Deep Learning Based Prediction Framework For Univariate Time-Series Smart Home Energy Consumption
The Smart Home System (SHS) concept emerged due to the advent of the Internet of Things (IoT). IoT sensors embedded in electronic devices or equipment enable devices to interact with their surrounding environment. Sensor-enabled IoT devices generate data at prescribed intervals and transmit it to re...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2021
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The Smart Home System (SHS) concept emerged due to the advent of the Internet of Things (IoT). IoT sensors embedded in electronic devices or equipment enable devices to interact with their surrounding environment. Sensor-enabled IoT devices generate data at prescribed intervals and transmit it to remote servers, where several platform applications represent it to the end-users. The transmission of IoT data is on wireless medium; therefore, the sensors-derived data often suffer from missing or incomplete data quality problems due to connection errors, sensor faults, and security attacks. This research addresses the problem of missing values in the IoT based SHS electricity consumption univariate time-series data source. The data sources of SHS appliances contains missing values large gaps that caused performance and credibility issues. The research developed a deep learning-based prediction framework for univariate time-series datasets of SHS electricity consumption to forecast large gaps of missing value. The framework utilised a hybrid Convolutional Neural Network-Long Short Term Memory (CNN-LSTM) neural network to forecast missing values large gaps from given input univariate time-series dataset of electricity consumption. This research examined a variety of statistical, machine learning, and deep learning univariate time-series forecasting models to find a single highperforming forecasting model for the prediction framework. Furthermore, the research observed the high performance in deep neural network (DNN) models for one-step and multistep forecasting in a comparative experiment. The performance of the DNNs was evaluated using three criteria: predictive accuracy, training loss, and training time. According to the experimental results, the CNN and Convolutional LSTM (ConvLSTM) outperformed the other models in all three performance measurement metrics. |
|---|