Designing a Russian–Chinese Omnichannel Logistics Network for the Supply of Bioethanol

Abstract

This research considers an Artificial Intelligence (AI)-driven omnichannel logistics network for bioethanol supply from Russia to China. As a renewable, low-carbon transport fuel, bioethanol plays a critical role in energy diversification and decarbonization strategies for both Russia and China. However, its flammability and temperature sensitivity impose stringent requirements on transport infrastructure and supply chain management, making it a typical application scenario for exploring intelligent logistics models. The proposed model integrates information, transportation, and financial flows into a unified simulation framework designed to support flexible and sustainable cross-border (CB) logistics. Using a combination of machine learning, multi-objective evaluation, and reinforcement learning (RL), the system models and ranks alternative transportation routes under varying operational conditions. Results indicate that the mixed corridor through Kazakhstan and Kyrgyzstan achieves the best overall balance of cost, time, emissions, and customs reliability, outperforming single-country routes. The findings highlight the potential of AI-enhanced logistics systems in supporting low-carbon energy trade and CB infrastructure coordination.

1. Introduction

The International Energy Agency projects a significant increase in global bio-fuel consumption by 2040, with bioethanol becoming a leading alternative to gasoline in the transport sector [1,2]. Bioethanol reduces carbon dioxide emissions [3], is renewable, and can be produced from widely available raw materials—all of which align with global sustainability goals.

Russia’s biofuel industry remains at an early stage of development. During the peak of the national “biofuel enthusiasm,” numerous investment projects for bioethanol production were announced across various regions, including Penza Oblast, Omsk Oblast, Rostov Oblast, Tomsk Oblast, and Altaysky Kray [4]. By 2023, Russia produced approximately 120,000 tons of bioethanol and 700,000 tons of biodiesel, representing only a minor share of the country’s total fuel consumption and indicating limited penetration of biofuels within the national energy mix [5]. Export volumes remain modest, reflecting low integration of bioethanol into cross-border energy trade and emphasizing the early-stage commercialization of the industry.

As of January 2025, Russia’s annual ethanol production is estimated at 5.6 billion liters, primarily used for the production of vodka. There are nearly 140 plants in the country, with a total capacity of 9.5 billion liters. Modern production technologies, however, are used only at 10 to 12 industrial sites [6]. Russia still lacks large-scale dedicated fuel ethanol facilities; most existing distilleries are small, energy-intensive, and low in automation, making fuel-grade ethanol production economically challenging in many cases.

Nevertheless, Russia’s vast arable land, substantial untapped biomass resources—including agricultural and forestry by-products [7]—growing urbanization, and cold-climate requirements for ethanol-blended fuels indicate significant potential for accelerated bioethanol market development. At the same time, China, as the world’s largest energy consumer, is actively promoting the decarbonization of its transportation sector. Its policy to expand ethanol–gasoline blending (E10) has created a growing domestic market for bioethanol. The E10 standard (GB 18351-2017) [8]—fuel containing 10% bioethanol blended with 90% gasoline—has already been implemented in multiple Chinese provinces as part of national efforts to reduce greenhouse gas emissions and dependence on fossil fuels [9,10]. E10 is compatible with most modern gasoline engines and does not require vehicle modifications. Its widespread adoption reflects both technical feasibility and regulatory support, and underscores the need for stable and scalable supply chains for bioethanol distribution.

Compared with electric energy options, which dominate long-term decarbonization pathways but require extensive charging infrastructure and fleet replacement, bioethanol provides an immediate, compatible solution for existing internal combustion vehicles, serving as a transitional approach complementary to electric mobility.

In this context, the potential relevance of Central Asian transit corridors, particularly those passing through Kazakhstan and Kyrgyzstan, deserves preliminary attention. These routes are increasingly recognized for their role in connecting Russian supply regions with Chinese demand centers under the Belt and Road framework. While the present study does not assume their superiority in advance, acknowledging their strategic position helps to frame the logistics design problem more realistically and provides a contextual motivation for including such corridors in the simulation framework, with systematic validation deferred to Section 3.

Establishing a CB logistics network for bioethanol between Russia and China supports both countries’ strategies for energy diversification and long-term objectives to reduce carbon intensity in the transport sector. It also enhances bilateral trade under the framework of the Belt and Road Initiative (BRI). However, existing CB logistics systems are mainly designed for conventional commodities and are poorly suited to bioethanol’s specific requirements, such as temperature stability, emissions control, and rapid distribution [11].

Moreover, conventional linear supply chains are increasingly inadequate in addressing today’s market complexity. As demand becomes more volatile, the need arises for a logistics network that is flexible, scalable, and capable of real-time coordination across multiple channels [12,13,14,15]. In particular, a logistics network must integrate information, transportation, and financial processes to function efficiently across borders.

This study focuses on designing an omnichannel logistics network for transporting bioethanol from Russia to China. It emphasizes small-batch transportation strategies and explores the application of AI to support decision-making. The research scope includes supply-side resources in the Kirov region, major road and railway corridors to China, and the logistical infrastructure required for CB energy trade.

Existing research on CB logistics optimization has mainly focused on either conventional multimodal planning or discrete improvements in carbon emission reduction, route scheduling, or customs facilitation. For instance, studies such as those by Li [16] and Pavlenko [17] have developed multimodal transport models considering cost and time constraints. Recent empirical research has confirmed that digital platforms significantly enhance transparency and operational efficiency in CB trade. For instance, Irawan [18] demonstrated that blockchain-enabled smart contracts and integrated digital payment systems reduce procedural uncertainty and increase the trust level among international trading parties. However, these approaches often treat logistics performance indicators in isolation and lack a systemic, multi-criteria evaluation framework. In parallel, the application of AI in logistics [19]—particularly ML and RL—has gained attention [20,21]. However, their integration into dynamic decision-making systems for international energy product logistics remains limited.

Prior to the implementation of Artificial Intelligence (AI)-driven logistics interventions, bioethanol exports were typically organized through fragmented, single-channel transport routes with minimal digital coordination. These routes relied heavily on manual scheduling and customs procedures, resulting in prolonged clearance times, low vehicle utilization, and increased exposure to operational risks. Within the broader energy and logistics chain, bioethanol, as a renewable and low-carbon alternative to gasoline, plays a strategic role in achieving national carbon neutrality goals and promoting diversification of the energy supply structure. However, its physical characteristics—flammability, temperature sensitivity, and the need for specialized handling—impose stricter requirements on transport infrastructure, safety protocols, and supply chain coordination. This complexity makes it an ideal case for exploring intelligent logistics models aimed at enhancing reliability, reducing environmental impacts, and supporting sustainable bilateral energy trade.

