1.1. Walkability modeling & agent-based modeling and simulation#

Walkability modeling (WM) involves simulating pedestrian movement in constrained environments, which makes it a valuable tool for understanding real-world situations and enhancing our knowledge of pedestrian behavior.¹⁵ This method allows for a detailed analysis of programs and patterns, suggesting improvements to existing structures for greater efficiency.^19,20 In the context of urban development, WM primarily focuses on assessing how human behavior impacts physical activity, providing insights that are crucial for urban planning and design.²¹

On the other hand, agent-based modeling is used to simulate the complex nature of walkability and its relationships.^22,23,24 Agent-based modeling and simulation (ABMS) provides a deep understanding of existing walking habits that directly affect physical activity in urban contexts.^15,25 By examining agent trails generated by ABMS, we can see how pedestrian pathways (Routes) and city features (Nodes) are interconnected, promoting livability and physical activity while also preparing for future mitigation efforts. This modeling method is particularly essential in developing countries, where factors influencing physical activity need to be considered by city stakeholders.²⁶

Integrating ABMS and WM offers significant benefits, including the ability to observe and forecast phenomena.²⁷ Computational methods within ABMS provide advantages such as dynamic feedback mechanisms and realistic behavior modeling.¹⁴ However, challenges arise when translating data, including behavior and ethnographic input,²⁸ into the ABMS environment.^24,29 Nonetheless, ABMS in WM allows for forecasting in complex scenarios, deepening our understanding of factors influencing human mobility in urban settings.^30,31

Recent academic attention has focused on integrating ABMS and WM, particularly in transit-oriented development (TOD), and identifying potential “Nodes” or specific public spaces as key feature.^32,33 Notable studies in this area include Badland et al. (2013), who introduced a web-based tool for evaluating and enhancing neighborhood walkability, emphasizing improvements like pedestrian simulations and considering factors such as residential density and traffic flow.³⁴ Herrmann et al. (2017) explored the impact of walkability on walking frequency in neighborhoods, advocating for interventions like reducing parking areas and increasing street-level tree coverage.³⁵ Lu (2017) conducted a comprehensive study on walkability modeling, using GIS methods like Map Overlay and Cost Distance Analysis to identify strategies for promoting walking across various dimensions.³⁶ Kanchik (2017) examined the health impacts of enhanced walkability modeling, revealing connections between architectural elements, perceived safety, physical activity, and overall well-being.³⁷ Additionally, Deng and Zhang (2018) emphasized their research on neighborhood walkability models, providing urban planners with standardized benchmarks derived from a walking speed index and walkability maps,³⁸ thus facilitating urban improvement strategies.³⁹

The studies mentioned above aim to improve neighborhood walkability through pedestrian simulations, benchmarking, and promoting walking to understand its impact on factors like walking frequency, physical activity, safety, and well-being, ultimately enhancing residents’ quality of life in cities. Although these studies do not specifically simulate individual or group movements iteratively in ABMS, they lay the groundwork for future research, including our ongoing research, which aims to enhance accessibility and connectivity. Additionally, they provide a theoretical foundation for understanding how ABMS can be used to parameterize route simulations considering multiple criteria such as greenness, distance to transit, safety perceptions, physical activities, health, and destination accessibility. These studies uncover the complex factors that impact walking behaviors, neighborhood walkability, and the interconnectedness of elements like buildings, safety perceptions, physical activities, and health.

Furthermore, prior research has highlighted the potential of integrating ABMS into upcoming studies to explore a shared vision.^40,41,42 This approach leads to a deeper understanding and supports data-driven decisions as reliable problem-solving methods,⁴² which can ultimately enhance walkability and promote healthier living environments.³⁴ There is also a growing interest in ABMS within architectural research due to its ability to replicate complex structures and behavioral dynamics.^41,43 However, current ABMS applications mainly focus on understanding agent capabilities and behaviors, analyzing interaction patterns, and simplifying model representations.⁴⁴ As a result, in our study, we will further enhance the ABMS method, especially by using ABM to model walkability and optimize movement in urban areas through a novel method that extensively utilizes pedestrian pathways (Routes) along with urban architecture, public spaces, and green spaces (Nodes). This will uncover new facets of walkability modeling within transit-oriented movement.

