A carregar...
8 resultados
Resultados da pesquisa
A mostrar 1 - 8 de 8
- Assessing the relevance of load profiling information in electrical load forecasting based on neural network modelsPublication . Sousa, J.C.; Neves, L.P.; H.M. JorgeThe article is focused on evaluating the relevance of load profiling information in electrical load forecasting, using neural networks as the forecasting methodology. Different models, with and without load profiling information, were tested and compared, and, the importance of the different inputs was investigated, using the concept of partial derivatives to understand the relevance of including this type of data in the input space. The paper presents a model for the day ahead load profile prediction for an area with many consumers. The results were analyzed with a simulated load diagram (to illustrate a distribution feeder) and also with a specific output of a 60/15 kV real distribution substation that feeds a small town. The adopted methodology was successfully implemented and resulted in reducing the mean absolute percentage error between 0.5% and 16%, depending on the nature of the concurrent methodology used and the forecasted day, with a major benefit regarding the treatment of special days (holidays). The results illustrate an interesting potential for the use of the load profiling information in forecasting.
- Short-term load forecasting using information obtained from low voltage load profilesPublication . Sousa, João; Pires Neves, Luís; Humberto M.M. JorgeRecent researches in load forecasting are quite often based on the use of neural networks in order to predict a specific variable (maximum demand, active electric power or hourly consumption) using past values of the same variable and other exogenous factors proved to influence the value being predicted. This work aims to explore different input patterns in neural networks incorporating information derived from load profiles of different consumers' classes.
- Load forecasting based on neural networks and load profilingPublication . Sousa, João; Pires Neves, Luís; H. M. JorgeThis work presents a novel perspective of load forecasting based on neural networks and load profiling. In addition to the variables that are typically used to predict future load demand, such as past load values, meteorological variables, seasonal effects or macroeconomic indexes, it is expected that the use of load profiles and detailed information of individual consumers could favor the forecasting process. The methodology can be extended to different temporal horizons being predicted and the eventual threat of overparametrization is attenuated by the use of neural networks since the complexity of the model does not necessarily depends on the number of its weights and biases, as some of these parameters might be found irrelevant in the process. Another way to reduce the risk of overparametrization and overfitting is through the use of a considerable number of data points (whenever historical data is available) to train the network.
- Forecasting Heating and Cooling Energy Demand in an Office Building using Machine Learning MethodsPublication . Godinho, Xavier; Bernardo, Hermano; Oliveira, Filipe Tadeu; Sousa, JoãoForecasting heating and cooling energy demand in buildings plays a critical role in supporting building management and operation. Thus, analysing the energy consumption pattern of a building could help in the design of potential energy savings and also in operation fault detection, while contributing to provide proper indoor environmental conditions to the building's occupants.This paper aims at presenting the main results of a study consisting in forecasting the hourly heating and cooling demand of an office building located in Lisbon, Portugal, using machine learning models and analysing the influence of exogenous variables on those predictions. In order to forecast the heating and cooling demand of the considered building, some traditional models, such as linear and polynomial regression, were considered, as well as artificial neural networks and support vector regression, oriented to machine learning. The input parameters considered in the development of those models were the hourly heating and cooling energy historical records, the occupancy, solar gains through glazing and the outside dry-bulb temperature.The models developed were validated using the mean absolute error (MAE) and the root mean squared error (RMSE), used to compare the values obtained from machine learning models with data obtained through a building energy simulation performed on an adequately calibrated model.The proposed exploratory analysis is integrated in a research project focused on applying machine learning methodologies to support energy forecasting in buildings. Hence, the research line proposed in this article corresponds to a preliminary project task associated with feature selection/extraction and evaluation of potential use of machine learning methods.
- Data Acquisition and Monitoring System for Legacy Injection MachinesPublication . Silva, Bruno; Sousa, João; Alenya, GuillemNowadays, companies must embrace the concept of Digitalization and Industry 4.0 to remain competitive in the market. The reality is that most of them do not have their industrial devices prepared to access their data on a real-time basis. As most companies do not have the possibility to renew all their legacy devices and because these devices are still very productive, a retrofit solution is of high interest. In this work, we propose an affordable procedure that allows data collection and monitoring of older injection machines, as a contribution towards legacy devices integration. The developed system neither requires additional proprietary modules, nor contractual annual fees for different devices, sharing the same interface across different machine manufacturers and also contributing to uniform data collection. Evaluation was carried out in a real shop floor, monitoring the injection parameters for different machine models, validating the effectiveness of the developed system.
