Task scheme generation system based on situation assessment

Abstract

The invention relates to the field of crime scheme generation, and particularly discloses a task scheme generation system based on situation assessment, which is characterized in that high-level task requirements and current situation key features are converted into formalized specification problems, and an integrated reasoning engine is combined to perform deep fusion and collaborative reasoning on a rule library and a case library, so that the task scheme generation efficiency is improved. The system can generate a scientific and feasible task scheme on the basis of fully understanding a complex battlefield environment. Compared with a traditional method, the scheme not only improves the intelligent level of task scheme generation, but also enhances the accuracy and adaptability of case matching through mechanisms such as semantic embedding coding and the like, improves the diversity and tactical value of the scheme, and is suitable for popularization and application. And the urgent demand of a commander on high-quality and multi-selection decision support in a complex and changeable battlefield environment is met, so that the scientificity, timeliness and reliability of task commanding are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of crime plan generation, and more specifically, to a task plan generation system based on situation assessment. Background Technology

[0002] Rapid and accurate situation assessment and the generation of scientifically feasible mission plans are crucial for gaining the initiative. Traditional plans rely on commanders' experience and are formulated manually, a cumbersome and inefficient process, susceptible to subjective influence, and unable to meet the demands of high-intensity, fast-paced environments. In the face of highly information-intensive environments, effectively extracting key situational information and supporting the intelligent generation of mission plans has become a core issue that urgently needs to be addressed. Therefore, building an intelligent system based on real-time situational assessment that can automatically generate diverse, high-quality mission plans is of great significance for improving the scientific rigor, timeliness, and reliability of mission command.

[0003] To meet the urgent needs of modern missions for intelligent decision-making systems, academia and industry have conducted relevant research. Early systems mostly adopted rule-based reasoning methods, generating solutions through pre-set mission instructions and expert knowledge. However, limited by the flexibility and completeness of the rule base, they struggled to cope with complex and ever-changing environments. Subsequently, case-based reasoning techniques were introduced, generating new solutions using historical cases. However, their effectiveness relies on a high-quality case library and precise retrieval and adjustment mechanisms, making matching difficult in unique situations, the adjustment process complex, and the solutions lacking innovation. To further improve the level of intelligence, some research has attempted to integrate rule-based and case-based reasoning to build hybrid reasoning systems. However, they still face challenges in deep integration of knowledge representation and reasoning collaboration, making it difficult to achieve true complementary advantages. In addition, existing systems have limited performance in terms of solution diversity, often outputting a single "optimal solution," while commanders often need multiple candidate solutions with distinct mission characteristics and significant differences to cope with uncertainty. Relying solely on parameter adjustment or randomization methods is insufficient to meet this requirement.

[0004] Therefore, how to build an intelligent system that can deeply integrate multi-source knowledge, fully understand the situation, and generate diverse task solutions remains a significant challenge for the field. Summary of the Invention

[0005] This application is made in order to solve the above-mentioned technical problems.

[0006] According to one aspect of this application, a task scheme generation system based on situation assessment is provided, comprising: a task requirement acquisition module for acquiring high-level task requirements input by a user; a current situation key feature acquisition module for acquiring current situation key features collected by a situation awareness module; an information transformation module for converting the high-level task requirements and the current situation key features into formalized normative problems; a preliminary task scheme generation module for inputting the formalized normative problems, a task rule base, and a task case base into an integrated inference engine to search for solutions that satisfy the formalized normative problems to obtain a preliminary task scheme skeleton; and a detail filling module for filling in the details of the preliminary task scheme skeleton to obtain detailed candidate task schemes.

[0007] Compared with existing technologies, this application provides a task plan generation system based on situation assessment. By integrating a reasoning engine and fusing a rule base and a case base, the system overcomes the limitations of single reasoning methods: semantic embedding encoding and dynamic field mapping technologies improve the accuracy of case retrieval and adaptability to unique situations, solving the problem of difficult matching in traditional case reasoning; the feasibility verification mechanism of the rule base compensates for the complexity of case adjustments, ensuring the rationality of the basic plan. Simultaneously, the system innovatively introduces a mutation-driven detail filling mechanism, generating differentiated details such as specific unit allocations and time nodes based on the initial plan skeleton, rather than relying on simple parameter adjustments, thereby outputting diverse candidate plans with distinct characteristics to meet the needs of multiple contingency plans in actual combat. Furthermore, by transforming high-level tasks and key real-time situational features such as enemy and friendly positions, terrain, and weather into structured inputs, the system supports the deep fusion of multi-source knowledge and a deep understanding of the situation, significantly improving the intelligence and timeliness of plan generation. Attached Figure Description

[0008] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 A block diagram of a task scheme generation system based on situation assessment according to an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the data flow of a task scheme generation system based on situation assessment according to an embodiment of this application.

[0011] Figure 3 A block diagram of the information conversion module in the task scheme generation system based on situation assessment according to the embodiments of this application is provided.

[0012] Figure 4 This is a block diagram of the preliminary task plan generation module in a task plan generation system based on situation assessment according to an embodiment of this application.

[0013] Figure 5 This is a schematic diagram of the data flow in the preliminary task plan generation module of the task plan generation system based on situation assessment according to an embodiment of this application. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0016] This application is made in response to the problems mentioned above in the background art. Figure 1 A block diagram of a task scheme generation system based on situation assessment according to an embodiment of this application. Figure 2 This is a data flow diagram illustrating a task scheme generation system based on situation assessment according to embodiments of this application. Specifically, as shown... Figure 1 and Figure 2 As shown, the task scheme generation system 100 based on situation assessment according to an embodiment of this application includes: a task requirement acquisition module 110, used to acquire high-level task requirements input by a user; a current situation key feature acquisition module 120, used to acquire current situation key features collected by a situation awareness module; an information conversion module 130, used to convert the high-level task requirements and the current situation key features into formalized normative problems; a preliminary task scheme generation module 140, used to input the formalized normative problems, a task rule base, and a task case base into an integrated inference engine to search for solutions that satisfy the formalized normative problems to obtain a preliminary task scheme skeleton; and a detail filling module 150, used to fill in the details of the preliminary task scheme skeleton to obtain detailed candidate task schemes.

[0017] Specifically, the task requirement acquisition module 110 is used to acquire high-level task requirements input by the user. In particular, in one example of this application, the high-level task requirements include the target area, time constraints, available troop types and quantities, prohibited areas, and effectiveness indicators. It should be understood that the generation of any task plan must revolve around clear strategic objectives and task intentions. High-level task requirements clearly define the expected final state, available resources, constraints to be followed, and success criteria for the system. To ensure that the situation assessment-based task plan generation system has clear guidance for targeted situation analysis to generate task plans that meet specific needs, it is necessary to acquire the high-level task requirements input by the user. Specifically, the target area, time constraints, available troop types, prohibited areas, and effectiveness indicators together constitute the basic framework of the task problem: the target area defines the spatial scope of the action; the time constraint specifies the urgency of the action; available troop types clarify the resources that can be allocated; prohibited areas mark the no-go zones; and effectiveness indicators provide a basis for subsequent plan evaluation and selection. Therefore, accurately acquiring and understanding these high-level mission requirements is the foundation and prerequisite for all subsequent intelligent processing, including information transformation, preliminary plan generation, and detail filling, ensuring that the final generated mission plan is consistent with the commander's strategic intentions.