1.2. Research gap#

To fill the existing research gap in WM, a new approach is essential. This approach involves analyzing the behavioral patterns of individuals and relevant urban features, translating them into computational data, and establishing customized parameters that align with walkability principles, creating livable cities, promoting health, and integrating transportation systems like TOD.^{15,19,20,21,32}

Utilizing crowdsourcing platforms such as OpenStreetMap (OSM) and Google Maps Typical Traffic.^45,46,47,48 in conjunction with field walkability observations for ABMS, is crucial for understanding the unique characteristics of each city.^{12,49,50,51,52,53} By incorporating quantitative data and algorithms, we can enhance the simplified route trails generated by ABMS^15,54,55 preparing them for optimization using a multi-objective approach to effectively tackle urban walkability challenges.⁵⁶ To ensure practical applicability, clustering methods are then applied to organize various approaches, algorithms, methods, and related elements iteratively, based on theoretical frameworks and practical contexts within WM. This clustering process aims to define Nodes and Routes effectively when implemented contextually.

The approach mentioned above is termed the dynamic multi-layer (DML) method, which integrates algorithms such as Laplacian smoothing for trail refinement,^57,58 A-star for finding the shortest path,^59,60,61 NSGA-2 for evolutionary multi-objective optimization (EMOO)^62,63 and iterative clustering to refine trails and determine optimal solutions into a single, optimal solution. The Method section will provide detailed explanations of this DML approach.

In summary, implementing this layered strategy in DML within WM can address current limitations, optimize Nodes and Routes for pedestrian-centric urban design, and enhance connectivity in urban communities.

The paper is structured as follows: The Case study section introduces the context in which the DML method is applied to address urban walkability challenges. The Method section outlines the formulation and explanation of the DML method, including dependencies between each layer parameter. The Discussions section examines the key findings of the DML method and its practical implications. The Limitations section identifies potential constraints and opportunities within the study. Finally, the Conclusion section concludes and emphasizes the significance of the research, highlighting the importance of the DML method and providing implementation recommendations to relevant stakeholders.

2.1. Urban mobility challenges in Kalideres, Jakarta#

Jakarta, Indonesia’s capital, grapples with significant urban walkability challenges. Specifically, the city suffers from a lack of pedestrian-friendly infrastructure, exemplified by poorly maintained footbridges that force pedestrians into risky behaviors like jaywalking.⁶⁴ A study conducted by Leather et al. at the Asian Development Bank (ADB) using the Global Walkability Index (GWI) revealed low ratings across residential, educational, commercial, and public transport zones, underscoring Jakarta’s substantial deficiencies compared to other Asian cities.^65,66

In the context of Kalideres, located in West Jakarta, there is a notable transit volume compared to other regions in Jakarta, as depicted in Figure 1. The area also contends with a dense population, inadequate law enforcement, and a disjointed urban transport system between the Kalideres Terminal and Station.⁶⁸ Consequently, Kalideres experiences frequent traffic congestion, prolonged vehicle idling, and the presence of unlicensed informal vendors.⁶⁹ As of June 2022, Kalideres has been officially designated for development into a major transportation hub,⁷⁰ aiming to establish seamless connectivity to Tangerang City in the west and serve as a crucial transit point for national routes, including those leading to destinations like Sumatra Island and Bali Island via the Indonesian National Route 1, a segment of the Asian Highway 2 (AH2).⁷¹ This development holds potential benefits for achieving the research objectives.

Figure 2 provides a satellite image showcasing Kalideres, highlighting the density of residential areas (including informal and landed housing), commercial zones (ranging from micro to major enterprises), wetlands (such as paddy fields), and transportation systems (including paratransit, traditional, and rapid transit), as well as industrial zones (ranging from small to large-scale industries) overflowing in this area. Additionally, the central terminal’s location underscores the necessity of a proper transit system to accommodate the rapidly growing population.³²

From a connectivity standpoint, arterial roads face congestion due to the mix of pedestrians, private vehicles, and public transport, leading to bottlenecks. Figure 3 illustrates that pedestrian movement is restricted by a car-oriented system implemented in this area, resulting in longer travel distances and durations. This contradicts the principles of walkability advocated by TOD and the sustainable city concept.^32,74,75 Therefore, the fundamental need for spacious sidewalks to enhance pedestrian and cyclist movements becomes evident.

2.2. Physical activities and green spaces challenges#

Emergent challenges related to physical activity and green spaces are also evident in Jakarta.^76,77 For instance, 65% of pedestrian overpasses remain underutilized.⁶⁵ Issues such as jaywalking and inadequate urban features at locations like Kalideres Terminal restrict physical activity opportunities and contribute to less stimulating environments, potentially leading to stress and risky behaviors.⁷⁸ To address these challenges, it is recommended to integrate visual attractions along pedestrian pathways (Routes) to enhance the overall experience and promote healthier behaviors.⁷⁸ Additionally, implementing TOD principles in areas like Kalideres and prioritizing urban green open spaces (UGOS) can enhance attractiveness and encourage pedestrian activity.