This study, therefore, focuses on conceptual design rather than physical implementation or infrastructure construction, which clarifies the architectural scope and distinguishes design from implementation aspects. Building upon this clarified scope, it is important to note that, although recent studies have examined aspects of biofuel logistics and Russia–China trade, several research gaps remain:

• Existing models often focus on linear supply chains and fail to consider omnichannel approaches that integrate multiple transport paths and decision points;
• There is limited analysis of regional suitability between bioethanol production sites in Russia and export-oriented transportation routes;
• AI applications in this domain have mostly been limited to general supply chain tasks, with little attention to price prediction, route optimization, or demand forecasting tailored to bioethanol;
• Research has emphasized static optimization without accounting for the dynamic evolution of logistics networks from pilot projects to full-scale operations.

To address the identified research gaps, this study proposes a methodological framework for the Russian–Chinese bioethanol supply logistics network based on the omnichannel concept. Previous CB logistics studies have typically relied on deterministic or static optimization models that assume fixed operating conditions and lack adaptability to regulatory or demand uncertainty. Based on our experience in CB logistics, such methods are insufficient to handle the requirements of hazardous cargo and the fluctuations in carbon emission policies.

Unlike these traditional approaches, the approach emphasizes adaptive coordination and coordinated optimization in CB logistics by integrating multiple nodes, transport modes, and flows, including information, goods, and capital. Within this framework, AI-assisted decision-support modules are developed based on deterministic multi-objective optimization to evaluate predefined scenario configurations, thereby enhancing operational flexibility and responsiveness without relying on dynamic system simulation or stochastic modeling. To reflect the phased nature of the supply chain’s development, the study develops a transition model that supports the shift from small-batch shipments to large-scale operations. In addition, sustainability metrics are embedded into the logistics planning process to balance environmental and economic objectives through carbon accounting and green performance evaluation.

The authors aim to design a conceptual model of an omnichannel logistics network for supplying bioethanol from Russia to China.

Unlike conventional linear logistics systems, the model combines physical transportation, data flow, and financial coordination into a single, adaptive simulation environment. It uses AI methods, including RL, to refine routing decisions in response to operational factors such as delivery time, emissions, and customs procedures. The study’s original novelties are summarized in five key points. First, it brings the omnichannel concept—commonly applied in consumer goods logistics—into the energy sector, where supply chains often lack flexibility. Second, it integrates sustainability and algorithmic decision-making to support route selection in complex international contexts. Third, it embeds sustainability metrics, including life-cycle carbon analysis, into the core of the routing decision process, enabling more environmentally informed logistics planning. Fourth, it operationalizes the model through a data-driven AI simulation environment, which allows continuous strategy adjustment and provides real-time feedback. Finally, the study applies the framework to the Eurasian corridor between Russia and China, where simulation results confirm its feasibility and robustness under uncertain cross-border conditions. These contributions position the model as a scalable and adaptable solution for managing cross-border energy supply chains facing operational and policy uncertainties.

Building on these elements, the study first presents the methodological framework and simulation design underpinning the research, then validates the model’s applicability, robustness, and impact on environmental performance through route optimization results, and finally discusses the scientific significance, practical value, and contribution to sustainable cross-border energy trade, highlighting the role of AI-enabled logistics in improving operational efficiency and achieving sustainability objectives.

2. Methodology

The study builds upon conventional logistics network design approaches and integrates the omnichannel logistics concept to develop and simulate an AI-driven, end-to-end logistics network. The model focuses on road transportation as the primary mode [22,23], aiming to enable seamless and responsive CB supply of bioethanol from Russia to China.

To systematically construct an omnichannel logistics network for bioethanol supply [24], the authors developed a three-tiered coordination model comprising an information layer, financial layer, and logistics layer [25]. Each of the three layers reflects a key operational domain within the logistics system: information processing, financial transactions, and material flow (see Figure 1). Their integration through adaptive coordination mechanisms enables different subsystems to interact in real-time and align resource allocation across the network. Data from IoT-enabled sensors and digital customs platforms feed into the information layer, triggering adaptive decision updates. Leveraging this information, the AI module employs RL to reallocate shipments across channels, prioritize vehicles compliant with flammable liquid regulations, and adjust dispatch frequencies in response to inspection regimes or demand fluctuations. This structure improves both the responsiveness and long-term viability of the CB logistics configuration.

The logistics network was adjusted to reflect actual conditions, taking into account available resources and market requirements. The design is based on a three-layer coordination model and starts from operational bioethanol facilities located in Russia. It relies on the country’s road transport infrastructure to support CB deliveries of smaller, more frequent batches to China. This configuration suits the needs of early-stage logistics deployments, where the ability to adjust quickly is crucial. The technical feasibility of using E10 fuel is well established, as most modern gasoline engines are compatible with ethanol–gasoline blends up to 10% without requiring mechanical adjustments. This compatibility has facilitated the expansion of ethanol use in transport without major infrastructure changes at the consumer level [8,26,27]. Consequently, logistics planning must ensure a consistent supply of bioethanol to blending facilities and regional fuel terminals to meet increasing demand in line with China’s nationwide ethanol mandate [28].

This study utilizes simulated data to evaluate the performance of the proposed omnichannel logistics network for CB bioethanol supply. The use of simulation is motivated by several factors. First, due to the early-stage development of the Russia–China bioethanol trade, real-time operational data, such as shipment volumes, exact customs clearance times, or CB bottlenecks, are either unavailable, incomplete, or not publicly accessible. Second, simulation allows for controlled variation in key parameters to test the robustness and adaptability of the decision-making model under different policy, cost, and environmental scenarios.

