A Multiobjective Deep Learning Solution for Optimizing Cooling Rates of Urban Courtyard Blocks in Hot Arid Zones

The optimization of courtyard blocks in high temperature regions presents a complex undertaking that involves balancing multiple competing design objectives. Recent research has shown the effectiveness of deep learning techniques in enhancing design goals, heating performance, and energy efficacy in urban courtyard blocks. These proposed frameworks based on deep learning have exhibited promising outcomes, holding the potential to transform urban planning practices in hot and arid zones. Nonetheless, further investigation is necessary to assess the reliability and scalability of these methodologies in the real-world.

Before searching for a comprehensive assessment of the existing literature related to city courtyard cooling, it's essential to set up a clean knowledge of key principles in this area.

1) Thermal Comfort: The concept of thermal comfort is fundamental to our discussion. It characterizes the psychological state of individuals about their satisfaction or dissatisfaction with the thermal environment in which they are situated. Factors contributing to thermal comfort include temperature, humidity, wind speed, and the interaction between these variables [8]. Ensuring thermal comfort is very important as it directly affects the overall well-being and productivity of the inhabitants of urban spaces.

2) Sustainable Design: Sustainable urban courtyard design aims to create environments that strike a balance between human habitation and environmental conservation. Its main objectives are to minimize the consumption of resources, reduce the ecological footprint, and raise the overall quality of life for the urban population. Sustainable design involves a multifaceted strategy, including optimizing energy efficiency, minimizing emissions, and promoting environmentally friendly practices [9]. The ultimate goal is to create urban courtyards that are environmentally responsible, economically viable, and socially rich.

3) Energy Efficiency: Energy efficiency constitutes a central tenet within the realm of urban courtyard design. This theory revolves around efforts to reduce energy consumption without compromising the functionality or performance of designed spaces. In the context of urban courtyards, achieving energy efficiency involves a careful analysis of building guidelines, material selection, and spatial layout. By reducing heating and cooling loads through judicious design decisions, energy-efficient urban courtyards can significantly contribute to sustainability goals, reduce urban heat island effects, and reduce pressure on energy resources [10].

The interconnectivity of these important key concepts is shown in Figure 2.

2.png

Figure 2. Interconnectivity of key concepts

Guedouh and Zemmouri [11] investigate the courtyard's impact on building morphology in the thermal and dazzling surroundings in hot and arid regions. They stress the importance of design aspects in increasing energy efficiency and occupant comfort through their study. Their findings show the potential of courtyard buildings in creating beneficial microclimates in arid environments, and they suggest that further improvements in the study can help to have accurate results.

Xu et al. [12] also use genetic deep learning algorithms to optimize the urban layout for improved microclimate performance in a cold region of China. The research considers optimization parameters that include courtyard design variables, building location, simulation settings, mechanical system, and thermal characteristics. They were able to optimize the microclimatic conditions by up to 80 percent. This happens because the optimized design creates more wind corridors and reduces the amount of heat that is trapped in the urban area.

Hasehzadeh Haseh, Khakzand, and Ojaghlou [13] investigate the best thermal characteristics of courtyards in the hot and desert climates of Isfahan. The authors investigate the effect of the courtyard design features on thermal performance, considering parameters such as shading, ventilation, and surface materials. The study provides valuable insights for architects and urban planners to enhance the heat comfort of courtyard spaces in arid regions. In a recent study, the authors M’Saouri El Bat et al. [14] propose a methodology to optimize the form of urban courtyards by combinaning outdoor zonal approach and building energy modeling. Their study demonstrates the possibility for combining multiple computational methodologies to enhance the efficacy of energy and thermal performance comfort of urban courtyard designs. The findings help to design sustainable urban planning techniques in arid regions, but this model doesn’t give us the exact magnitude of cooling rates or loads required in a courtyard if designed.

Khan et al. [15] focuses on the optimization of courtyard house design for passive cooling in hot, dry regions. Their study investigates the impact of design parameters, such as courtyard size, shape, and orientation, on cooling performance. By employing computer simulations and performance metrics, they offer design guidelines for achieving energy-efficient and comfortable living environments in arid climates. Ibrahim et al. [16] address the multi-objective optimization of city courtyard blocks in hot arid zones. The authors propose an optimization framework taking into account aspects such as thermal comfort, daylighting, and energy consumption. By further research and employing simulation techniques and genetic algorithms, they provide insights into designing sustainable and adaptive courtyard blocks in arid regions.