Aligning WM simulations with TOD principles supports the development of UGOS through modeling, indirectly improving transportation efficiency by promoting human-oriented movement and enhancing public health and well-being through increased green spaces.^79,80,81 Integrating UGOS with public facilities like commuter rest areas provides socio-economic benefits. As of December 2022, Jakarta had limited public space (PS) and only one transit hub under construction.⁸² Additionally, the UGOS footprint in Jakarta has decreased to 110.45 km² ⁸³. Therefore, strategically placing UGOS as public spaces at intervals of 400–800 meters based on TOD principles is necessary to ensure community liveliness and well-being.^73,84,85

These challenges highlight the importance of further exploring the DML method for WM development. Insights from WM can guide the placement of urban features, particularly public spaces acting as Nodes for rest areas. A comprehensive urban design approach focusing on pedestrian and cycling amenities aligned with TOD principles can help reduce pollution and traffic incidents in the city.⁸⁶ Analyzing junctions, interactions, and crowd dynamics within WM can uncover patterns to inform advanced modeling methods for addressing challenges at critical intersections.

3.1. Research framework#

The dynamic multi-layer (DML) method is conceptually formulated based on the theoretical framework and the preceding case study. This method employs interconnected dynamic strategies operating layer by layer, with dependencies between sub-layers as depicted in Figure 4. It’s important to note that while the methods outlined are partially contextualized within the selected experimental case study,^{64,68,69,83,87,88} adjustments may be necessary in different contexts.⁵³ Customization according to the project’s nature, context, or established environment will be highlighted in subsequent paragraphs when relevant.

The DML method provides a solution for cities aiming to plan and design based on best practices in urban planning and design principles. It integrates planning for UGOS and PS to promote healthy urban environments,^{11,13,80,81,85} focuses on transit hub (TH) placement according to walkability principles in TOD^32,67,73,86 and implements physically-oriented linkages between TH and PS with UGOS included for optimized accessibility within TOD radius principles.^73,75,85,89 In conclusion, this method cohesively integrates multiple approaches to orchestrate an optimal solution (design guideline).

Figure 4 illustrates the six-layer DML approach, with each layer managing distinct input-output data interactions. Subsequent layers build upon insights from previous ones, processing data using predetermined algorithms and parameters. Layers 1 to 4 prioritize walkability data generation, while Layer 5 focuses on simulation and optimization, and Layer 6 manages data clustering based on simulation outcomes. The following overview outlines how the DML addresses the walkability challenge through integrated TH and PS placements, with detailed explanations of each layer’s operation within the selected case study provided in subsequent sections.

1. Layer 1: Data gathering & modeling preparation, focuses on understanding the environment and human behavior through detailed data gathering. It establishes a unique context profile and characteristics (UCPC) by considering parameters such as typical traffic,⁴⁹ walkability, behavior, negative–positive, and solid–void.^49,90
2. Layer 2: Predicting the dynamic model of a complex system, builds upon Layer 1 and predicts human movement behavior within the environment using dynamic modeling techniques to customize the settings of attraction nodes, generating results referred to as human trails and potential composition (HTPC).^50,91
3. Layer 3: Simplifying the trails, aims to simplify the complex HTPC generated in Layer 2 for easier interpretation and analysis. It utilizes the Laplacian smoothing algorithm to refine the trails iteratively.^57,58
4. Layers 4 and 5: Defining and optimizing the objectives. Layer 4 sets clear goals for the optimization process in Layer 5 based on a clear understanding of the context and human behavior. It utilizes the A-star algorithm^59,61 for efficient pathfinding^92,93,94,95 and defines objective parameters related to people, attraction nodes, PS, spatial formation, and TOD principles. Layer 5 focuses on optimizing the objectives formulated in Layer 4 by minimizing values between refined parameters. It aims to identify solution candidates using the NSGA-2 algorithm.^62,63
5. Layer 6: Clustering & selecting the best solution, solution candidates undergo clustering and evaluation based on predetermined criteria from Layers 1 to 5. Solutions are considered using various approaches such as Machine-,⁹⁶ Algorithm-,⁹⁷ Collaborative-,⁹⁸ and Context-driven⁵³ methods, guiding the implementation of the most suitable solution for the selected context.⁹⁹

3.2. Layer 1: Data gathering & modeling preparation#

Layer 1 involved analyzing traffic congestion and pedestrian flow using field walkability observations (Figure 3). Digital tools like Rhino,¹⁰⁰ Grasshopper,¹⁰¹ and Caribou¹⁰² were utilized to map the environment and create a base map (Figure 5). Observing emergent behaviors, we used crowdsourced data from Google Maps Typical Traffic to track traffic congestion,^{47,48,103,104} influencing pedestrian movements collected from 6 a.m. to 10 p.m., as shown in Figure 6a.