From a logistics network design perspective, small-batch transportation supports multi-node and multi-corridor optimization, facilitates AI-based simulation and scheduling, and suits the high-frequency delivery operations typical of early-stage or sensitive markets (such as China’s emerging bioethanol adoption market), where demand uncertainty, corridor reliability, and customs-related risks must be managed. In this study, small-batch transport is defined as a single tanker truck shipment, typically carrying 25–35 tons of bioethanol, a scale that enables the simulation to capture the operational characteristics of early cross-border bioethanol trade while providing a benchmark for evaluating future transitions toward bulk transport modes.

The simulated dataset is constructed based on representative values drawn from open-source logistics statistics, academic studies on energy transport corridors, and government-reported parameters associated with road networks, fuel prices, and bioethanol infrastructure in Russia and China. To validate the credibility of simulated data, we benchmarked key logistics parameters against available statistics for CB energy transport.

The subsequent evaluation employs an explicit weighting scheme for the five performance criteria, namely transportation time (25%), transportation cost (30%), carbon emissions (20%), customs clearance convenience (15%), and total transportation distance (10%).

By simulating 1000 iterations for each routing scenario, the evaluation process ensures convergence of results, thereby offering statistically robust evidence for the comparative performance of alternative routes.

To enhance the adaptability of the omnichannel cross-border logistics system under uncertainty, a reinforcement learning (RL) module is embedded into the proposed decision-making framework. Rather than aiming for complex end-to-end learning, the RL module functions as a dynamic adjuster that complements the multi-criteria decision-making (TOPSIS) layer. Its purpose is to optimize resource allocation and routing strategies in real time, driven by the objective of maximizing the composite performance function $F\left(R_i\right)$.

The design of the RL module includes three core components—state space, action set, and reward function—which are described in detail in

Table 1. Technical Components and Parameters of the RL Module.

Component	Definition	Configuration in This Study
State Space (S)	A set of variables describing the logistics system status at a given time	Composed of five core variables aligned with the composite performance score function $F\left(R_i\right)$: transportation cost, time, carbon emissions, customs clearance score, and distance. Additionally includes real-time environmental factors such as vehicle availability and border inspection intensity.
Action Set (A)	Operations that the agent can take under a given state	Reallocating shipment volumes across three predefined cross-border corridors (Kazakhstan, Kyrgyzstan, Hybrid) and adjusting dispatch frequency
Reward Function (R)	Immediate feedback signal guiding the agent’s learning	$R_t=F\left(R_i\right)$, where $F\left(R_i\right)$ is the weighted composite score based on the five performance KPIs
Policy (π)	Rule for selecting actions	ε-greedy strategy balancing exploration and exploitation
Update Algorithm	Method for learning value or policy functions	Q-learning with an adaptive learning rate
Discount Factor (γ)	Trade-off between short-term and long-term rewards	$\gamma$ = 0.9 (emphasizing long-term performance objectives)
Training Parameters	Learning rate, iteration number, convergence criteria	1000 iterations per scenario; learning rate α = 0.1 (with decay); convergence threshold defined as $\Delta Q<{10}^{-3}$

. The reward function is directly defined as the composite score $F\left(R_i\right)$, thereby ensuring full alignment between the agent’s learning goal and the evaluation framework.

The module is implemented using a standard Q-learning algorithm with an ε-greedy exploration policy. This setup enables effective learning under uncertainty and supports rapid adjustment in response to fluctuating cross-border constraints. Additional parameters such as the learning rate (α = 0.1), discount factor (γ = 0.9), and convergence threshold ($\Delta Q<{10}^{-3}$) are used to ensure robustness and replicability.

It is important to note that Q-learning and ε-greedy strategies are not the methodological focus of this work but are selected for their maturity and stability in dynamic decision environments. The methodological contribution lies in integrating this RL mechanism within an AI-enhanced, multi-layered logistics coordination framework.

Table 1 presents a summary of the technical components, implementation logic, and parameter settings used in this study.

This approach emphasizes model behavior, decision dynamics, and sensitivity to parameter changes rather than empirical prediction. Consequently, while the results provide insights into optimal route structures and system performance patterns, they are not intended to reflect real-world performance outcomes with statistical precision. The proposed methodology is designed to evaluate the performance of the AI-based routing model in terms of its structure, adaptability, and decision logic. This provides a foundation for applying the system to real-world data once the bioethanol trade between Russia and China becomes more developed.

To explore both the financial and environmental aspects of the logistics network, the authors used Life Cycle Assessment (LCA) [29,30] to estimate carbon output at each stage of the distribution chain, including packaging, warehousing, international transport, and final delivery within China. When paired with a cost analysis, the results help identify which transport strategies perform best under varying shipment sizes and market scenarios. The combined method provides a more solid foundation for selecting routes that strike a balance between economic objectives and environmental responsibilities within an omnichannel logistics framework.

Figure 2 illustrates the overall flowchart of the study’s methodology. The diagram presents the sequential research steps, including defining the scientific purpose and scope, developing the methodological framework (three-layer coordination model), executing simulation and analysis, evaluating results (stability and sustainability assessments), and discussing scientific and practical implications. Side annotations specify the key elements of each step, providing an intuitive overview of the study’s logic and structure.

3. Results

As a starting point in developing a logistics network for bioethanol trade between Russia and China, the authors focused on selecting appropriate origin cities. The effectiveness of the entire network depends on this choice, which must reflect real access to supply sources, available transportation routes, and a supportive policy context [31]. To guide this process, a structured evaluation was conducted using five analytical categories:

• Resource Availability—presence of production facilities and access to feedstock.
• Transport Capacity—strength of road networks and border connectivity.
• Geographic Access—distance to major export corridors toward China.
• Regulatory Conditions—administrative efficiency and trade openness.
• Regional Stability—socio-political predictability in the area.

Using the defined criteria, five representative Russian cities were selected and evaluated as potential starting points for bioethanol exports to China. A comparative analysis was conducted to assess each city’s strengths and limitations across key dimensions, including the availability of resources, transport infrastructure, proximity to border crossings, regulatory environment, and regional stability.

Table 2. Comparative analysis of starting cities for Russian bioethanol supply.