[0018] In one possible example, the task requirement acquisition module 110 is implemented as follows: First, the system needs to provide a user interface that allows commanders or authorized users to input various task parameters. This interface can be graphical, containing designated input boxes, drop-down menus, map interaction tools, etc., so that users can clearly and accurately express their task intentions. The acquisition process is as follows: The input consists of various high-level task requirements entered by the user through the interactive interface. The system will prompt the user to input sequentially or in parallel: 1. Target area: The user may define this by plotting one or more areas on an electronic map (e.g., by drawing polygons, circles, or specifying existing geographic reference points), or by entering a precise geographic coordinate range. For example, the user inputs the target area as Hill A, with center coordinates of 118°30'E, 32°05'N, and a radius of 2 kilometers. 2. Time limit: The user inputs the allowed start and end times of the task, or the duration. For example, the user sets a time limit of 24 hours from the current moment, during which control of Hill A must be completed. 3. Available Troop Types and Quantities: Users select available troop units and their quantities from the system's preset troop database, or input them manually. For example, a user specifies available troops as 2 infantry battalions, 1 artillery battery, and 3 unmanned reconnaissance aircraft. 4. Restricted Areas: Users mark areas on the electronic map that must be avoided during mission operations, such as civilian areas, areas of important friendly facilities, or areas restricted by specific international law. For example, a user marks the residential area of ​​City B as a restricted area. 5. Performance Indicators: Users select or input key indicators for measuring mission success and their expected values. For example, a user sets performance indicators as successfully capturing Hill A, enemy manpower losses exceeding 70%, and friendly casualty rates below 10%. After the user completes and confirms all necessary information input, the system integrates this input information into a structured data object or record as the output of this step. For example, the output could be a set of mission requirement data containing all the specific values ​​and descriptions mentioned above.

[0019] Specifically, the current situation key feature acquisition module 120 is used to acquire the current situation key features collected by the situation awareness module. In particular, in one example of this application, the current situation key features include the location of key enemy units, terrain obstacles, weather conditions, and the current status of friendly forces. Correspondingly, any effective mission plan must be based on an accurate and comprehensive understanding of the current actual environment. That is, the situation is dynamic and changing, including factors such as the force deployment of both sides, geographical limitations, and weather conditions, which together constitute the specific constraints and opportunities for mission actions. Therefore, this step, by acquiring these preliminarily collected and processed key feature information from the situation awareness module, provides an objective and realistic basis for subsequent situation assessment and plan generation. Without this step, the generated mission plan will be detached from reality, not only difficult to execute but potentially leading to catastrophic consequences. Specifically, the location of key enemy units, terrain obstacles, weather conditions, and the current status of friendly forces in the example reveal the threat source, action constraints, environmental impact, and one's own capabilities, respectively, and are indispensable core information for decision-making.

[0020] In one possible example, the current situation key feature acquisition module 120 is implemented as follows: First, it initiates a data request to the situation awareness module (another component of the system or an external interface) or receives data pushed by it. The situation awareness module is responsible for collecting raw data through various sensors such as radar, satellites, UAVs, ground reconnaissance units, and information sources such as intelligence reports, meteorological services, and friendly / blue force tracking systems, and performs preliminary processing, fusion, and refinement to form structured key feature information. The data stream collected by the situation awareness module includes, for example: 1. Enemy key unit locations: The situation awareness module may provide a list containing multiple enemy unit records. Each record includes the unit ID, type (e.g., T-90 tank platoon, S-400 missile site), latest detected geographic coordinates (e.g., 118°25′10″E, 32°08′30″N), activity status (e.g., moving, silent), and information confidence level (e.g., 85%). 2. Terrain obstacles: The situation awareness module may provide a list containing terrain feature layers or obstacles. For example, identifying river section X (coordinates A to B) with a current water level 2 meters above average, making it unsuitable for wading; and mountain range Y (coordinates C to D) with an average slope exceeding 40 degrees, unfavorable for heavy equipment movement. 3. Weather conditions: The situational awareness module provides weather reports for the target area and its surroundings. For example, the weather forecast for the target area for the next 6 hours: cloudy, visibility 2 km, southwest wind force 3, 40% probability of showers. 4. Current status of friendly forces: The situational awareness module obtains information through reports from friendly units or the Blue Force Tracking System (BFT). For example, our 1st Battalion, located at coordinates E, has a 90% troop readiness rate, 70% ammunition load, and 60% fuel remaining; its current mission is to defend point Z. This information from the situational awareness module is then received, analyzed, and integrated. The system will obtain this data according to predefined interface protocols, such as API calls and message queue subscriptions, and store it in an internal unified data structure. The integrated data is the set of key features of the current situation. This set contains all the specific information obtained above, such as a structured data object containing a list of enemy units, a list of terrain obstacles, a weather report object, and a list of friendly forces status.

[0021] Specifically, the information transformation module 130 is used to transform the high-level task requirements and the key features of the current situation into formalized normative problems. It should be understood that the high-level task is described in natural language, while the situation features may come from heterogeneous data sources; this raw information is difficult to use directly by computer algorithms. Through formal transformation, this descriptive, unstructured information is mapped into mathematical or logical expressions with clear semantics and a unified structure. This eliminates the ambiguity of natural language, enabling the precise definition and quantification of complex factors such as task objectives, constraints, available resources, and the environment. This lays the necessary foundation for the subsequent integrated inference engine to perform rigorous logical deduction, search, and solution, and ultimately generate a preliminary task scheme skeleton, ensuring the logic and reliability of the entire generation process.

[0022] Specifically, in a particular example of this application, Figure 3 A block diagram of the information conversion module in the task scheme generation system based on situation assessment according to the embodiments of this application is provided. Figure 3 As shown, the information transformation module 130 includes: a description information extraction unit 131, used to extract an initial state description, a target state description, and a constraint description from the high-level task requirements and the key features of the current situation; and a formalized normative problem construction unit 132, used to construct the formalized normative problem based on the initial state description, the target state description, and the constraint description.

[0023] Accordingly, in order to further refine and structure the relatively macroscopic and diverse input information into the basic building blocks required for the subsequent formal normative problem construction, this application extracts initial state descriptions, target state descriptions, and constraint descriptions from the aforementioned high-level mission requirements and the key characteristics of the current situation. Specifically, the initial state description defines the starting point for problem solving, namely the baseline of the current battlefield and our forces; the target state description clarifies the expected final effect of the mission action; and the constraint description defines the rules and restrictions that must be followed during the action. This decomposition helps to clearly define the boundaries of the problem and the solution space, transforming complex mission requirements and battlefield environmental factors into logical elements that machines can understand and operate. Only by clearly separating and precisely expressing these elements can the formal normative problem construction unit effectively organize them into a complete and unambiguous normative problem representation that can be processed by the integrated inference engine, thereby ensuring the relevance and feasibility of subsequent mission solution generation.