Figure 6a displays Google Maps Typical Traffic data transformed into a ranking schema categorizing traffic densities into three levels: low, moderate, and high. Each route is assigned “Total points” and “Rank” per day per 1-hour timestep, allowing us to identify route segments affected by significant bottlenecks.⁴⁸ This enables ABMS representation of agents and targets in the environment. Concurrently, Figure 6b depicts:

• T1: Primary target for high-density areas, totaling 16 nodes.
• T2: Secondary target for moderately dense regions, totaling 22 nodes.
• AT: Transit-centric agent for arrival/departure epicenters, totaling 25 nodes.
• AR: Resident-centric agent for nodes from residential areas, totaling 49 nodes.

3.3. Layer 2: Predicting the dynamic model of a complex system#

Layer 2 employs the ABMS framework to predict complex movement patterns. This framework generates representation configurations based on behavior data collected during field walkability observations, creating unique agent configurations for AR, AT, T1, and T2 as informed by Layer 1 insights. Utilizing Grasshopper¹⁰¹ and the open-source framework Quelea,¹⁰⁵ the simulation demonstrates the iterative behavioral functionalities of each agent group, as outlined in Table 1. To ensure simulation consistency, Table 2 summarizes the global settings employed for the ABMS.

After 300 iterative steps per iteration, Figure 7 presents superimposed results that unveil patterns across iterations. In Iteration 1, agent navigation exhibits a smoother flow. By Iteration 2, agents cluster in bustling zones, and by Iteration 3, dense congregations emerge around areas such as T1 and T2. These dense areas, particularly those near high-traffic zones, become congestion hotspots due to their population density. The simulations also identify primary axes at road segment intersections, providing insights for urban renewal and infrastructure enhancements aligned with walkability principles.

3.4. Layer 3: Simplifying the trails#

Layer 3 abstracts data from three iterations in Layer 2, synthesized in Figure 7 and elaborated in Figure 8. The methodology utilizes the Laplacian smoothing technique,^57,58 originally developed by Field⁵⁵ and refined by Pryor and Zwierzycki,¹⁰⁶ focusing on average calculations of nearby nodes. Its algorithm includes:

1. Maintaining consistent spacing of 25m between nodes along curve trails.
2. Identifying uniformly segmented spaces as modifiable modal points.
3. Categorizing modal points based on their position: start, end, or intermediate nodes.
4. Iteratively adjusting intermediate nodes using a loop method based on their relative positions and average values,¹⁰⁷ simulating magnet-like behavior with the loop algorithm from Anemone.¹⁰⁸
5. Choosing the final outputs, known as smoothed curve trails, to preserve essential data while avoiding excessive refinement.

Figure 8 illustrates the algorithm procedure, particularly spotlighting the 10th iteration for its well-maintained balance between the original trail’s complexity and simplicity. In this stage, the primary trails begin to blend harmoniously with the environment, paving the way for further analysis in the subsequent layer. The next layer concentrates on defining mesh space through the shortest path algorithm (SPA), in line with TOD Standard 3.0⁷³ and Sustainable Development Goal 11,¹⁰⁹ highlighting urban design enhancement and resilient transportation infrastructure.

3.5. Layers 4 and 5: Defining and optimizing the objectives#

Layers 4 and 5 synergize to optimize solution candidates effectively. Layer 4, depicted in Figure 9, illustrates the evolution of simplified trails across six iterations. Its algorithmic steps include:

1. Smoothing the simple trails to rationalize geometries for faster calculations.
2. Offsetting all curves to create a bounding box for the outer boundary of trails.
3. Further simplifying the trails to ensure consistency and computational efficiency.
4. Utilizing simplified boundary lines to generate a triangulation mesh for graph preparation for the shortest path algorithm (SPA).
5. Transforming the triangulation mesh into a mesh of near-equilateral triangles using the TriRemesh component from Kangaroo2.¹¹⁰
6. Simulating the shortest path using the A-star algorithms for SPA, including graphs for the path, start and end points, and employing the shortest path component from Arachne¹¹¹ for simulation.