City	Advantages	Disadvantages	Comprehensive Evaluation
Kirov	Abundant agricultural and forestry resources, with infrastructure for bottling plants, an economically stable, and a favorable policy environment.	Located far from the Russian–Chinese border (approximately 4000 km), with a long transportation cycle.	Initial small-scale supply is appropriate, with favorable overall conditions.
Novosibirsk	Transportation port with excellent road conditions and well-established routes to Manzhouli.	Bioethanol resources are relatively weak and require external raw materials.	Expansion in the medium to long term may be considered, and mid-term transition points should be set.
Irkutsk	Near the Outer Baikal Port, the highway is relatively closer to China.	Predominantly heavy industry, with insufficient bioethanol resources.	Well suited as a logistics hub, but unsuitable for direct supply.
Tomsk	Well-developed scientific research and agricultural resources, with potential for developing new processes.	Relatively poor transportation accessibility.	Advantages in technical support, but unsuitable for initial launch locations.
Chita	Exceptionally close to the Outer Baikal border crossing, with easy access to the Chinese market.	Regional economies are underdeveloped, and supply chain infrastructure is poorly developed.	Suitable for transit, unsuitable as a departure city.

summarizes the advantages and disadvantages of each city based on this multi-criteria evaluation. The screening process enabled a structured assessment of candidate locations and provided a basis for selecting the most viable starting point for the logistics network.

To facilitate comparison across evaluation criteria, the authors visualized the scoring results using radar charts (Figure 3) and average score bar charts (Figure 4). These diagrams help compare the advantages and limitations of each proposed city, offering measurable justification for selecting a suitable origin point for the logistics system.

While Kirov and Novosibirsk exhibited comparable overall ratings, Kirov was selected due to its stronger performance in key criteria critical for early-stage deployment— reliable access to bioethanol feedstock, existing production infrastructure, and a predictable regulatory environment. These factors make Kirov a practical choice for piloting small-batch distribution and for subsequent operational scale-up.

Kirov Oblast possesses a robust agricultural resource base, producing approximately 586 thousand tons of grains and legumes, 108 thousand tons of potatoes, and 63 thousand tons of vegetables in 2024 (

Table 3. Gross harvest of crop products in Kirov Oblast in 2024 by category (thousand tons) (Based on the data from https://43.rosstat.gov.ru/storage/mediabank/) (accessed on 30 April 2025).

Crop Category (Weight After Processing)	All Farms	Agricultural Enterprises	Household Farms	Peasant (Farm) Enterprises and Individual Entrepreneurs
Grains and Legumes	586.4	547.1	0.9	38.4
Potatoes	107.7	12.0	85.8	9.9
Vegetables (open and protected ground)	62.8	1.2	59.6	1.9

). Agricultural sales data confirm the stability of this resource supply, with 320 thousand tons of grain, 30 thousand tons of potatoes, and more than 6 thousand tons of vegetables sold in 2024, reflecting a positive growth trend compared with 2023 (

Table 4. Sales of agricultural products in Kirov Oblast by category (thousand tons) (Based on the data from https://43.rosstat.gov.ru/storage/mediabank/) (accessed on 30 April 2025).

Product Category	2024 (Thousand Tons)	2024 as % of 2023
Grain of cereals and legumes	319.7	107.8%
Potatoes	30.2	118.4%
Vegetables	6.2	111.9%

). These categories represent available feedstock that could be used for bioethanol production in the future. However, it should be emphasized that current bioethanol output in Kirov remains modest and is largely confined to pilot or experimental projects. The present analysis therefore focuses on the region’s agricultural resource potential and its suitability as a supply node in the conceptual logistics network [32].

Improving transport route selection plays a central role in making the Russia–China bioethanol supply chain more efficient and environmentally sound. At the current stage, where small-batch and high-frequency road transport is dominant, route selection must balance time efficiency, cost, carbon emissions, and the reliability of border crossings.

To explore routing alternatives, the authors developed an AI-supported system that generates and tests multiple CB transport configurations. The initial setup assumes Vyatka (Kirov) as the point of origin and examines three road-based options crossing through Central Asian regions. These alternatives are summarized in

Table 5. Candidate paths.

Route Number	Route Name	Description
Route 1	Kazakhstan Channel	Kirov → Kazakhstan → Alashankou → China
Route 2	Kyrgyzstan Channel	Kirov → Kyrgyzstan → Irkeshtam → China
Route 3	Hybrid Channel	Kirov → Kazakhstan → Kyrgyzstan → China

.

To complement the overview in Table 5, Figure 5 provides a schematic visualization of the candidate transport corridors. Rather than indicating precise operational routes, the map highlights conceptual geographic alternatives, showing how the corridor could run exclusively through Kazakhstan, exclusively through Kyrgyzstan, or follow a hybrid channel combining both. This schematic representation allows readers to better grasp the spatial logic of alternative pathways and the implications of different border-crossing choices.

Figure 5 shows the schematic representation of candidate transport corridors from Kirov to China, including the Kazakhstan Channel (Route 1, solid blue line), Kyrgyzstan Channel (Route 2, solid orange line), and Hybrid Channel (Route 3, dashed green line). The map was generated by the authors via GeoJSON.io.

To test how the logistics system performs in uncertain and variable conditions, the authors implemented an AI-driven simulation tool for route selection. This model compares several candidate paths by analyzing key factors, including delivery time, transportation expenses, environmental footprint, and the stability of customs-related operations at border points.

Three candidate routes are considered: a direct route via Kazakhstan, a route via Kyrgyzstan, and a hybrid route combining both corridors. As illustrated in Figure 6, the model integrates these options into a unified decision-making framework. It links the selection of the departure city, the routing alternatives, and relevant input data into an AI-driven simulation environment. The system then performs multi-objective evaluation to determine the optimal path under varying operational scenarios [33].

Figure 6 illustrates the structure of the AI-based decision framework used in this study. At the core of the model is a three-tiered collaboration structure—information flow, financial flow, and physical logistics flow—that enables coordination among customer requirements, financial settlements, and transportation execution. All candidate routes are passed through a standard data interface that consolidates key metrics—such as delivery time, economic cost, emissions, and customs reliability—for analysis. The architecture supports simultaneous simulations involving multiple routing scenarios, criteria, and algorithmic settings. This capability demonstrates the system’s ability to adapt and scale effectively within dynamic CB conditions.

The evaluation relies on five main indicators that reflect both financial and environmental considerations:

Total distance covered $\left(D_i\right)$;

Shipment duration $\left(T_i\right)$;

Logistical expenditures $\left(C_i\right)$;

Greenhouse gas emissions $\left(E_i\right)$;

Ease of customs procedures $\left(S_i\right)$.