[0024] In one possible example, the description information extraction unit 131 is implemented as follows: This unit first acquires the aforementioned high-level mission requirement data, including details such as the target being Hill A, its center coordinates being 118°30′E, 32°05′N, a radius of 2 kilometers, a time limit of 24 hours from the current moment requiring control of Hill A, available forces being 2 infantry battalions, 1 artillery battery, and 3 unmanned reconnaissance aircraft, a prohibited area being the residential area of ​​City B, and performance indicators such as successful capture of Hill A, enemy personnel losses exceeding 70%, and friendly casualty rates below 10%; simultaneously acquiring... The key characteristics of the current situation are taken, including, for example, the latest detected geographical coordinates of the enemy T-90 tank platoon are 118 degrees 25 minutes 10 seconds east longitude and 32 degrees 08 minutes 30 seconds north latitude, the activity status is moving, the information confidence level is 85%, the terrain obstacles are such as the X section of the river, the coordinate range is A to B, the current water level is 2 meters higher than the average and cannot be forded, the weather conditions are such as the weather in the target area is cloudy in the next 6 hours, the visibility is 2 kilometers, the southwest wind is level 3, and the current status of friendly forces is such as our 1st Battalion, located at coordinate E, the troop integrity rate is 90%, the ammunition load is 70%, the fuel is 60%, and the current mission is to defend point Z, etc.

[0025] Next, the input information is deeply analyzed and extracted based on a series of pre-defined parsing rules and semantic mapping logic. These rules are pre-defined according to military domain knowledge and mission planning requirements. For example, one rule may specify that the current status information of friendly forces in the key features of the current situation is mapped as a whole to the deployment of friendly forces in the initial status description, while also incorporating the types and quantities of available forces in the high-level mission requirements into the initial friendly resources; another rule may stipulate that the target area and some effectiveness indicators are combined to define the target status description; the rest, such as time constraints, prohibited areas, and some effectiveness indicators, are classified as constraint descriptions.

[0026] Then, through preset parsing rules and semantic understanding logic, the above input information is analyzed and extracted to generate: initial state description, such as: {current deployment of friendly forces (our 1st battalion, location coordinates E, troop readiness rate 90%, ammunition 70%, fuel 60%, mission defense point Z); our available resources (type infantry battalion, quantity 2); our available resources (type artillery company, quantity 1); our available resources (type unmanned reconnaissance aircraft, quantity 3); enemy situation (unit T-90 tank platoon, location [118°25′10″E, 32°08′30″N], status moving, confidence level 85%); battlefield environment terrain (X river segment, characteristic: cannot be waded through, reason: water level is 2 meters higher than average); battlefield environment weather (visibility 2 km, wind direction: southwest wind level 3)}. The target state is described, such as: {Core mission objective (achieve control of Hill A, geographical range [center coordinates 118°30' E, 32°05' N, radius 2 km]); Expected mission outcome (enemy manpower loss, greater than 70%)}. Constraints are described, such as: {Mission time limit (total mission duration, less than or equal to 24 hours); Operation area restrictions (prohibited areas, residential areas of City B); Acceptable friendly casualties (casualty rate, less than 10%)}. These outputs will serve as inputs to the formalized normative problem construction unit in the next stage.

[0027] It can be understood that the structured semantic information extracted in the previous stage is integrated and transformed into a standardized, rigorous mathematical or logical form that can be directly processed by a computer inference engine. Although the initial state, target state, and constraint descriptions are relatively clear, they are still separate descriptive fragments. The role of the formalized problem construction unit is to fuse these fragments into a unified, well-defined problem framework. This transformation is key to achieving automated solution generation. It provides an unambiguous input with strict syntax and semantics for the subsequent integrated inference engine, enabling the engine to use its internal search algorithms, rule-based reasoning, or case-matching mechanisms to explore feasible action sequences within this formalized problem space—that is, the initial task solution skeleton.

[0028] In one possible example, the formal specification problem construction unit 132 is implemented as follows: the process of this unit is to map these descriptions into a predefined formal problem representation framework. For example, if a classic planning problem representation (such as a PDDL-like structure) is adopted, the construction process may be as follows: First, the domain of the problem is defined as pre-defined, including action types (such as movement, attack, reconnaissance) and their preconditions and effects. Here, a problem instance is constructed. Second, each atomic fact in the initial state description is transformed into the initial state (Init) part of the formal problem instance. For example, the existing deployment of friendly forces (our Battalion 1, location coordinates E,...) may be transformed into a series of initial predicates such as (at unit1 loc_E), (type unit1 infantry_battalion), (readiness unit1 90). Our available resources (type infantry battalion, quantity 2) may be transformed into (=(available_infantry_battalions) 2).

[0029] Then, each objective condition in the objective state description is transformed into the goal part of the formalized problem instance. For example, the core objective (achieving control of Height A,...) might be transformed into (is_controlled loc_A_high_ground). The expected objective outcome (enemy manpower loss, greater than 70%) might be transformed into (>=(enemy_loss_percentage) 70).

[0030] Finally, the time limit, regional prohibition and cost acceptance range in the constraint description are incorporated into the problem definition as global constraints, such as (<= total_mission-time 24) and (not (is_in_forbidden_zone B_city_residential)).

[0031] The final output formal specification problem is a plain text file containing a structured description that follows the syntax of a specific formal language (such as the text style of PDDL).

[0032] Specifically, the preliminary task solution generation module 140 is used to input the formalized normative problem, the task rule base, and the task case base into the integrated inference engine to search for solutions that satisfy the formalized normative problem to obtain a preliminary task solution skeleton. Accordingly, in order to combine the clearly defined problem (formalized normative problem) with available knowledge (task rule base and case base), this application uses an intelligent inference mechanism (integrated inference engine) to solve the problem and generate a preliminary solution that meets the task requirements. That is, the formalized normative problem provides the engine with a clear solution objective, initial conditions, and boundary constraints. The task rule base contains universally applicable task principles and tactical logic, providing a foundation for constructing compliant action sequences. The task case base supplements valuable historical experience and proven effective action patterns, facilitating heuristic search and rapid generation of high-quality solution fragments. Through systematic searching, rule deduction, and case reference, the integrated inference engine searches for a macroscopic action sequence within the problem space defined by the formalized normative problem that can reach the target state from the initial state and satisfy all constraints. This sequence is the preliminary task solution skeleton, a necessary prerequisite for subsequent refinement work.

[0033] Figure 4 This is a block diagram of the preliminary task plan generation module in a task plan generation system based on situation assessment according to an embodiment of this application. Figure 5 This is a schematic diagram of the data flow in the preliminary task plan generation module of the task plan generation system based on situation assessment, according to an embodiment of this application. Figure 4 and Figure 5 As shown, the preliminary task solution generation module 140 includes: a similar case fragment retrieval unit 141, used to retrieve similar case fragments from the task case library using the integrated reasoning engine based on the initial state description in the formalized normative problem; a preliminary solution skeleton generation unit 142, used to integrate the similar case fragments into the current solution construction path after adaptive processing to obtain a preliminary solution skeleton prototype; and a verification unit 143, used to perform feasibility verification on the preliminary solution skeleton prototype using the integrated reasoning engine based on the task rule library, and determine the preliminary solution skeleton prototype as the preliminary task solution skeleton after the feasibility verification is passed.

[0034] It is understandable that the initial state description in formalized normative problems precisely depicts the current battlefield starting point. By matching this initial state with the starting conditions of a large number of historical cases or tactical fragments stored in the mission case library, the integrated inference engine can quickly locate the past situation most similar to the current situation. This retrieval of similar cases allows the system to draw inspiration from proven or valuable experiences, extracting reusable action sequences or tactical patterns. This not only significantly improves the efficiency of solution generation, avoiding a global search from scratch, but also introduces combat-tested wisdom, providing a solid foundation and effective shortcut for generating high-quality, insightful preliminary mission solution frameworks, especially providing valuable heuristic guidance when dealing with complex and uncertain battlefield environments.