The papers examined within this research explore the use of optimization methods aimed at enhancing the thermal comfort of urban courtyards in diverse climatic contexts. These queries use a variety of optimization algorithms, including genetic algorithms, multi-objective evolutionary algorithms, and artificial neural networks. In addition, a range of datasets, including measurements, simulation-generated data, and historical records, serve as the foundation for these investigations. Findings from these studies underscore the effectiveness of optimization techniques for improving the thermal comfort experienced within urban courtyards.

4.1 Dataset description

The dataset has been sourced via the University of Bath, United Kingdom [17]. This dataset contains several observations of hot and desert zone urban courtyard blocks. The columns of data indicate the geometric features of the blocks: elongation, distances, and orientation. The height of the block is represented by the following nine columns: southwest, southeast, west, east, northwest, northeast, south, north, and the average height. The last column reflects the blocks' cooling per square meter (cooling/m²). The size of this dataset is substantially large, containing about 708 rows and 13 columns, respectively. The dataset can be used to optimize multi-objective urban courtyard blocks for hot and arid zones to predict cooling per square meter while keeping the required FAR (Floor Aspect Ratio) and geometric and height aspects of the blocks. The dataset contains a variety of values for various geometric and height features that can be used to construct and optimize urban courtyard blocks in hot and arid climates.

Table 1 below displays various building design characteristics, in a hot arid zone, including elongation, distances, orientation, height, FAR, and cooling load per unit area.

Table 1. Design parameters with description

4.2 Proposed work

The aim of our methodology is to predict the cooling rates/m² with the help of other variables as specified in the dataset description. To achieve this, we are using three algorithms, namely, RBFN (Radial Basis Function Network (RBFN), MLP Regressor, and RNN LSTM (Long Short Term Memory).

These algorithms are used in contrast to other prominent Deep Learning algorithms like CNN (Convolutional Neural Network) and ANN (Artificial Neural Network) because the algorithms we choose work efficiently on the type of data we are using. Additionally, the performance of these algorithms was relatively better. The entire code was run on Google Colab. Figure 3 shows the basic diagram to depict the workflow of our methodology.

3.png

Figure 3. Proposed methodology

4.2.1 Algorithm 1: MLP Regressor

MLP Regressor is a computational model based on neural networks and backpropagation. This is applied for training purposes. This methodology is designed to address regression tasks using supervised learning techniques [18].

The following Table 2 represents the libraries that are imported for the successful implementation of algorithms:

Table 2. Libraries used for MLPRegressor

The proposed strategy involves a series of sequential steps:

1. Data Preprocessing: The dataset is loaded. Feature scaling is applied to normalize the input features. The dataset is divided into training and validation subsets (80% training and 20% validation).

2. Model Architecture: An MLP (Multi-Layer Perceptron) regressor model is constructed. The architecture consists of two hidden layers with 50 and 25 units, respectively, followed by a ReLU activation function.

3. Training: The MLP model is trained with a maximum of 1000 iterations. The random state is set to 42 for reproducibility.

4. Evaluation: The model's performance is evaluated on the validation set. Evaluation metrics include MSE, RMSE, and MAE.

5. Prediction: The trained MLP model is used to make predictions on new data. Table 3 represents the parameters used in MLP Regressor Algorithm.

Table 3. Parameters used for MLPRegressor

4.2.2 Algorithm 2: RNN Long Short Term Memory

LSTMs are RNNs that can learn long-term dependencies by selectively retaining or discarding input via a gating mechanism [19].

The following Table 4 shows the libraries that are used in this algorithm for successful implementation:

Table 4. Libraries used for RNN LSTM

The following phases are included in the proposed methodology for this architecture:

1. Data Preprocessing: The dataset is loaded, consisting of features and the target variable, cooling/m². Feature scaling is performed to normalize the input features. The dataset is split into training and validation sets (80% training and 20% validation).

2. Model Architecture: An RNN LSTM (Recurrent Neural Network with Long Short-Term Memory) model is constructed. The architecture comprises a single LSTM layer with 50 units, followed by a dense layer with ReLU activation. The model uses the Adam optimizer to minimize the loss function.

3. Training: The model is being trained for 50 epochs with a batch size of 32. During training, the Mean Squared Error (MSE) is used as the loss function to minimize the prediction error.