To formalize the route evaluation process, we define a composite score function $F\left(R_i\right)$ that aggregates these indicators into a unified decision metric:

$$F\left(R_i\right)=w_1{\displaystyle \frac{D_{min}}{D_i}}+w_2{\displaystyle \frac{T_{min}}{T_i}}+w_3{\displaystyle \frac{C_{min}}{C_i}}+w_4{\displaystyle \frac{E_{min}}{E_i}}+w_5{S}_i$$

(1)

where $w_j\ge 0$ are the indicator weights with ${\sum }_{j=1}^5{w}_j=1$; the ratio terms normalize cost-type indicators so that lower values yield higher scores, while the customs score $S_i\in \left[0,1\right]$ is treated as a direct benefit indicator. The optimal route $R^{\ast }$ is then obtained as follows:

$$R^{\ast }=\mathit{arg}\underset{i\in \mathcal{R}}{\mathit{max}} F\left(R_i\right)$$

(2)

These criteria serve as the basis for ranking each option under various operational scenarios.

To enable comparison across indicators, the study applies a Weighted Aggregation Model [34,35], which standardizes all variables on a [0, 1] scale. Based on the practical priorities of CB logistics management, a set of relative weights is assigned to each KPI. A linear scoring function is then used to calculate a composite score for each route and determine the overall ranking.

To improve prediction accuracy under variable conditions, the model incorporates a Random Forest (RF) Regression algorithm [36,37], which forecasts key input values affected by external fluctuations. These predictions are fed into the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) model [38] (Figure 7), which ranks the routes based on their distance from an ideal performance profile. This combination of weighted scoring and machine learning enables robust and interpretable route selection.

The overall process of route prediction and ranking is illustrated in Figure 7, which shows how it feeds into the TOPSIS ranking procedure for robust route selection.

In the simulation environment, the system ingests standardized route indicators along with predictive variables generated by the RF model. These inputs are processed through a weighted aggregation function and subsequently evaluated by the TOPSIS multi-criteria ranking model (with fixed weights), producing an initial ranking of routing schemes. It is important to emphasize that RL in this study is not employed as an independent route-searching algorithm. Instead, during successive iterative cycles, the RL module observes multidimensional system states, including transportation cost, delivery time, carbon emissions, and customs reliability, and adjusts cargo distribution and dispatch frequency across predefined CB corridors. The optimization objective is to maximize the composite performance score output by the TOPSIS model, thereby enhancing the consistency and robustness of routing decisions under uncertainty while preserving the stability of evaluation criteria. It should be stressed that the primary contribution of this research lies in the design and methodological validation of the CB logistics network, rather than in the engineering development of the RL algorithm itself. The simulation structure is depicted in Figure 6. It shows how input data, performance indicators, machine learning predictions, and decision algorithms interact in a feedback-enabled loop to identify optimal solutions.

To enhance the robustness and reliability of the route evaluation process, the model executes 1000 independent simulation runs for each input configuration. The results are averaged, and statistical error analysis is applied to assess the consistency of route performance. This study is a proof-of-concept simulation using multiple scenarios to test model behavior under different parameter assumptions. Due to limited industry data, no empirical parameter calibration has been conducted; therefore, the results reflect theoretical potential rather than actual operational performance. To reduce distortion caused by extreme values, the Interquartile Range (IQR) method [39,40] is used to identify and remove outliers from the dataset [41].

The integrated AI-driven framework (Figure 8) enables multi-route evaluation through standardized KPI scoring, machine learning-based input prediction, and multi-round simulation using RL. The conceptual framework adjusts scoring weights and outputs comprehensive rankings and sensitivity matrices, which are then used to inform logistics scheduling and resource allocation strategies [42,43].

Two operational scenarios illustrate this adaptability:

(1)

When hazardous material inspection is tightened on the Kazakhstan corridor, loads are reallocated to the hybrid Kazakhstan–Kyrgyzstan route using vehicles pre-certified for stricter standards.

(2)

When CB carbon quotas are temporarily tightened, the system prioritizes lower-emission vehicle fleets, such as trucks equipped with advanced emission-control technologies, and dynamically consolidates shipments while adjusting driving schedules to minimize idle time and reduce total carbon intensity.

To further analyze the simulation outcomes, Figure 9 illustrates the average performance of the three candidate routes across five normalized indicators. The results show that Route 3 maintains a well-rounded score profile, especially with respect to emissions, cost, and customs-related stability.

Figure 10 displays the rankings derived from the TOPSIS method, underscoring the system’s capacity to resolve multi-objective trade-offs. Across multiple simulation runs, Route 3 consistently emerges as the most favorable alternative.

Figure 11 illustrates how route score distributions change after outlier elimination, showing that Route 3 maintains lower dispersion and a more compact interquartile span, indicating improved stability under variable input conditions. In the initial results (left panel), Routes 1 and 2 exhibit wider score ranges and multiple extreme outliers, indicating performance fluctuations under certain conditions. Route 3 shows a relatively higher median score and fewer extreme values. After outlier removal (right panel), score distributions become more compact, with Route 3 demonstrating the narrowest interquartile range and highest median value.

Figure 9, Figure 10 and Figure 11 reflect the key outcomes of this analysis stage. These results confirm the robust and stable performance of Route 3 compared with the alternative routes. In the boxplots, the central line denotes the median, the box boundaries represent the first and third quartiles, the whiskers extend to 1.5 × IQR, and the dots indicate outliers.

This integrated modeling process improves the scientific rationality and transparency of route decision-making [42,43], providing a reliable basis for digital and intelligent decision-making by stakeholders in the Russian–Chinese omnichannel logistics network, including logistics operators, transportation dispatch platforms, policy regulators, and end purchasers in China.

The research conducted a route grouping statistical analysis based on simulated data. Simulation data were grouped by route number (Route 1, Route 2, Route 3), and the mean and standard deviation of key indicators, including transportation distance, transportation time, transportation cost, carbon emissions, and customs clearance convenience score, were calculated for each group. This was used to identify the relative strengths and weaknesses of each route in terms of efficiency, economy, and environmental friendliness. To ensure the scientific and practical validity of the evaluation framework, the weighting scheme introduced earlier was grounded in a hybrid approach that combined expert scoring with literature benchmarking. A panel of five logistics professionals independently assessed the importance of the five performance indicators, and the averaged results were normalized to yield the final weights: transportation time (25%), transportation cost (30%), carbon emissions (20%), customs clearance convenience (15%), and total transportation distance (10%). These weights were then applied to calculate composite scores for each sample and to identify outstanding candidate path combinations.