[0035] Specifically, in a specific example of this application, the similar case fragment retrieval unit 141 includes: an initial state semantic encoding subunit, used to perform semantic embedding encoding on the initial state description to obtain an initial state semantic embedding encoding vector; a case fragment semantic encoding subunit, used to perform semantic embedding encoding vectors on each case fragment in the task case library to obtain a set of case fragment semantic embedding encoding vectors; a semantic similarity calculation subunit, used to calculate the semantic similarity between the initial state semantic embedding encoding vector and each case fragment semantic embedding encoding vector in the set of case fragment semantic embedding encoding vectors; and a similar case fragment selection subunit, used to select the case fragment corresponding to the one with the largest semantic similarity as the similar case fragment.

[0036] Correspondingly, raw initial state descriptions (whether natural language fragments or structured data) are not suitable for direct large-scale, deep similarity matching. Semantic embedding encoding techniques, such as those using deep learning models, can map high-dimensional, sparse text information into a low-dimensional, dense vector space, making semantically similar initial states also close in distance within the vector space. This allows for the capture of deep semantic relationships behind words, phrases, and even entire descriptions. By generating semantic embedding encoding vectors for initial states, a foundation is laid for accurate and efficient semantic similarity calculations with case fragments in the task case library, enabling more intelligent retrieval of truly relevant historical experiences.

[0037] In one possible example, the initial state semantic encoding subunit is implemented as follows: First, it takes structured text describing the initial state presented in the formal specification problem as input. This means that even if the initial state description in the formal specification problem is stored internally in a more structured form (e.g., key-value pairs, JSON fragments), it is transformed or serialized into a natural language text sequence through a pre-defined strategy before being fed into a pre-trained deep learning semantic encoding model, which may be fine-tuned for the military domain. The training of this model typically includes two stages: First, pre-training, using large-scale general text corpora (e.g., Wikipedia, news articles) to teach the model universal language representations through self-supervised tasks (e.g., masked language models predicting masked words, or next-sentence prediction judging sentence coherence). Subsequently, to meet the need for generating high-quality sentence-level embeddings, such as the Sentence-BERT model, a specific training objective is adopted, such as using Siamese networks or triplet network structures, employing contrastive learning methods to minimize the distance between semantically similar sentence pairs in the vector space and maximize the distance between dissimilar sentence pairs. Finally, to adapt to specific applications, fine-tuning is performed using data such as professional military literature, mission regulations, and historical cases. Supervised learning or further contrastive learning enhances the model's understanding and encoding accuracy of military terminology, tactical concepts, and complex situations, optimizing its performance in that specific domain. Afterward, the trained Transformer-based architecture, such as Sentence-BERT, is used for processing. Its core architecture includes: an input embedding layer, which segments the text (e.g., using WordPiece), adding its corresponding word embedding and position embedding to each word to generate an initial word vector; a multi-layer Transformer encoder, each layer consisting of a multi-head self-attention mechanism sublayer and a feedforward neural network sublayer; the multi-head self-attention mechanism enables the model to capture the dependencies between different words within the text sequence, assigning different weights to different words; the feedforward network performs a non-linear transformation on the representation of each position; and residual connections and layer normalization exist between layers. Pooling layers are necessary because the Transformer encoder generates a context-dependent vector for each input word. A pooling operation is needed to aggregate the outputs of all words into a fixed-size sentence (or paragraph)-level vector. For example, Sentence-BERT often uses average pooling to obtain the semantic representation of the entire input text. Ultimately, the model's output is a fixed-dimensional dense vector, which is the initial state semantic embedding encoding vector.

[0038] It is understandable that the task case library stores a large number of fragments describing past battles, engagements, or tactical operations, which contain valuable practical wisdom. To enable the computer to understand the deeper meaning of these case fragments and make meaningful comparisons with the initial state description that needs to be addressed, these case fragments, existing in text or structured form, need to be converted into numerical vector representations within the same semantic space as the initial state description. By generating a semantic embedding encoding vector for each fragment in the case library, the system constructs a vectorized knowledge base of experience. This allows for the rapid and accurate identification of historical case fragments that are semantically most similar to the current situation by calculating the distance between vectors, thus providing powerful reference and inspiration for generating task solutions.

[0039] In one possible example, the semantic encoding subunit for case fragments is implemented as follows: The task case library is a pre-built database, where each case fragment may contain a description of a historical task scenario, such as enemy and friendly troop deployments, terrain features, tactical action sequences and their results, etc. This information may be stored in structured or semi-structured text. This subunit processes each case fragment in the case library one by one. For each case fragment, text preprocessing is first performed, integrating its key information into a coherent text sequence. For example, a case fragment may describe: In the early stages of Campaign A, our X Division encountered the enemy's Z Armored Brigade holding a stalemate in area Y. The terrain was hilly, and the weather was clear. Our X Division adopted a tactic of frontal containment and flanking maneuvers... This description will be organized into text suitable for model input.

[0040] Subsequently, this subunit encodes the text sequence of each case segment using the same or compatible pre-trained semantic embedding model (e.g., the Sentence-BERT model based on the Transformer architecture) as the initial state description. This model architecture also includes: an input embedding layer that segments the text and generates initial word vectors containing word embeddings, position embeddings, and segment embeddings; a multi-layer Transformer encoder that captures deep semantic dependencies of the text through multi-head self-attention and a feedforward neural network; and a pooling layer that aggregates the multiple word vectors output by the encoder into a fixed-dimensional vector representing the semantics of the entire case segment. For example, average pooling can be used. For each case segment in the case library, after processing by the above model, a dense vector of fixed dimensions is output; this vector is the case segment semantic embedding encoding vector. Finally, the output of this step is the set of these individual vectors, i.e., the set of case segment semantic embedding encoding vectors, providing the data foundation for subsequent similarity calculations.

[0041] Accordingly, to assess the correlation between the current initial battlefield situation and the numerous historical experiences stored in the mission case library, it is necessary to calculate the semantic similarity between the initial state semantic embedding encoding vector and the semantic embedding encoding vectors of each case fragment. That is, by converting the complex initial state description and each case fragment into vector representations within a unified semantic space, calculating their semantic similarity transforms this abstract similarity into a concrete numerical value. This value objectively reflects the degree of semantic closeness between each historical case fragment and the current problem. This provides a direct basis for accurately and efficiently selecting the most valuable experiences from a massive database of cases.

[0042] Specifically, as the foundation for constructing the task scheme, the preliminary scheme skeleton is obtained by fusing retrieved similar case fragments into the current scheme construction path. However, since the initial state description used as a retrieval reference belongs to the configuration of a formalized normative problem, there will be certain semantic domain differences between it and the similar case fragments used as retrieval targets. Furthermore, considering the fundamental nature of the retrieved similar case fragments, the initial state semantic embedding encoding vector obtained by semantic embedding encoding the initial state description actually needs to serve as a latent index field, while the case fragment semantic embedding encoding vector serves as a strengthened semantic domain. Therefore, it is necessary to reconstruct the initial state semantic embedding encoding vector and the case fragment semantic embedding encoding vector obtained through semantic embedding encoding based on their respective attributes.

