In modern naval warfare, the deployment of carrier-based drone formations has revolutionized anti-ship strike operations. As a researcher in this field, I have focused on addressing the critical challenge of target selection when a drone formation executes anti-ship missions. The limited number of drones that can be launched from an aircraft carrier necessitates a precise and efficient method to prioritize and select targets from a complex maritime target system. This paper presents a comprehensive model based on discrete Dynamic Bayesian Networks (DBN) to enhance target selection for carrier-based drone formations, ensuring that strikes are directed at the most impactful targets within the enemy’s maritime target architecture.
The concept of a target system is fundamental to understanding maritime warfare dynamics. I define a target system as an organic whole composed of individual targets (subsystems, elements, or parts) that interact through specific operational mechanisms to perform a certain function. Extending this, a target architecture comprises multiple target systems interacting to form a higher-level organic entity. This architecture exhibits multi-level and multi-faceted characteristics: vertically, elements have hierarchical relationships where lower-level elements form the foundation of higher-level ones; horizontally, elements at the same level can be divided into interrelated, constrained, and interacting parts. For drone formations, this means that striking a target is not just about destroying an individual unit but about affecting the entire system’s functionality and resilience.
Target selection, in the context of drone formation operations, involves identifying and prioritizing key targets within this architecture to achieve operational intent. Specifically, for carrier-based drone formations, it refers to the decision-making activity that determines the most necessary and suitable maritime targets for attack, based on the operational objectives, battlefield situation, available target information, and the performance of the drones and their weapon systems. The drone formation’s effectiveness hinges on this selection process, as it directly influences whether the strike can degrade or collapse the enemy’s target system.
The characteristics of target selection for carrier-based drone formations are distinct. First, the可选范围 is vast due to the unique advantages of drones in penetrating防空 threats where manned aircraft might be vulnerable. Second, high planning requirements are essential because drones, despite their advancements, have limitations in临机决策 compared to manned platforms; thus, target selection must be meticulously planned to reduce decision-making complexity. Third, predictive capability is crucial given the dynamic and uncertain maritime battlefield, where targets are highly mobile; the drone formation must anticipate target states to allocate time and space for effective engagement. Fourth, emphasis on key points is imperative since the limited number of deployable drones necessitates focusing strikes on critical nodes within the target architecture. These factors underscore the need for a robust, adaptive model that can handle uncertainty and prioritize targets effectively.
To address these challenges, I propose a three-level functional model for target selection in carrier-based drone formation cooperative anti-ship strikes: “Maritime Battlefield Situation Information Extraction – Target Selection Preprocessing – Target Selection Fine Processing.” This model structures the decision-making process into manageable stages, ensuring that the drone formation operates with enhanced situational awareness and precision.
The first level, Maritime Battlefield Situation Information Extraction, involves gathering and analyzing data from reconnaissance assets such as侦察机,预警机, and satellites. While not the primary focus of this paper, it provides the foundational input for subsequent stages. In practice, this stage yields a comprehensive situational picture of the maritime environment, including target positions, movements, and identities.
The second level, Target Selection Preprocessing, performs a qualitative analysis based on the situation set from the first level. It integrates domain expert knowledge, current battlefield information, and historical case libraries (e.g., target特征信息库 and target行为分析库). This stage roughly identifies primary operational海域 and generates an initial target list for further refinement. For example, by matching current data with historical patterns, the drone formation can filter out less critical targets and focus on those with higher潜在价值.
The third level, Target Selection Fine Processing, conducts real-time, precise situational reconnaissance on the preliminary target list and海域. Using discrete DBN methods, it evaluates and predicts target priority based on three core aspects: target value, target threat, and attack feasibility. The final target selection is made by applying predefined thresholds. This stage leverages the dynamic capabilities of DBN to incorporate temporal data and uncertainty, making it ideal for the evolving maritime battlefield where the drone formation must adapt quickly.
Discrete Dynamic Bayesian Networks (DBN) extend static Bayesian networks by incorporating time, allowing modeling of stochastic processes that evolve over time. A DBN consists of a prior network \(B_1\) for the initial time slice and a transition network \(B_{\rightarrow}\) for relationships between time slices. For a finite time period \(1, 2, \ldots, T\), the DBN can be unrolled into a network over variables \(X[1], X[2], \ldots, X[T]\). The conditional probability distribution at time \(t\) given the previous state is:
$$P(X_t | X_{t-1}) = \prod_{i=1}^{N} P(X_i^t | Pa(X_i^t))$$
where \(X_i^t\) represents the \(i\)-th variable at time \(t\), and \(Pa(X_i^t)\) denotes its parent nodes. This formulation enables reasoning under uncertainty by updating beliefs as new observations arrive. For the drone formation, this means continuously refining target priorities based on real-time data, enhancing the accuracy of selection despite incomplete or noisy information.
I constructed a discrete DBN model for target selection with a network structure that captures causal relationships among key factors. The model is divided into three sub-models: Target Value Analysis, Target Threat Analysis, and Target Attack Feasibility Analysis, which are integrated to determine the overall target selection priority.
The Target Value Analysis Model considers intrinsic value (Int) and system value (Sys). Intrinsic value is further broken down into economic value (Eco), political value (Pol), and psychological value (Psy). System value assesses the target’s role in the architecture: key node (Key), important node (Imp), or general node (Gen). The states for these variables are defined as high (H), medium (M), low (L) for value-related nodes, and {key, Imp, Gen} for system value. The joint probability influences the root node Target Value (Val) with states {H, M, L}.