To evaluate the robustness of route selection under varying operational priorities and regulatory uncertainties, a comprehensive sensitivity analysis of the evaluation criteria weights was conducted. Three predefined policy-related scenarios were introduced to reflect real-world conditions:

Policy Tight, simulating enhanced customs inspection and regulatory restrictions;

Incident, representing sudden delays caused by border congestion or emergency closures;

Low-Carbon Policy, emphasizing environmental objectives and carbon emission reductions.

In addition to these targeted scenarios, six random perturbation scenarios (Random_1–Random_6) were generated, where all five evaluation criteria weights were randomly adjusted and normalized to create unstructured variations. These random scenarios do not correspond to any specific policy phase but serve as stress tests to ensure that route selection is not overly sensitive to arbitrary changes in weight distribution.

Furthermore, to capture systematic but bounded uncertainties, ±5% and ±10% perturbations were applied independently to each criterion in the baseline weighting scheme, generating 20 additional scenarios. These perturbations simulate mild and substantial fluctuations in decision priorities, such as temporary shifts in time sensitivity, cost concerns, customs facilitation emphasis, emission control priorities, or route distance considerations. Combining all predefined, random, and perturbation-based scenarios resulted in a total of 30 weight-adjustment conditions.

Figure 12 presents the robustness evaluation of route performance under different scenario-based weightings.

As shown in Figure 12a, the heatmap summarizes route scores across 30 weighting scenarios, ordered as baseline, policy-driven configurations, ±5% perturbations, ±10% perturbations, and random weights. The figure visualizes the relative performance of all three routes under varied decision priorities. Notably, Route 3 consistently demonstrates superior scores, indicating its robustness across diverse preference structures.

Figure 12b illustrates the Route 3 advantage gap, defined as the score margin between Route 3 and the second-best route in scenarios where Route 3 is ranked first. This margin provides direct evidence of performance resilience, showing that Route 3’s superiority is not only frequent but also statistically significant under both structured policy shifts and unstructured random variations.

The results demonstrate that Route 3 consistently outperforms the alternatives, ranking first in over 90% of all scenarios. Even in cases where Route 3 did not rank first, its performance gap relative to the top route was minimal, and its advantage in predefined policy scenarios remained statistically significant. These findings confirm that Route 3’s superiority is robust to both structured policy shifts and unstructured random disturbances, validating its strategic suitability for hazardous bioethanol CB transport.

From a practical perspective, these outcomes highlight the importance of implementing routing frameworks that are adaptive to external uncertainties. The AI-driven module enables real-time adjustment of decision weights in response to policy changes, market fluctuations, or operational disruptions. Accordingly, it is recommended that logistics operators perform the following:

• Continuously monitor and adjust decision weights in response to policy or market changes.
• Allocate vehicles and optimize node configurations with greater flexibility.
• Establish contingency plans for high-uncertainty scenarios.

To enhance route selection robustness, the proposed model integrates multi-criteria evaluation with AI-based modules, including predictive forecasting, adaptive learning, and outlier filtering. Rather than reiterating technical processes, this section emphasizes their collective outcome: consistent identification of the Kazakhstan–Kyrgyzstan corridor as the most balanced option across cost, environmental, and operational dimensions. The results reflect the system’s ability to adjust to conflicting objectives and input uncertainties while maintaining transparent decision logic.

Figure 13 visualizes the sensitivity of each route to changes in input variables. Color intensity indicates the relative variance, with darker shades representing greater sensitivity. Route 1 exhibits notable sensitivity in customs clearance and transportation cost, while Route 2 shows moderate variation across multiple indicators. In contrast, Route 3 consistently displays the lowest variance across all performance dimensions, confirming its high level of stability and operational reliability.

This outcome points to the AI module’s effectiveness in identifying routing configurations that remain viable despite fluctuating parameters. The selected pathway, which passes through Kazakhstan and Kyrgyzstan, is incorporated into subsequent logistics planning steps, helping to fine-tune vehicle allocation and node positioning. Overall, the integration of RL with multiple evaluation models improves the system’s ability to navigate complex decision environments, strengthens result stability, and contributes to more transparent route optimization in CB logistics.

The functional components of the AI-enabled modules and their representative outputs are summarized in

Table 6. Adaptation of the evaluation framework for selected renewable energy sources.

Component	Technology Category	Purpose in Framework	Representative Output
Data Reliability Enhancement	Data Preprocessing	Improve the integrity and consistency of heterogeneous inputs	Standardized and validated dataset
Forecasting Module	Predictive Modeling	Anticipate variations in key external parameters	Scenario-specific input estimates
Integrated Performance Scoring	Multi-criteria Evaluation Model	Consolidate multi-dimensional criteria into comparable scores	Ranked performance matrix of route alternatives
Adaptive Optimization Engine	Dynamic Policy Adjustment	Adjust decision priorities dynamically under changing contexts	Context-sensitive routing recommendations

. These components establish a coherent analytical foundation for handling uncertain inputs and supporting adaptive route planning under changing operational conditions.

The authors suggest scientific discussion regarding AGI-driven resource allocation in a wide sense. The researchers suggest that AI-driven logistics carry potential value in reducing systemic inefficiencies and, when combined with predictive circular economy models and dynamic taxation systems, may generate positive impacts at the macroeconomic level. While this perspective lies beyond the empirical scope of the present study, it underscores the forward-looking significance of integrating circular economy concepts with adaptive policy instruments. International AI Committee (IAIC, https://iaic.world/) is able to consider the mentioned issues.

4. Discussion and Implications

The simulation analysis reveals that each of the three candidate routes follows a distinct performance profile. Route 1 demonstrates strength in terms of travel time and distance, but lacks reliability in customs processing. Route 2 performs favorably in customs-related metrics, but it suffers from elevated transportation expenses and emissions, which reduce its appeal in green transition contexts. Route 3 achieves the most consistent balance across all criteria and shows stable results across multiple simulations, making it the most viable candidate for CB bioethanol transport.