[0043] Based on this, in a specific example of this application, the semantic similarity calculation subunit is used to: obtain a transformation matrix based on the initial state semantic embedding encoding vector and the case fragment semantic embedding encoding vector, i.e.: ,in, This is the initial state semantic embedding encoding vector. For the semantic embedding encoding vector of the case fragment, It is matrix multiplication. It is the transformation matrix, that is, setting the initial state semantic embedding encoding vector as... And the semantic embedding encoding vector of the case fragment is This requires establishing an implicit projection interaction unification from the latent index field to the enhanced semantic state domain. That is, the local non-uniform mapping patterns caused by projection into the nonlinear state domain are subject to continuous diffusion constraints. Thus, the transformation matrix is ​​first obtained; the transformation matrix is ​​then constrained by local nonlinear transformations to generate a continuous diffusion analytic matrix, i.e.: , ,in, To calculate the L2 norm of a vector, It is a scaling factor. It is the value of an exponential function with the natural constant e as its base. It is a dot product by position. It is added based on the position point. It is a continuous diffusion analytic matrix, that is, through... and The transformation matrix is ​​subjected to a low-rank nonlinear constraint on its 2-norm. The local characteristic distribution is subjected to continuous diffusion; the continuous diffusion analytical matrix is ​​dynamically mapped onto the transformation matrix to obtain the dynamic mapping matrix, i.e.: ,in, This is the mapping constraint factor, a hyperparameter that is pre-set based on experience or determined through experimental tuning. It aims to optimize the effect of dynamic field mapping and ensure that local identity constraints are satisfied. For example, 0.1 is used here only as an example. yes The inverse matrix, It is a dynamic mapping matrix, so that the transformation matrix can be applied... While performing continuous diffusion analysis, local conservation constraints are achieved by simultaneously mapping based on its inverse matrix; based on the dynamic mapping matrix, the semantic field state of the case fragment semantic embedding encoding vector is adjusted to obtain the case fragment semantic embedding enhanced encoding vector, i.e.: ,in, It is a semantic embedding enhancement encoding vector for case fragments, thus enabling the semantic embedding encoding vector in the initial state to be implemented. In the case of a latent index field, the semantic embedding encoding vector of the case fragment serves as the field state transmission in the enhanced semantic state domain. This is achieved through continuous diffusion resolution to realize implicit projection interaction transmission, while simultaneously achieving mapping homogenization based on local conservation constraints. This improves the accuracy of semantic similarity calculation between the initial state semantic embedding encoding vector and the case fragment semantic embedding encoding vector under their respective attribute affiliations. The semantic similarity between the initial state semantic embedding encoding vector and the enhanced semantic embedding encoding vector of the case fragment is calculated using cosine similarity. ,in, yes The Middle 1 eigenvalue, yes The Middle 1 eigenvalue, yes and The number of eigenvalues, This refers to semantic similarity. In other words, this step uses the cosine similarity formula to calculate the complex semantic relationship between the initial state semantic embedding encoding vector and the case fragment semantic embedding reinforcement encoding vector obtained after semantic field adjustment, transforming it into a single scalar value ranging from -1 to 1 (or 0 to 1 if the vector is non-negative). .Should The value represents the semantic similarity between the two, and its direct effect is to provide the system with a clear and comparable metric. This metric can accurately measure the degree of semantic matching between the current initial state and each potentially related historical case fragment, thereby supporting the subsequent similar case fragment selection subunit to sort and filter cases based on this similarity k value, and finally select the case fragment most relevant to the current situation.

[0044] Accordingly, semantic similarity calculation provides a quantitative indicator of the correlation between each case fragment and the current initial state. Therefore, in order to accurately select the single case that best fits the current battlefield initial situation from a large amount of historical experience, this application uses the case fragment corresponding to the highest semantic similarity score as the similar case fragment. By selecting the case fragment with the highest similarity score, the system can ensure that the extracted historical experience has the highest consistency and relevance to the current problem at the semantic level. This provides a high-quality, high-matching starting point for the heuristic generation of subsequent task plans or the construction of plan frameworks, thereby improving the efficiency and effectiveness of plan generation and ensuring that the experience referenced is the most relevant.

[0045] In one possible example, the similar case segment selection subunit is implemented as follows: First, the input is a set of data, where each pair contains a case segment and a corresponding semantic similarity score k. For example, the input might be: {(case segment A, similarity 0.85), (case segment B, similarity 0.92), (case segment C, similarity 0.78)}. This subunit iterates through these similarity scores, finding the maximum value through comparison operations. For example, it first considers 0.85 (corresponding to case segment A) as the current maximum value, then compares it to 0.92 (corresponding to case segment B), finding that 0.92 is greater than 0.85, so it updates 0.92 to the maximum value and considers case segment B as the current best match. Next, it compares 0.78 (corresponding to case segment C), and since 0.78 is less than 0.92, the maximum value and the corresponding case segment remain unchanged. After completing all comparisons, the output is the case segment with the maximum semantic similarity score, which in this example is 0.92, i.e., case segment B. This output similar case segment is the historical experience that the system considers most similar to the current initial state.

[0046] It is understandable that while directly retrieved similar case fragments are highly relevant, they often fail to perfectly match the specific details of the current mission, such as time, location, available forces, or precise sub-targets. Therefore, to transform purely historical experience into usable tactical components for the current specific mission and effectively integrate them into the mission plan being built, adaptive processing is required. This involves adjusting case elements to adapt to the current situation, ensuring immediate usability. The fusion operation involves embedding this adjusted, instructive tactical template or action sequence into the logical chain of the mission plan, which is progressively deduced by the integrated inference engine based on formalized normative problems. This significantly accelerates the planning process, avoiding the need to design all tactical details from scratch. By referencing and modifying proven patterns, a preliminary framework containing core action steps and structure can be quickly formed, laying a solid foundation for subsequent refinement, evaluation, and optimization.

[0047] In one possible example, the preliminary scheme skeleton generation unit 142 is implemented as follows: First, adaptive processing is performed. The input to this stage is a similar case fragment from the previous step. If the selected similar case fragment describes a historical mission action: Historical Battalion A attacked enemy Unit R at location P using tactic Q, taking S hours and achieving effect T. The adaptive processing unit will refer to the initial state description of the current mission, such as: Our Battalion 1 is located at coordinate E, with a troop readiness rate of 90%... 2 infantry battalions, 1 artillery battery, and 3 unmanned reconnaissance aircraft are available. The enemy T-90 tank platoon is located... The X section of the battlefield is uncrossable due to the water level being 2 meters higher than the average... Weather visibility is 2 kilometers, southwest wind force 3, as well as the target state description and constraint description. It will adjust the elements of the original case fragment according to preset mapping rules and heuristic logic. For example, if the currently available unit is Our Battalion 1, then the historical Battalion A in the case will be replaced with Our Battalion 1; if the target location is M, then location P will be adjusted to M. If the original tactic Q involves fording the Y river, but the current battlefield's X river section is impassable, the tactic will be modified. This might involve finding an alternative route or requesting bridge support, or, without suitable adjustments, reducing the tactic's applicability. These adjustment rules are pre-defined. For example, a rule could be set: if the original terrain is plains and the current terrain is mountains, adjust the movement speed parameter to 60% of its original value, prioritizing tactical variations suitable for mountain missions. The processed output is an adjusted similar case snippet where key elements are matched to the current mission context.