The Target Threat Analysis Model evaluates threat based on target speed (Rate), course (Cou), and distance (Dis). Speed states are {H, M, L}; course states are {Toward, Parallel, Leave}; distance states are {Far, M, Close}. These feed into the Target Threat (Thr) node with states {H, M, L}. For instance, a target moving quickly toward the drone formation at close range would have a high threat probability.
The Target Attack Feasibility Model assesses how suitable a target is for drone strikes, considering target vulnerability (Vul), mobility (Man), and counterattack capability (Con). Vulnerability includes tactical vulnerability (ease of detection) and structural vulnerability (susceptibility to damage). States for these nodes are {H, M, L}, influencing the Attack Feasibility (Fea) node with states {H, M, L}. A target with low mobility and weak counterattack capability would be more feasible for the drone formation to engage.
Integrating these sub-models, I built the overall Target Selection Priority Model, where the root node Priority (Pri) has states {H, M, L}. The network structure encodes dependencies: for example, high target value and high threat contribute to high priority, but low attack feasibility might reduce it. The discrete DBN allows these relationships to evolve over time, capturing dynamic changes in the battlefield that affect the drone formation’s decision-making.
To operationalize the model, I defined Conditional Probability Tables (CPTs) based on expert knowledge, historical data, and simulation. For nodes without parents, initial state probabilities are assigned uniformly to avoid bias. For complex nodes, intermediate nodes are used to simplify CPTs. For instance, the CPT for Priority given Target Value and Target Threat might be defined as follows:
| Priority (Pri) | P(Val=H|Pri) | P(Val=M|Pri) | P(Val=L|Pri) | P(Thr=H|Pri) | P(Thr=M|Pri) | P(Thr=L|Pri) |
|---|---|---|---|---|---|---|
| High (H) | 0.80 | 0.20 | 0.00 | 0.90 | 0.10 | 0.00 |
| Medium (M) | 0.35 | 0.50 | 0.15 | 0.30 | 0.50 | 0.20 |
| Low (L) | 0.15 | 0.15 | 0.70 | 0.10 | 0.20 | 0.70 |
Transition probabilities between time slices for the Priority node are also defined, such as:
| P(Pri(t+1)|Pri(t)) | High | Medium | Low |
|---|---|---|---|
| High | 0.75 | 0.20 | 0.05 |
| Medium | 0.30 | 0.50 | 0.20 |
| Low | 0.06 | 0.14 | 0.80 |
These probabilities enable the DBN to update beliefs over time. For example, if a target has high priority at time \(t\), it is likely to remain high at \(t+1\), but uncertainties from battlefield changes can alter this. The drone formation uses these updates to adapt its target list dynamically.
To validate the model, I conducted a simulation using Netica, a Bayesian network analysis tool. The scenario involves a carrier-based drone formation facing eight maritime targets identified through reconnaissance: one destroyer, two frigates, one amphibious landing ship, two missile boats, and two supply ships. After preprocessing, the fine processing stage applies the DBN model for real-time priority assessment. For Target 1 (destroyer), input parameters over three time slices are shown in the table below, representing probability distributions for factors like economic value (e.g., (45%, 33%, 22%) for H, M, L states at time 1).
| Time Slice | Economic Value (Eco) | Political Value (Pol) | Psychological Value (Psy) | System Value (Sys) |
|---|---|---|---|---|
| Time 1 | (45, 33, 22) | (70, 20, 10) | (40, 30, 30) | (55, 23, 22) |
| Time 2 | (50, 33, 17) | (75, 15, 10) | (45, 30, 25) | (60, 28, 12) |
| Time 3 | (60, 30, 10) | (78, 15, 7) | (50, 30, 20) | (65, 30, 5) |
Feeding these into the DBN model, the inference results show that the probability of Target 1 having high priority increases over time: 33.3% at time 0 (initial uniform distribution), 49.8% at time 1, 64.9% at time 2, and 83.8% at time 3. This demonstrates the model’s ability to refine assessments as more data becomes available, crucial for the drone formation’s operational timing. Similarly, priorities for all targets are computed, leading to a sorted list. The table below summarizes the results, which can guide the final target selection for the drone formation.
| Target Name | Target ID | Probability of High Priority | Priority Rank |
|---|---|---|---|
| Destroyer | 1 | 83.8% | 2 |
| Frigate 1 | 2 | 81.5% | 4 |
| Frigate 2 | 3 | 82.4% | 3 |
| Amphibious Ship | 4 | 65.8% | 8 |
| Missile Boat 1 | 5 | 85.7% | 1 |
| Missile Boat 2 | 6 | 73.3% | 7 |
| Supply Ship 1 | 7 | 80.4% | 5 |
| Supply Ship 2 | 8 | 78.7% | 6 |
This simulation confirms the model’s effectiveness in handling dynamic maritime environments. Compared to traditional methods like expert scoring or fuzzy综合评价, the discrete DBN approach offers better real-time adaptability and reliability because it incorporates temporal dependencies and continuously updates probabilities based on incoming observations. For the drone formation, this means more accurate target selection under uncertainty, enhancing mission success rates.
In conclusion, the three-level functional model combined with discrete DBN provides a robust framework for target selection in carrier-based drone formation cooperative anti-ship strikes. By modeling target value, threat, and attack feasibility within a dynamic probabilistic network, the approach enables the drone formation to prioritize targets effectively, even in complex, evolving battlefields. Future work could focus on refining the model parameters through machine learning from historical data or expanding the network to include more factors like weather conditions or drone resource constraints. Overall, this research contributes to advancing autonomous decision-making for drone formations, ensuring they can execute precise strikes that maximize operational impact while minimizing risks. The integration of such models into real-time command systems will be key to leveraging the full potential of drone formations in naval warfare.