Importantly, the present simulation is anchored in small-batch transport scenarios, consistent with the pilot and exploratory nature of current cross-border bioethanol operations. This focus allows the framework to capture the technical and institutional challenges inherent in early-stage logistics design. While bulk transport is likely to generate more pronounced economies of scale and efficiency gains, such conditions extend beyond the empirical scope of this study and are identified as a promising avenue for future investigation.

The proposed framework combines logistical, informational, and financial components into a unified simulation platform powered by AI. This structure incorporates multi-model evaluation and RL to adjust decision parameters dynamically. As the system registers fluctuations in delivery time, emission levels, or customs performance, it updates route priorities accordingly. This makes the tool relevant for decision-making in uncertain, policy-sensitive environments, especially for planners, regulators, and logistics service providers.

Russian–Chinese CB logistics networks have faced unprecedented pressure and complexity in recent years, especially with the rise in total freight volume between the two countries driven by the BRI. In 2024, Russian–Chinese bilateral trade reached USD 244.8 billion, representing a 1.9% year-on-year increase and setting a record high for the second consecutive year [44]. For example, the total volume of goods imported and exported through the Manzhouli Railway Port in 2024 exceeded 22 million tons, representing a year-on-year increase of 939,400 tons, a 4.5% increase [44]. As of 25 October 2024, the total cargo throughput at Manzhouli Port was 20.057 million tons, representing a year-on-year increase of 9.4%. The railway port processed 18.08 million tons, up 5.4% year-on-year; the highway port processed 1.977 million tons, up 65.8% year-on-year [45]. Russian–Chinese energy cooperation is an essential cornerstone of practical cooperation between the two countries, with energy trade accounting for more than one-third of total trade between Russia and China [46]. Russia has surpassed Saudi Arabia to become China’s largest supplier of crude oil, accounting for 21.7% of China’s total crude oil imports [47]. Driven by carbon neutrality strategies and stimulated by green energy substitution policies, bioethanol, a representative of gasoline additives and low-carbon fuels, has been gradually included in the list of policies prioritized for encouragement, demonstrating significant growth potential and strategic value.

Beyond the quantitative growth of Russian–Chinese trade, the geopolitical context (particularly conflicts and the resulting political and economic realignments) has introduced new dynamics that directly and indirectly affect the strategic outlook of renewable energy cooperation. Several implications can be identified:

• International sanctions have limited access to certain equipment and technologies relevant to bioethanol production and transport infrastructure, reinforcing Russia’s shift toward self-sufficiency and domestic technology solutions.
• The redirection of Russian trade flows toward Asian markets, including China, has strengthened the strategic interest in developing eastbound supply chains, including for renewable fuels like bioethanol.
• Despite logistical and financial disruptions in broader energy markets, bioethanol remains a relatively underdeveloped but strategically significant sector for Russia’s green export diversification, especially in light of growing demand for low-carbon fuels in China.

Although bioethanol accounts for less than 1% of the current CB trade between Russia and China, its role in green and low-carbon transition and energy strategy coordination is becoming increasingly prominent, possessing significant modeling value and practical significance. First, from the perspective of policy drivers, bioethanol, as a clean energy alternative, aligns with China’s “dual carbon” goals and Russia’s renewable energy export strategy, making it a pilot product for transforming energy cooperation between the two countries. Secondly, the technical complexity of transporting bioethanol is significantly higher than that of general cargo. Its high flammability, temperature sensitivity, and requirement for specialized vehicles and storage facilities amplify risk points in the logistics system, directly affecting the coordination of customs clearance, scheduling, and supervision, which highlights its strategic potential as a primary channel for transporting bioethanol and other temperature-controlled hazardous products.

Under the context of deepening trade cooperation between Russia and China, the transportation hub’s operational load continues to increase, especially with the growing demand for energy and hazardous products (such as bioethanol), and bottlenecks in customs clearance and warehousing have become increasingly prominent.

Currently, only a few ports (Zabaykalsk–Manzhouli, Blagoveshchensk–Heihe, and Pogranichny–Suifenhe) have Russian warehousing facilities [48], while most ports are constrained by limited infrastructure capabilities, frequently causing vehicle delays and customs clearance delays. The average customs clearance cycle has increased from one day to seven days. In addition, issues such as inconsistent equipment standards [49], inadequate multimodal transport, and fragmented information systems have exacerbated transportation uncertainties. To alleviate these bottlenecks, Russia and China are accelerating the construction of joint transportation hubs, with a focus on the Heilongjiang River basin and the Far East region, to create modern logistics hubs. These new hubs introduce mechanisms such as temperature control, fire prevention, and appointment-based inspections to enhance the efficiency and compliance of customs clearance for high-sensitivity goods, providing hardware support and institutional frameworks for the CB transportation of special goods, including bioethanol.

Under the BRI, infrastructure investments in border highways, dry ports, and digital customs systems have the potential to significantly improve multimodal connectivity and reduce transit times. Customs harmonization—through mutual recognition of inspection standards, pre-clearance protocols, and electronic data interchange (EDI)—can substantially reduce clearance uncertainties, as exemplified by the Manzhouli border crossing, where a regulatory system supported by institutional innovation and digital coordination is being developed that is capable of handling special cargo, such as bioethanol. As a UN Class 3 flammable liquid (UN1170) [50,51], bioethanol imposes higher requirements on temperature control, safety, inspection, and customs clearance during CB transportation. The Manzhouli Port is actively advancing the “Smart Port” initiative, introducing system innovations such as a 24 h appointment-based inspection mechanism [52], risk classification, and prior data review. These improvements are transforming the customs clearance model for hazardous goods from a manual approval process to system-driven, early-warning, and intelligent identification processes. The integration of cross-departmental joint inspection mechanisms with current digital platforms has already improved the supervisory efficiency and compliance of bioethanol and other controlled goods.