[0048] Next, the fusion process is performed. The current solution construction path refers to a high-level or mid-level task action sequence or plan framework generated by the integrated inference engine based on a formal problem, through task decomposition or preliminary planning. For example, it might be a list containing main task phases: [Phase 1: Battlefield Reconnaissance] -> [Phase 2: Firepower Preparation] -> [Phase 3: Main Attack - Occupy Target Area] -> [Phase 4: Consolidate Defense]. The fusion module analyzes which link or sub-objective in the current solution construction path the adjusted similar case fragment is best suited to achieve. For example, if the adjusted similar case fragment specifically describes a complete attack operation, it might be used to fill in [Phase 3: Main Attack - Occupy Target Area] in detail. The fusion operation inserts the action steps, resource allocation, and temporal relationships in the adjusted case as specific tactical templates for achieving the objectives of that phase into the path. The output is the initial framework of the plan, which is a more specific and realistic mission plan framework. For example: [Phase 1: Order 3 drones to conduct detailed reconnaissance of the target area and the enemy T-90 tank platoon] -> [Phase 2: The artillery company conducts precision strikes on the identified threats and the forward defense of the target based on the reconnaissance results] -> [Phase 3: Our 1st Battalion (90% of the force) refers to the adjusted case fragment tactics (such as frontal containment and flank attack sequence) to carry out the attack on the target area...] -> [Phase 4: After the 1st Battalion occupies the area, it uses the remaining force to organize a ring defense and reports the consumption of ammunition and fuel.]

[0049] Accordingly, although the preliminary scheme framework incorporates experience from similar cases, its adaptability to the current specific situation, the coordination between internal actions, and the degree to which it meets various constraints all need to be verified through rigorous rule-based checks. Therefore, to ensure that the preliminary scheme generated from case lessons conforms to military logic, resource constraints, and mission principles at a fundamental level, this application conducts a feasibility check on the aforementioned preliminary scheme framework framework. This check aims to filter out scheme fragments that are clearly infeasible or logically flawed, providing a solid and reliable foundation for subsequent detailed planning or decision-making.

[0050] In one possible example, the verification unit 143 is implemented as follows: The integrated inference engine loads a task rule base. This rule base is a predefined, structured collection of knowledge, containing general mission principles, tactical regulations, equipment performance parameters, resource consumption standards, preconditions and effects of actions, and specific constraints such as prohibited action areas and time limits. For example, the task rule base may contain the following rules: R001: For a single UAV reconnaissance mission, if the target distance is greater than 50 kilometers, at least one medium-range UAV is required, with a fuel consumption of 5 units per hour and a minimum guaranteed flight time of 2 hours; R002: When an infantry battalion performs a strong attack mission, the hourly ammunition consumption shall not be less than 15% of the basic ammunition; R003: The total duration of the mission operation shall not exceed the user-set 24-hour limit; R004: No action may enter the preset residential area of ​​City B. The values ​​of these rules (such as 50 kilometers, 15%, 24 hours) are preset based on military regulations or expert experience.

[0051] The integrated inference engine analyzes each stage and action in the initial plan's skeletal framework step by step. For Stage 1, UAV reconnaissance, the engine checks resource availability (3 UAVs > 1, satisfied) based on the initial state description and rule R001. It then assesses fuel consumption and flight time based on the target location and our deployment location, estimating distance if known, and whether these are within the range of fuel carried by Battalion 1 or available from the base. For Stage 3, our Battalion 1's attack, the engine refers to the initial state's 70% ammunition reserve. Based on rule R002 and the estimated attack duration (e.g., 2 hours), it calculates the estimated ammunition consumption: 2 hours × 15% per hour = 30%. Therefore, the remaining ammunition is 70% - 30% = 40%, which is higher than the minimum mission reserve requirement (e.g., greater than 0%), thus passing this step. Simultaneously, the engine checks whether the total estimated time for all stages satisfies rule R003 and whether all action paths violate rule R004.

[0052] If all actions in the preliminary scheme skeleton meet the constraints defined in the task rule base, or if the deviation from the rules is within a preset tolerance threshold (e.g., the rule base allows for a 5% fluctuation in certain resource consumption), then the feasibility check is passed. At this point, the preliminary scheme skeleton is officially confirmed as the preliminary task scheme skeleton. Its output content is consistent with the input skeleton in structure and text, but its internal state has changed from a prototype to a verified preliminary skeleton, possessing higher credibility and priority for subsequent processing.

[0053] Specifically, the detail filling module 150 is used to fill in the details of the preliminary task plan skeleton to obtain detailed candidate task plans. It should be understood that while the preliminary plan skeleton indicates the main task phases and core actions, it lacks the precise details required at the execution level, such as unit assignments down to the single digit, time arrangements down to the minute, inter-unit coordination mechanisms, and specific targets and timing of fire support. To transform the feasibility-verified preliminary task plan skeleton from a relatively macro-level action framework into a specific, operational task plan with clearly defined elements, detail filling is necessary to ensure that each stage of the plan has clearly defined responsible units, time nodes, and coordination requirements, thereby enabling the plan to be further evaluated, extrapolated, and ultimately executed.

[0054] Specifically, in a specific example of this application, the detail filling module is used to: identify the variation points in the preliminary task plan skeleton; and fill in the details of each variation point to obtain the detailed candidate task plan, wherein the details include specific unit allocation, precise time nodes, coordinated actions, and firepower allocation.

[0055] In one possible example, the detail filling module 150 is implemented as follows: First, identify the variation points in the preliminary mission plan skeleton. Variation points refer to the parts of the plan skeleton that still need to be refined and specifically determined. In the above example, ordering three UAVs is a variation point, which three need to be specified, for example, selected from the UAV-001, UAV-002, and UAV-003 available in the initial situation, and the specific reconnaissance route, altitude, and start and end time. Similarly, the artillery company's precision strike based on the reconnaissance results is also a variation point, which needs to be specified, and if there are multiple artillery companies, one should be designated. Here, it is the specific target number, coordinates, ammunition type, firing data, and firing window for the artillery company's Alpha strike. Our 1st Battalion... executing... the frontal containment and flank attack sequence in the attack also needs to be broken down into the specific tasks, attack directions, time coordination points, and fire support requirements of each subordinate company and platoon.

[0056] Then, the details of each mutation point are filled in. For Phase 1, the system will specifically allocate UAVs from available resources, for example, assigning UAV-001, UAV-002, and UAV-003 to reconnaissance missions, and setting precise time nodes, such as reconnaissance from H-2 hours to H-1 hours, and outputting specific reconnaissance areas and lists of targets of interest. For Phase 2, according to preset firepower allocation rules, such as prioritizing high-threat targets, concentrating firepower on obstacles in the main attack direction and weapon performance in the mission rule library, the mission of Artillery Battery Alpha will be refined as follows: from H-30 minutes to H-10 minutes, fire 15 armor-piercing rounds at the enemy T-90 tank platoon position (coordinate G1) discovered by reconnaissance, and fire 30 high-explosive rounds at point M (coordinate G2) on the target's defensive front. For Phase 3, the actions of our 1st Battalion will be detailed as follows: 1st Company (specifically assigned from within the 1st Battalion) will launch a feint attack in the Y1 direction at time H; 2nd and 3rd Companies will launch the main attack in the Y2 direction on the flank at time H+5. Coordination will stipulate that the attack by 2nd and 3rd Companies will be signaled by 1st Company drawing enemy main fire, or will be launched on time after artillery fire has completed its designated suppression. Fire support will also be specified, such as Artillery Company Alpha suppressing the enemy's pre-positioned blocking position G3 in the Y2 direction from time H+3 to time H+15.

