Over 33% of final energy consumption is used in buildings which leads to nearly 40% of total direct and indirect CO2 emissions in the world. While energy consumption is steadily rising globally, managing building energy utilization by on-site renewable energy generation can help responding this demand. This paper proposes a deep learning method based on a discrete wavelet transformation and long short-term memory method (DWT-LSTM) and a scheduling framework for the integrated modelling and management of energy demand and supply for buildings. This method considers several factors including a two-step electricity price, uncertainty in climatic factors, availability of renewable energy resources (wind and solar), energy consumption patterns in buildings, and the non-linear relationships between these parameters. This novel method analyzes and continuously learns from data patterns based on hourly, daily, weekly and monthly intervals. The method enables monitoring and controlling renewable energy generation, the share of energy imports from the grid, employment of saving strategy based on the user priority list, and energy storage management. The main objective of the proposed method is to minimize the reliance on the grid and electricity cost, especially during the peak days. The results demonstrate that the proposed method can forecast building energy demand and energy supply with a high level of accuracy, showing a 3.63- 8.57% error range in hourly data prediction for one month ahead. The results also show that the method can supply 304 days (83.2%) of a year without reliance on energy grids, decreasing 87.2% in energy demand on one hand and exporting annually 7777 kWh to the grid on the other hand. In addition, the rescheduling framework decreased the imported electricity cost with the higher electricity tariff by 98 %. The combination of the deep learning forecasting, energy storage, and scheduling algorithm enables reducing annual energy import from the grid from 6709 to 858 kWh (84.3%).

This paper presents the development of air quality low-cost sensors (LCS) with improved accuracy features. The LCS features integrate machine learning based calibration models and virtual sensors. LCS performances are analyzed and some LCS variables with low performance are improved through intelligent field-calibrations. Meteorological variables are calibrated using linear dynamic models. While, due to the non-linear relationship to reference instruments, fine particulate matter (PM2.5) are calibrated using non-linear machine learning models. However, due to sensor drifts or faults, carbon dioxide (CO2) does not present correlation to reference instrument. As a result, the LCS for CO2 is not feasible to be calibrated. Hence, to estimate the CO2 concentration, mathematical models are developed to be integrated in the calibrated LCS, known as a virtual sensor. In addition, another virtual sensor is developed to demonstrate the capability of estimating air pollutant concentrations, e.g. black carbon, when the physical sensor devices are not available. In our paper, calibration models and virtual sensors are established using corresponding reference instruments that are installed on two reference stations. This strategy generalizes the models of calibration and virtual sensing which then allows LCS to be deployed in field independently with a high accuracy. Our proposed methodology enables scaling-up accurate air pollution mapping appropriate for smart cities.

The time-series forecasting is a vital area that motivates continuous investigate areas of intrigued for different applications. A critical step for the time-series forecasting is the right determination of the number of past observations (lags). This paper investigates the forecasting accuracy based on the selection of an appropriate time-lag value by applying a comparative study between three methods. These methods include a statistical approach using auto correlation function, a well-known machine learning technique namely Long Short-Term Memory (LSTM) along with a heuristic algorithm to optimize the choosing of time-lag value, and a parallel implementation of LSTM that dynamically choose the best prediction based on the optimal time-lag value. The methods were applied to an experimental data set, which consists of five meteorological parameters and aerosol particle number concentration. The performance metrics were: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and R-squared. The investigation demonstrated that the proposed LSTM model with heuristic algorithm is the superior method in identifying the best time-lag value.