is applied to handle the sharp spikes and fluctuations in historical load demand data. In addition, particle swarm optimization (PSO) is used to train the CFBPN in ensemble network for better forecast accuracy. In proposed forecast framework, historical load demand data, temperature, wind speed and humidity are applied as ensemble model inputs. The performance of proposed model is analyzed for one day and 12 hours ahead load demand forecast of summer (S), autumn (A), winter (W) and spring (SP) days. The proposed forecast model provides higher forecast accuracy compared to autoregressive (AR) and wavelet transformed backpropagation neural network (WT+BPNN).

The ability of Artificial Neural Network (ANN) for estimating the natural gas demand load for the next day and month of the populated cities has shown to be a real concern. As the most applicable network, the ANN with multi-layer back propagation perceptrons is used to approximate functions. Throughout the current work, the daily effective temperature is determined, and then the weather data with the gas consumption data of the last days are used for network training. It is shown that nearly 93% and 98.9% of the result is in a good agreement with the real data for the daily gas load forecasting and those of the monthly respectively. These results clearly show the capability of the presented networks. The method, however, can further be developed for prediction of other required information in various industries.

Grey wolf-optimized artificial neural networks used in DC–AC hybrid distribution networks, to regulate the energy consumption, is presented in this study. Energy management system that takes into consideration, the distributed generation, load demand, and battery state of charge are being considered. The artificial neural network have been trained, utilising the profile data, based on the energy storage system’s charging and discharging characteristics, under various distribution network power conditions. Moreover, the error rate was kept, well under 10%. The suggested energy management system, that employs an artificial neural network, has been trained to function in the optimal mode, utilising grey wolf optimization for each grid-connected power converter. Small-scale hybrid DC/AC microgrids have been developed and tested, in order to simulate and verify the proposed energy management system. The grey wolf optimized neural network energy management system has been proven to provide 99.48 % efficiency, which is superior when compared to other methods existing in the literatures.

A novel two-phase approach to manage the daily operation of proton exchange membrane (PEM) fuel cells for residential applications is presented in this paper. Conventionally, the performance optimization is carried out offline since it is a time-consuming process and needs high computational capabilities. To simplify the management process and to enable online parameter updating, the paper suggests a new technique using artificial neural networks (ANNs). First, a database is extracted by performing offline optimization processes at different load demands and natural gas and electricity tariffs using a genetic algorithm (GA). Then, the obtained results are used for the offline training and testing of the ANN, which can be used onsite to define the settings of the fuel cell. The tariffs and load demands as inputs of the ANN can be easily updated online to enable the ANN to estimate new optimal or quasioptimal set points after each variation in operating points. The agreement between ANN decisions and optimal values as well as the achieved reduction in operating costs encourage the implementation of the proposed technique to achieve both fast online adaptation of settings and near optimal operating cost. This technique is applicable for different distributed generating units (DGUs), which are expected to spread within the power systems in the near future.

This paper presents an investigation of an artificial neural network (ANN) to predict the flutter characteristics of a simple wing that under went multiple weight variation changes. The weight variations were used to simulate the addition of non-structural weight or structural repair weight. The test case wing had 24 designated repair areas where repair weights were added as lumped masses. Weight and location of the repairs were randomly generated to increase the sample size and help the generalization of the ANN. The ANN was evaluated for both the flutter speed and flutter frequency for multiple levels of structural damping. The number of test cases required to train the neural network was also evaluated for both the error between the input and output results as well as the generalization of the neural networks.

This study focuses on the production of durable and high-quality concrete that aligns with the United Nations sustainable development goals (SDGs). Specifically, it aims to fulfil SDG 9 (industry, innovation and infrastructure) and SDG 11 (sustainable cities and communities). However, producing fibre-reinforced ultra-high-performance self-compacting concrete (FRUHPSCC) presents a challenge in achieving the desired mechanical properties. As a result, constructing numerous trial samples increases cost and time. To address this issue, an artificial neural network (ANN) can accurately predict the FRUHPSCC's mechanical properties. The study utilised garnet and basalt aggregates, nanosilica, steel fibre and other components to make FRUHPSCC and tested its compressive and tensile strengths and microstructure. By utilising a data set of experimental results, five types of ANN were developed with different training algorithms, as were five hybridised types of ANN employing the grasshopper optimisation algorithm (GOA), that predicted the compressive strength of this type of concrete. The results indicated that their predictions were highly accurate, and the hybridisation of ANNs with GOA increased prediction accuracy further. Notably, the network that combined the training function trainlm and GOA produced the highest prediction accuracy, showing that ANNs can predict FRUHPSCC's compressive strength accurately while reducing production cost and time.

Electricity load forecast needs to ensure minimum load wastage and requires intelligent decision-making systems to accurately predict future load demand. Learning capability, robustness and ability to solve nonlinear problems make ANN widely acceptable. For accurate short-term load forecast (STLF), an ensemble soft computing approach, namely ANN-PSOHm, composed of an artificial neural network (ANN) and particle swarm optimization (PSO) with homeostasis based mutation is presented in this article. To enhance the learning strength of ANN, PSO undergoes homeostasis mutation to bring greater diversity among solutions by exploring in wider search space. To demonstrate the effectiveness of ANN, three case studies on load data of NEPOOL region (courtesy ISO New England) are performed. The experimental results show that ANN-PSOHm improves accuracy by 11.57% MAPE over ANN-PSO.

This paper presents the development of Short Term Load Forecasting (STLF) model using Artificial Neural Network (ANN). STLF is required for electric power planning and electricity market planning. The proposed model predicts the load demand of Connecticut in the U.S. using hourly historical electric load and weather data. For improving the load prediction accuracy, we consider two main issues that are seasons and weather factors. Each season has different load demand patterns, thus the weather factors are differently applied in each season. The proposed model uses the composited weather factor which consists of temperature and dew point. The temperature and dew point weather factors are selected through the correlation coefficient to obtain the meaningful data among the weather factors. The selected weather factors adjust the level of the pitch which is the predicted load demand of one day ahead. The proposed model improves the forecasting accuracy both in summer and winter.

Application of hybrid computational intelligence models in short-term bus load forecasting