[0057] The final output is a detailed candidate mission plan, a more complete sequence of instructions that includes all the details mentioned above. For example: [Start time H-2:00, End time H-1:00, Units UAV-001, UAV-002, UAV-003, Mission: Conduct continuous reconnaissance of area R1 and the position of the enemy T-90 tank platoon, and report to the command post] -> [Start time H-0:30, End time H-0:10, Units Artillery Battery Alpha, Targets G1, G2, Mission: Target G1 [Fire 15 armor-piercing shells, fire 30 high-explosive shells at G2]->[Phase 3.1: Start time H:00, Unit 1, Battalion 1, Company 1, Mission: Conduct a feint attack on the enemy along the Y1 direction to attract the enemy's attention; Phase 3.2: Start time H+0:05, Unit 1, Battalion 2, Company 3, Coordination conditions: Feint attack successful or artillery preparation ends, Mission: Launch the main attack on the target area along the Y2 direction, Artillery Company Alpha suppresses G3 from H+0:03 to H+0:15...]->[Phase 4:...]

[0058] In summary, the situation assessment-based task plan generation system 100 based on the embodiments of this application is explained. By integrating a reasoning engine and fusing a rule base and a case base, the system overcomes the limitations of a single reasoning method: semantic embedding encoding and dynamic field mapping technologies improve the accuracy of case retrieval and its adaptability to unique situations, solving the problem of difficult matching in traditional case reasoning; the feasibility verification mechanism of the rule base compensates for the complexity of case adjustments, ensuring the rationality of the basic plan. Simultaneously, the system innovatively introduces a mutation-driven detail filling mechanism, generating differentiated details such as specific unit allocations and time nodes based on the initial plan skeleton, rather than relying on simple parameter adjustments, thereby outputting diverse candidate plans with distinct tactical characteristics to meet the needs of multiple contingency plans in actual combat. Furthermore, by transforming high-level tasks and key real-time situational features such as enemy and friendly positions, terrain, and weather into structured inputs, the system supports the deep fusion of multi-source knowledge and a deep understanding of the situation, significantly improving the intelligence and timeliness of plan generation.

[0059] Specifically, in another specific example of this application, a task scheme generation system based on situation assessment is proposed, comprising: a situation awareness module for collecting and processing multi-source heterogeneous data such as reconnaissance and surveillance, electronic countermeasures, and cyber intelligence to construct a unified situation map; a task planning module for establishing a task dependency network, determining constraints and performance indicators, and allocating tasks and scheduling resources for cross-domain collaborative tasks based on task requirements and available resources; a scheme generation module for generating multiple candidate task schemes using a combination of rule-based reasoning and case-based reasoning; an evaluation and adjustment module for designing a dual-channel feedback mechanism and establishing a scheme iteration log; and a human-computer interaction module for automatically writing task commands, collaborative plans, and other documents based on natural language generation technology, and adjusting and confirming them through an interactive interface to form the final task scheme.

[0060] Specifically, the situational awareness module fully integrates various reconnaissance methods and technologies during implementation to achieve efficient collection and processing of multi-source heterogeneous data from reconnaissance and surveillance, electronic warfare, and cyber intelligence. By deploying various types of reconnaissance equipment, such as radar, drones, and sensors, it conducts comprehensive and multi-layered reconnaissance and surveillance of the battlefield environment, acquiring key information such as enemy troop deployment, equipment configuration, and movement trajectories in real time. In the field of electronic warfare, it intercepts enemy electromagnetic signals using electronic reconnaissance equipment, analyzes their spectrum characteristics, communication content, and jamming patterns, and grasps the enemy's electronic warfare capabilities and intentions. For cyber intelligence, it uses network monitoring technology and data mining algorithms to collect and analyze information in cyberspace, including signs of enemy network dynamics, intelligence transmission, and hacker attacks. After collecting this multi-source heterogeneous data, it uses data fusion technology, combined with analysis of enemy and friendly intentions and judgment of current battlefield actions, to filter, calibrate, correlate, and comprehensively process the data, removing redundant and contradictory information and extracting valuable content. Finally, the processed information is visualized to construct a unified situation map. This situation map can comprehensively and accurately reflect the real-time status of the battlefield, covering elements such as the troop distribution, weaponry, battlefield environment, and action trends of both sides. It provides a solid data foundation and intuitive battlefield understanding for subsequent modules such as mission planning and scheme generation.

[0061] The implementation of the mission planning module closely revolves around the mission objectives. Through in-depth analysis of the mission's nature, requirements, and actual battlefield conditions, a scientifically sound network of mission dependencies is established. First, the overall mission is comprehensively broken down and refined, clarifying the specific content, objectives, and execution sequence of each sub-task. Dependencies between tasks are identified; for example, some sub-tasks can only be carried out after other sub-tasks are completed, or multiple sub-tasks need to be coordinated to achieve specific effects. Simultaneously, various constraints, such as time limits, troop size, equipment quantity, and logistical support capabilities, are comprehensively considered. These constraints directly affect the mission execution method and resource allocation strategy. To measure the effectiveness and efficiency of mission execution, corresponding performance indicators are formulated, such as mission completion time, target destruction rate, and troop casualty rate, ensuring that mission execution meets the expected mission objectives. After determining mission dependencies, constraints, and performance indicators, task allocation and resource scheduling for cross-domain collaborative tasks are conducted in conjunction with situational awareness and available resource information provided by the situational awareness module. Based on the characteristics and advantages of each branch of the armed forces, sub-tasks are rationally allocated to the corresponding task units. At the same time, resources such as manpower, equipment, and materials are optimized to ensure that resources can be efficiently transferred and optimally utilized between different tasks and task units, forming a powerful synergy for collaborative tasks and providing a clear task framework and resource guarantee for subsequent plan generation.

[0062] The scheme generation module employs a combination of rule-based reasoning and case-based reasoning, fully leveraging existing task knowledge and experience to generate multiple candidate task schemes. First, a comprehensive rule base and case base are established. The rule base contains various task principles, tactical specifications, and weapon usage rules—scientific guidelines summarized from long-term task practice. The case base includes past successful and unsuccessful task cases, including detailed information such as background, task process, strategies adopted, and results achieved. When generating candidate schemes, based on the task requirements, constraints, and performance indicators determined by the task planning module, rule-based reasoning is used to deduce possible task strategies and action steps according to established task rules and logic. Simultaneously, case-based reasoning technology is used to search the case base for cases similar to the current task scenario, analyzing successful experiences and lessons learned from failures, drawing on feasible strategies and methods, and adjusting and optimizing them based on the actual battlefield situation. The results of rule-based reasoning and case-based reasoning are integrated to form multiple different candidate task schemes, each covering specific task objectives, action steps, troop deployment, and resource allocation. These candidate solutions fully consider the complexity and uncertainty of the battlefield, have different characteristics and advantages, provide a wide range of choices for subsequent evaluation and adjustment, ensure that battlefield changes can be addressed from multiple perspectives, and improve the pertinence and effectiveness of mission plans.