- A Data-Driven Approach to Forecasting Heating and Cooling Energy Demand in an Office Building as an Alternative to Multi-Zone Dynamic SimulationPublication . Godinho, Xavier; Bernardo, Hermano; Sousa, João C. de; Oliveira, Filipe T.Nowadays, as more data is now available from an increasing number of installed sensors, load forecasting applied to buildings is being increasingly explored. The amount and quality of resulting information can provide inputs for smarter decisions when managing and operating office buildings. In this article, the authors use two data-driven methods (artificial neural networks and support vector machines) to predict the heating and cooling energy demand in an office building located in Lisbon, Portugal. In the present case-study, these methods prove to be an accurate and appealing alternative to the use of accurate but time-consuming multi-zone dynamic simulation tools, which strongly depend on several parameters to be inserted and user expertise to calibrate the model. Artificial neural networks and support vector machines were developed and parametrized using historical data and different sets of exogenous variables to encounter the best performance combinations for both the heating and cooling periods of a year. In the case of support vector regression, a variation introduced simulated annealing to guide the search for different combinations of hyperparameters. After a feature selection stage for each individual method, the results for the different methods were compared, based on error metrics and distributions. The outputs of the study include the most suitable methodology for each season, and also the features (historical load records, but also exogenous features such as outdoor temperature, relative humidity or occupancy profile) that led to the most accurate models. Results clearly show there is a potential for faster, yet accurate machine-learning based forecasting methods to replace well-established, very accurate but time-consuming multi-zone dynamic simulation tools to forecast building energy consumption.
- Machine Learning Methods for Quality Prediction in Thermoplastics Injection MoldingPublication . Silva, Bruno; Sousa, João; Alenya, GuillemNowadays, competitiveness is a reality in all industrial fields and the plastic injection industry is not an exception. Due to the complex intrinsic changes that the parameters undergo during the injection process, it is essential to monitor the parameters that influence the quality of the final part to guarantee a superior quality of service provided to customers. Quality requirements impose the development of intelligent systems capable to detect defects in the produced parts. This article presents a first step towards building an intelligent system for classifying the quality of produced parts. The basic approach of this work is machine learning methods (Artificial Neural Networks and Support Vector Machines) and techniques that combine the two previous approaches (ensemble method). These are trained as classifiers to detect conformity or even defect types in parts. The data analyzed were collected at a plastic injection company in Portugal. The results show that these techniques are capable of incorporating the non-linear relationships between the process variables, which allows for a good accuracy (≈99%) in the identification of defects. Although these techniques present good accuracy, we show that taking into account the history of the last cycles and the use of combined techniques improves even further the performance. The approach presented in this article has a number of potential advantages for online predicting of parts quality in injection molding processes.
- End-to-End Management System Framework for Smart Public BuildingsPublication . Jesus, Ivo; Pereira, Tomás; Marques, Pedro; Sousa, João; Perdigoto, Luís; Coelho, PauloThis paper presents a project aiming to design a complete framework to measure energy (electricity and gas) and water consumptions in a local Parish Council building and an adjacent Sports Hall located in the central part of Portugal. The goal is an end-to-end solution, from data acquisition to data analysis. Besides acquiring and storing the data, the aim is to make this information available and valuable to enhance better decisions in building management actions, to enable detection of situations of anomalous consumption and also to promote building users' awareness. To pursue this goal, PLCnext technology solutions from Phoenix Contact are adopted. The system is based on a new generation industrial controller that communicates with energy and water meters distributed throughout the building using a standard Information Technology (IT) network. The solution explores Industry 4.0 concept, such as Cloud Data Management, Cybersecurity, and Machine Learning. With historic consumption records available, Machine Learning strategies are being used to predict load profiles in a short-term horizon and also planned to classify untypical consumption behaviors (for monitor and alarm purposes). This project is being deployed in partnership between Polytechnic of Leiria, EduNet International Education Network and involving the local Parish Council, owner of the monitored buildings.