After iterative refinement, the baseline trails have the potential to evolve into optimal routes, enhancing walkability and aligning with TOD and Sustainability Development Goals principles. These optimized routes play a crucial role in improving physical activities and upgrading public transportation in densely populated areas. As mentioned earlier, the design objectives will focus on promoting the use of PS and integrating them with UGOS, guided by the baseline trails segmented by length. It is noteworthy that in this experiment, the start point functions as the TH, serving as a central coordinate (the intersection), while intermediate and end points serve as PS coordinates, providing resting areas for commuters. Additionally, the function of this objective can be adjusted based on context, while in this study, only the functions TH and PS are utilized.

In Layer 5, the primary objective is to conduct an EMOO using NSGA-2 algorithm^62,63 to identify the best solutions in line with walkability principles. The fitness objectives (FO)¹¹² detailed in Table 3, along with considerations outl ined in Table 4, are crucial for determining optimal positions for PS aligned with TH central points and evaluating their compatibility with the environment. Integrated PS with UGOS¹¹³ can serve as Nodes or places for placemaking, enhancing community interaction and well-being, while Routes act as transit stops for facilitating smooth mobility and efficient transportation systems.^2,114,115

The genes¹¹⁶ for this simulation are constructed based on distances between potential TH and PS coordinate points, following TOD Standard 3.0 benchmarks.⁷³ These distances range from 300m to 500m in 10m intervals. After formulating the FO and genes, the EMOO is executed using customized settings detailed in Table 5.

Upon simulating with the settings in Table 5 (results shown in Figure 10), the evaluation utilizes Wallacei’s features116, including the Parallel Coordinate Plot (PCP), Standard Deviation Graph (SDG), and Fitness Value Graph (FVG). Figure 11 illustrates consistent results:

1. Most FO values converge, except for FO5, showing tolerable fluctuations due to minor variations in observed PS numbers, as seen in the PCP and SDG.
2. The 3D objective space indicates the Pareto front’s movement toward coordinates (0,0,0) with each simulation generation.¹¹⁶
3. The EMOO and analysis confirm the effectiveness of the evolutionary method in generating numerous solutions from a vast search space. The simulation’s efficiency is highlighted by a runtime duration of about 23mins and 42s.

With this comprehensive data, the next layer will employ solution clustering using ISOA techniques to select the best solution as a design guideline.⁹⁹

3.6. Layer 6: Clustering & selecting the best solution#

Layer 6 compiles the solution candidates output from Layer 5 and iterates them into the best optimal solution using the iterative solution optimization approach (ISOA) as a future design guideline. Figure 12 shows how ISOA combines four methods: Machine-, Algorithm-, Collaborative-, and Context-driven. The detailed ISOA formulations are outlined in Table 6, highlighting their objectives and flexibility in adapting to various problem contexts. We utilize Machine-driven solutions with Wallacei’s analytical features with customizable parameters, while Algorithm-driven solutions are based on Rhino & Grasshopper and can be adjusted parametrically. Collaborative- and Context-driven solutions are developed in Microsoft Excel as needed.

3.6.1. Machine-driven methods#

In this approach, Iter₀ consists of solution candidates from Layer 5, optimized through EMOO, resulting in 2,500 solutions. Figure 13 shows Iter₁, where the K-means clustering algorithm is applied to Iter₀, generating five clusters based on solution characteristics, resulting in 263 solutions distributed across these clusters.

3.6.2. Algorithm-driven methods#

In this approach, the progression through four stages, as referred to in Figure 14, is described as follows:

1. In Iter₂, 263 solutions from Iter₁ are analyzed based on five FO functions (${\text{FO}}_i$), each with a specific weight distribution ($w_{{\text{FO}}_i}$) as outlined in Table 3 and Table 4. The equation for Iter₂ is ${\text{Iter}}_2 = \text{Top}\left(\sum_{i=1}^{m}\frac{{\text{FO}}_i}{w_{{\text{FO}}_i}}\right)$, resulting in 81 solutions.
2. Iter₃ focuses on the top 20 values for each of the five objective functions, leading to ${\text{Iter}}_3 = \text{Top20}\left({\text{Iter}}_2\right)$ and the selection of 100 potential solutions, of which 56 are unique.
3. Iter₄ combines data subsets from Iter₂ and Iter₃ to create an integrated set of 181 solutions. Finally, Iter₅ identifies recurring index values as “dominant solution” values from the 181 solutions, resulting in a refined list of 29 solutions using ${\text{Iter}}_5 = \text{Top}\left(\text{Dominant Solution}, {\text{Iter}}_4\right)$.