Building on the above institutional reforms and ongoing infrastructure upgrades at border crossings, the simulation framework translates improvements in specific KPIs into measurable system-level outcomes. For instance, a 10% reduction in border clearance time—achievable through smart port infrastructure and streamlined inspection facilities—corresponds to an approximate 5% increase in the composite path score $F\left(R_i\right)$, while a 20% reduction yields an improvement of about 9%. Similarly, highway modernization that reduces average transportation time by 15% results in a 7% efficiency gain, and dry port expansion, lowering queuing delays and logistics costs by 5% produces a 6% increase in the weighted score. These scenario-based results indicate that targeted investments in border infrastructure, supported by digital customs innovations, not only enhance individual process efficiency but also generate cumulative performance gains, thereby reinforcing the systemic value of BRI-related projects.

Building upon these advancements, further measures are recommended to enhance network efficiency:

• Expanding CB single-window systems to fully integrate documentation and inspection processes.
• Promoting trusted trader programs, allowing pre-certified carriers to bypass redundant inspections.
• Establishing joint emergency response protocols to mitigate congestion risks during sudden policy shifts or border disruptions.

Beyond bioethanol, the evaluation framework has the potential to support logistics planning for other renewable fuels. However, different fuels possess unique physical and regulatory characteristics that require targeted adjustments to ensure accurate assessment.

Table 7. Methodical adaptations of the evaluation framework for selected renewable energy sources.

Fuel Type	Key Differentiators vs. Bioethanol	Required Model Adaptations	Indicative Weight Changes *
Biodiesel	Higher viscosity; poor cold flow; non-flammable	Introduce CFPP indicator; emphasize low-temperature operability and handling cost	+Cost, +Temperature control; −Safety and Customs
Green Ammonia	Toxic, corrosive; moderate flammability	Add corrosion-resistant equipment and worker safety indicators; adjust inspection complexity	+Safety, +Compliance; −Distance
Compressed Hydrogen	High-pressure risk; explosive	Introduce tank integrity, leak detection, emergency response time KPIs; reclassify transport mode risks	+Safety and Emission; +Risk KPI weight
Methanol	Toxic, corrosive; lower flash point than ethanol	Add corrosion control and PPE cost KPIs; raise worker protection emphasis	+Safety; +Handling

* “+” indicates an increase and “−” indicates a decrease in relative weight compared with the baseline bioethanol model.

summarizes key differentiators and corresponding model adaptations, highlighting how the framework maintains structural stability while providing flexibility for diverse renewable fuels.

These examples demonstrate that while the core structure of the framework remains applicable, targeted methodical adaptations are required to accommodate the technical and regulatory features of each fuel type. Specifically, three categories of adaptations are essential:

• Reweighting existing KPIs to reflect differing hazard profiles and transport sensitivities.
• Adding or removing risk-specific sub-indicators (e.g., leak detection for hydrogen, corrosion resistance for methanol).
• Aligning compliance checks with applicable international regulations, including IMDG [53], ADR [54], and RID [55].

Importantly, the generalization to other renewable fuels is not confined to simple parameter adjustments but entails a methodical refinement of the evaluation framework. By embedding a dedicated Risk KPI ($H_i$) and incorporating both baseline-proportional and risk-enhanced weighting schemes, the model preserves its structural integrity while systematically capturing hazard-specific characteristics. This methodical extension ensures robustness under diverse hazard profiles and strengthens the framework’s applicability in supporting logistics planning across a broad spectrum of renewable energy products.

At the core of the model is a three-tier structure that synchronizes data exchange, financial operations, and physical transport. The information layer processes order inputs, demand signals, real-time tracking data, and simulation feedback to support the overall system. It functions as the analytical backbone, enabling ongoing route assessment as operational inputs evolve.

The financial layer manages payments, pricing dynamics, and the execution of digital contracts. Its integration ensures that cost variability, liquidity risks, and market volatility are reflected in the system’s route evaluation logic. This enables technical options to be weighed against economic feasibility in real-time.

The physical layer covers the actual flow of bioethanol across borders, capturing variables such as network topology, vehicle deployment, timing, and environmental impact. The physical layer grounds the simulation in real-world logistics activity by supplying empirical data on environmental metrics and service-level constraints.

The interaction between the three functional layers ensures alignment across all decision-making processes. Information from the physical domain is continuously fed back into the system’s analytical engine, where it shapes financial logic and data-driven coordination. This feedback loop allows the network to be restructured dynamically in response to external changes—whether regulatory, infrastructural, or market-related. As a result, the model becomes more applicable for real-time, sustainability-focused logistics management in transnational energy contexts.

Looking ahead, the model’s flexibility can be improved by integrating live operational data and predictive analytics. A deeper connection between AI modules and execution platforms could support a closed-loop approach to adaptive route selection and deployment planning. Additionally, multi-stakeholder preference modeling and policy scenario analysis can support broader applications in regional green transition strategies.

These findings highlight the broader implication that a route consistently robust under dynamic simulations is better positioned to withstand policy fluctuations, customs uncertainties, and environmental constraints. This underscores the strategic importance of resilience in CB logistics design, suggesting that decision-makers should prioritize not only cost or time minimization but also long-term adaptability when evaluating transport corridors.

5. Conclusions

This study developed an AI-driven simulation framework for designing an omnichannel logistics network to support CB bioethanol supply from Russia to China. The model integrates data exchange, financial coordination, and physical transport processes, and employs a combination of multi-criteria assessment, route prioritization using TOPSIS, and RL to determine the most effective logistics strategies under diverse operational scenarios.

Its three-layer structure enhances the model’s responsiveness to external uncertainty. Among the routes analyzed, the Kazakhstan–Kyrgyzstan hybrid corridor consistently delivers the most balanced performance in cost, transit time, emissions, and customs-related factors. These findings confirm that AI-enabled simulation can support routing decisions in complex, multi-objective environments, enabling flexible operational adjustments and long-term system planning.

These operational and structural advantages translate into positive sustainability impacts. Environmental benefits include reduced empty mileage and avoidance of unnecessary detours through the integration of route optimization with small-batch bioethanol transportation, thereby lowering transport-related CO₂ emissions compared with baseline bulk transport configurations. Economic benefits arise from the small-batch, multi-corridor strategy, which mitigates inventory risk, enhances supply chain flexibility, and supports gradual market entry without immediate large-scale infrastructure investment. Social benefits include promoting cleaner fuel use and fostering regional cooperation in renewable energy trade through the expansion of cross-border bioethanol supply.

The methodology is scalable and can be extended to other renewable energy supply chains or different geographical contexts involving infrastructure constraints and policy variability.