[0063] The evaluation and adjustment module employs a dual-channel feedback mechanism to comprehensively and thoroughly evaluate the generated candidate mission plans. Based on the evaluation results, adjustments and optimizations are made, and an iteration log is established to record the entire process. The dual-channel feedback mechanism includes internal and external feedback. Internal feedback primarily comes from the mission command team and experts in various fields. They analyze and evaluate candidate plans from the perspectives of mission theory, tactical rationality, resource utilization efficiency, and risk assessment, pointing out problems and shortcomings, such as whether task allocation is reasonable, resource scheduling is optimized, and whether there are loopholes in the mission strategy. External feedback is obtained through simulated combat exercises, experience sharing with friendly forces, and the use of battlefield simulation systems. Simulated combat exercises can test the feasibility and effectiveness of plans in a near-real battlefield environment, identifying potential difficulties and challenges during actual execution. Experience sharing with friendly forces allows for learning from the successful practices and valuable experiences of other units in similar mission scenarios. Battlefield simulation systems can quantitatively evaluate various performance indicators of the plans through data simulation and analysis, such as mission completion time, casualties, and mission effectiveness. Based on information collected through dual-channel feedback, candidate solutions are comprehensively evaluated, and an evaluation index system is established to score and rank solutions from multiple dimensions. For issues identified during the evaluation, relevant personnel conduct in-depth analysis and discussion, formulate adjustment strategies, and modify and improve the solutions, such as adjusting task allocation, optimizing resource scheduling, and revising task strategies. Simultaneously, a solution iteration log is established, recording in detail the time of each evaluation, participating personnel, evaluation results, adjustment content, and reasons for adjustment, facilitating subsequent review and traceability. This provides historical data support for continuous solution optimization, ensuring that mission solutions can continuously adapt to changes in the battlefield environment and improve mission success rate and effectiveness.

[0064] The human-computer interaction module, based on natural language generation technology, automates the creation of documents such as task orders and coordination plans. These documents are then adjusted and confirmed through an interactive interface to form the final task plan. First, an advanced natural language generation system is developed. This system can automatically generate standardized, accurate, and clear documents, including task orders, coordination plans, and intelligence analysis reports, based on the specific content of the task plan, such as task objectives, action steps, troop deployment, and timelines. Natural language generation technology ensures that the document's language style conforms to military standards, and that the content is complete and logically rigorous, reducing the time and error rate of manual drafting. Simultaneously, a user-friendly and convenient interactive interface is designed, allowing operators to interact with the system in real time. After automatic document generation, operators can review the document content in detail and adjust various aspects based on actual needs and changes in battlefield conditions, such as modifying task descriptions, adjusting schedules, and changing troop deployments. The system can respond to operator adjustment requests in real time, providing intelligent prompts and suggestions to help operators quickly and accurately complete modifications. During the adjustment process, operators can interact with the system to confirm every detail of the plan, ensuring that the final task plan meets actual task requirements and command intentions. The human-computer interaction module enables efficient connection between mission plan generation and confirmation, improving the efficiency and accuracy of mission command, making mission plans more closely reflect actual battlefield conditions, and providing strong support for the smooth implementation of mission operations.

[0065] As described above, the situation assessment-based task plan generation system 100 according to the embodiments of this application can be implemented in various wireless terminals, such as servers with situation assessment-based task plan generation algorithms. In one possible implementation, the situation assessment-based task plan generation system 100 according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the situation assessment-based task plan generation system 100 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the situation assessment-based task plan generation system 100 can also be one of many hardware modules of the wireless terminal.

[0066] Alternatively, in another example, the situation assessment-based task plan generation system 100 and the wireless terminal can also be separate devices, and the situation assessment-based task plan generation system 100 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0067] Various implementations of this disclosure have been described above. The foregoing description is exemplary and not exhaustive. Furthermore, it is not limited to the disclosed implementations, and many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described implementations.

Claims

1. A task plan generation system based on situation assessment, characterized in that, include: The task requirement acquisition module is used to acquire high-level task requirements input by the user; the current situation key feature acquisition module is used to acquire current situation key features collected by the situation awareness module. The information transformation module is used to transform the high-level task requirements and the key features of the current situation into formalized normative problems; the preliminary task scheme generation module is used to input the formalized normative problems, the task rule base and the task case base into the integrated reasoning engine to search for solutions that satisfy the formalized normative problems to obtain the preliminary task scheme skeleton. The detail filling module is used to fill in the details of the preliminary task scheme skeleton to obtain detailed candidate task schemes.

2. The task plan generation system based on situation assessment according to claim 1, characterized in that, The high-level mission requirements include target area, time limit, type and number of available forces, prohibited areas, and effectiveness indicators.

3. The task plan generation system based on situation assessment according to claim 2, characterized in that, The key features of the current situation include the location of key enemy units, terrain obstacles, weather conditions, and the current status of friendly forces.

4. The task plan generation system based on situation assessment according to claim 1, characterized in that, The information transformation module includes: a description information extraction unit, used to extract initial state description, target state description and constraint description from the high-level task requirements and the key features of the current situation; and a formalized normative problem construction unit, used to construct the formalized normative problem based on the initial state description, the target state description and the constraint description.

5. The task plan generation system based on situation assessment according to claim 4, characterized in that, The preliminary task solution generation module includes: a similar case fragment retrieval unit, used to retrieve similar case fragments from the task case library using the integrated reasoning engine based on the initial state description in the formalized normative problem; a preliminary solution skeleton generation unit, used to integrate the similar case fragments into the current solution construction path after adaptive processing to obtain a preliminary solution skeleton prototype; and a verification unit, used to perform feasibility verification on the preliminary solution skeleton prototype using the integrated reasoning engine based on the task rule library, and determine the preliminary solution skeleton prototype as the preliminary task solution skeleton after the feasibility verification is passed.

6. The task plan generation system based on situation assessment according to claim 5, characterized in that, The similar case fragment retrieval unit includes: an initial state semantic encoding subunit, used to perform semantic embedding encoding on the initial state description to obtain an initial state semantic embedding encoding vector; a case fragment semantic encoding subunit, used to perform semantic embedding encoding on each case fragment in the task case library to obtain a set of case fragment semantic embedding encoding vectors; a semantic similarity calculation subunit, used to calculate the semantic similarity between the initial state semantic embedding encoding vector and each case fragment semantic embedding encoding vector in the set of case fragment semantic embedding encoding vectors; and a similar case fragment selection subunit, used to select the case fragment corresponding to the one with the largest semantic similarity as the similar case fragment.

7. The task plan generation system based on situation assessment according to claim 6, characterized in that, The semantic similarity calculation subunit is used to: obtain a transformation matrix based on the initial state semantic embedding encoding vector and the case fragment semantic embedding encoding vector; and to generate a continuous diffusion analytic matrix by subjecting the transformation matrix to a local nonlinear transformation constraint. The continuous diffusion analytic matrix is ​​dynamically mapped onto the transformation matrix to obtain a dynamic mapping matrix; based on the dynamic mapping matrix, the semantic field state of the case fragment semantic embedding encoding vector is adjusted to obtain the case fragment semantic embedding reinforcement encoding vector; the semantic similarity between the initial state semantic embedding encoding vector and the case fragment semantic embedding reinforcement encoding vector is calculated.

8. The task plan generation system based on situation assessment according to claim 1, characterized in that, The detail filling module is used to: identify the variation points in the preliminary task plan skeleton; and fill in the details of each variation point to obtain the detailed candidate task plan, wherein the details include specific unit allocation, precise time nodes, coordinated actions, and firepower allocation.

Images