4. Evaluation: The model's performance is assessed on the validation set. Evaluation metrics include MSE, Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE).

5. Prediction: The trained model is used to make predictions on new data. Parameters used for RNN LSTM algorithm are shown in Table 5.

Table 5. Parameters used for RNN LSTM

4.2.3 Algorithm 3: RBFN

The Radial Basis Function Network (RBFN) is another computational model used for regression tasks based on neural networks. It is distinct from the more common feedforward neural networks (such as multilayer perceptrons) in its architecture and activation functions. RBFNs are particularly useful when dealing with problems that involve nonlinear relationships in the data [20]. The libraries included in this model are shown in the Table 6:

Table 6. Libraries used for RBFN

The following steps were implemented here:

1. Data Preprocessing: The same dataset is used for the RNN LSTM and MLP model. Feature scaling is applied to normalize the input features. The dataset is divided into training and validation subsets (80% training 20% validation).

2. Model Architecture: An RBFN model is constructed. The architecture consists of Radial Basis Functions (RBF) as activation functions. The number of RBF neurons is determined based on the dataset or prior knowledge.

3.Training: The RBFN model is being trained. The center points and widths of the RBF neurons are typically learned during training. The training process aims to minimize the error between the RBFN model's predictions and the actual target values.

4. Evaluation: The model's performance is evaluated on the validation set. Evaluation metrics include Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). These metrics help assess the accuracy and performance of the RBFN model.  Parameters used for this algorithm are shown in an effective manner in Table 7.

5. Prediction: The trained RBFN model makes predictions on new or unseen data. The model applies the learned RBF activations to the input data to produce predictions.

Table 7. Parameters used for RBFN

Finally, deep learning techniques are used to optimize urban courtyard blocks in hot and arid climates to forecast cooling rates in these blocks to ensure their long-term viability. For this objective, neural network models such as RNN-LSTM, MLP Regressor, and RBFN are used. RBFN outperformed MLP Regressors and RNN LSTM in forecasting cooling rates in urban courtyard blocks, according to the findings. This is critical because accurate forecasting is required to maintain the long-term viability of these blocks in hot and arid areas. As a result, RBFN is a more effective choice for predicting cooling rates in such urban courtyard blocks. When we say these algorithms perform well, it is also important to understand that these algorithms hold certain limitations, including data constraints, the need for expertise, and the unavailability of large datasets for better predictions.

Further, we could continue the study by optimizing models that not only give us the cooling rates but also predict thermal comfort and thermal cooling levels accurately. Currently, the study is focused on hot arid zones using a “Cairo, Egypt” dataset; in the future, we can use this technique and modify parameter settings to fit into different temperature environments and use it to predict cooling and thermal rates on various datasets from diverse cities.

Designers, Architects, Urban city planners, and Environmentalists can all use this study to develop software/models that easily let them enter parameters such as area and size specifications of a plot and building and estimate the cooling rate, cooling load, and thermal comfort according to different specifications. They can use the results to make design changes to optimize load results to decrease the energy consumption of courtyard blocks and buildings, by which a greener, sustainable way of urban development can be achieved. Policy makers can also use it to regulate new design plans. Anticipating cooling loads based on building features encourages architects and builders to design and construct buildings that are better adapted to the local climate conditions. Sustainable building design promotes the use of passive cooling strategies, such as optimal orientation, shading, and insulation, which can reduce the reliance on energy-intensive cooling systems.

As climate change continues to bring rising temperatures and more frequent heat waves, the ability to accurately anticipate and manage cooling loads becomes critical for building resilience. Buildings designed to handle extreme heat efficiently are better equipped to provide a comfortable indoor environment during heatwaves, thus enhancing the resilience of communities to climate change impacts. Sustainable building practices and energy-efficient technologies can foster local economic development. The research can encourage the adoption of green building standards and the growth of industries related to energy-efficient construction and retrofitting. This aligns with the goal of conserving energy resources and reducing greenhouse gas emissions associated with energy production.

Overall, this research highlights the importance of using advanced machine learning techniques, such as neural networks, to optimize urban courtyard blocks in hot, arid environments. It underscores the significance of precise cooling rate predictions in ensuring the long-term viability of these blocks. In the future, these insights can inform the development of more sustainable urban design solutions in hot and arid zones. It aligns with global goals related to sustainable development and climate change adaptation, contributing to a more sustainable and resilient built environment.