3.6.3. Collaborative-driven methods#

In this approach, Iter₆ evaluates the 29 solutions from Iter₅ by visually categorizing them using DFC radar plots to establish clusters. Figure 15 illustrates how each solution is grouped based on visual similarities in objective patterns, resulting in the identification of four distinct clusters.

3.6.4. Context-driven methods#

In this approach, Iter₇ identifies the dominant cluster from Iter₆. Figure 16 demonstrates the application of a customized weight formula for each FO, similar to the customized weighted considerations in Iter₂. The weights assigned to Clusters 1 to 4 are: 240, 190, 160, and 200. An asterisk value after 240 indicates the primary cluster with the highest cumulative sum of weights among all clusters.

In Iter₈, each direction (east, south, and west) is assigned a route orientation (Routes) to measure distances between specific points using coordinates from Figure 17. The weighted route orientation ($W_{\text{Route}_n}$) is calculated as $W_{\text{Route}_n}=w_n \times \text{length of Route}_n$, where $w_n$ represents the weight assigned to Routeₙ. The weight distribution (${\text{Iter}}_8=\sum_{\text{direction}}W_{\text{Route}_{\text{direction}_1}}$) for the route orientations is [0.50, 0.20, 0.30] for east, south, and west directions, respectively. The sum of weighted route orientations in Table 6 is denoted as ${\text{Iter}}_8=\sum_{\text{direction}}W_{\text{Route}_{\text{direction}_1}}$. Based on this calculation, the last five solutions with a score of 402 are selected for the subsequent Iteration, Iter₉ (Table 7).

Solution index assessment for Iter₈. An asterisk (*) denotes selected solutions with the lowest value.

Gen	Idv	${\text{Route}}_{\text{{east}}_1}$	${\text{Route}}_{\text{{south}}_1}$	${\text{Route}}_{\text{{west}}_1}$	$W_{\text{Route}_{\text{east}_1}}$	$W_{\text{Route}_{\text{south}_1}}$	$W_{\text{Route}_{\text{west}_1}}$	${\text{Iter}}_8$
17	5	460	440	500	230	88	150	468
18	4	460	440	500	230	88	150	468
19	9	440	460	500	220	92	150	462
19	11	440	460	500	220	92	150	462
21	12	440	460	500	220	92	150	462
22	16	440	460	500	220	92	150	462
23	10	450	460	500	225	92	150	467
23	14	440	460	500	220	92	150	462
23	16	440	460	500	220	92	150	462
24	19	440	460	500	220	92	150	462
27	34	450	460	500	225	92	150	467
46	40	410	450	500	205	90	150	445*
47	0	410	450	500	205	90	150	445*
47	30	410	450	500	205	90	150	445*
47	47	410	450	500	205	90	150	445*
48	7	410	450	500	205	90	150	445*

In Iter₉, southern transit routes (Routes) are optimized using coordinates from Figure 17. The south weight distribution ($w_{\text {south}_n}$) for route orientations is [0.50, 0.30, 0.20] for south₁ to south₃ directions respectively. The sum of weighted route orientations in Table 8 is denoted as ${\text{Iter}_9} = \sum_{\text{direction (south)}} W_{\text{Route}_{\text{direction (south)}_1}}$. Solutions Gen46 Idv40 and Gen47 Idv30 with a score of 402 are selected due to their lowest score. Figure 18 illustrates one of these solutions meeting the set goals for each FO, serving as a leading design guideline for future TH and PS planning. (Note: Gen46 Idv40 and Gen47 Idv30 refer to the same solution with identical genes and FO values.)

Solution index assessment for Iter₉. An asterisk (*) denotes selected solutions with the lowest value.

Gen	Idv	${\text{Route}}_{\text{{south}}_1}$	${\text{Route}}_{\text{{south}}_2}$	${\text{Route}}_{\text{{south}}_3}$	$W_{\text{Route}_{\text{south}_1}}$	$W_{\text{Route}_{\text{south}_2}}$	$W_{\text{Route}_{\text{south}_3}}$	${\text{Iter}}_9$
46	40	450	350	360	225	70	108	402*
47	0	450	360	350	225	72	105	403
47	30	450	350	360	225	70	108	402*
47	47	450	360	350	225	72	105	403
48	7	450	360	350	225	72	105	403