14961

13465

كارشناسي ارشد

معماري كامپيوتر

اصفهان : دانشگاه صنعتي اصفهان

1398

سيزده، 72ص. مصور، جدول، نمودار

مجيد نبي، مهران صفاياني

مسعودرضا هاشمي، عبدالرضا ميرزايي

1398/06/02

كتابنامه

مهندسي كامپيوتر

مهندسي برق و كامپيوتر

1398/06/03

2553310

Energy consumption prediction in presence of missed sensor data Yazdan Ghaderi Sanjabi y ghaderi@ec iut ac ir June 2019 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Prof Majid Nabi nabi@cc iut ac ir Second Supervisor Prof Mehran Safayani safayani@cc iut ac ir Abstract Today electrical energy has become one of the basic human needs For everyday use such as the use of home grownequipment industrial equipment lighting heating and ventilation and much more electrical energy is needed ForecastingElectricity Consumption is important because of that energy supplier and energy consumer systems should be managed inefficient way Electricity supply companies need short term electricity consumption to carry out daily operations fulfillmentof commitments and planning of the transfer of electricity from a power company to commercial office and residential units Therefore a precise prediction of short term forecast electricity consumption is vital for an electricity supply company Byexpanding the machine learning algorithms and proving their effectiveness the tendency to use these algorithms in differentdomains In the short term prediction of electrical energy consumption the inputs are the amount of sensors recorded atdifferent places in a home Therefore any sensor may be interrupted and This interrupt will cause no amount to be recordedat that moment This creates an incomplete set of data In most of the works on short term energy consumption prediction the use of an incomplete data set for energy prediction has not been raised In this dissertation we provide several modelsthat can predict the amount of energy consumed in the next 10 minutes even if there are a set of data that does not contain anumber of data these models are in fact a hybrid model based on multi task Gaussian processes and various predictors suchas multi layer perceptron networks recurrent neural networks and convolutional neural networks with the aim of short termprediction of electric energy consumption The approach is in fact a supervised approach and its learning is done usingthe labeled data for a residential home Finally MGP lstm model which is one of the three proposed models shows betterresults than other models For example by assuming 50 missed data MGP lstm model could improve the accuracy of theresults compared to the lstm model in the RMSE and MAE criteria by 14 3 and 27 5 respectively Key Words Time series Prediction Short term Electrical Energy Consumption Forcasting Multi task Gaussian Process Neural Network Ariam Process Incomplete Dataset

مجيد نبي، مهران صفاياني

مسعودرضا هاشمي، عبدالرضا ميرزايي