An equipment simulation training modeling method for a whole battle position process

Abstract

The application discloses a kind of equipment simulation training modeling methods for whole war position process, specifically relates to equipment simulation training process management field, for solving the problem that instruction disordered retransmission and multi-channel feedback gap or conflict coexist in whole war position collaborative training cause node to advance difficult to determine and lack of available evidence in review;By decomposing training process into training process nodes and generating process tokens binding node metadata for nodes, organize instruction delivery and feedback collection with process tokens, perform semantic regulation to generate semantic skeleton for inbound instruction, and conduct double-domain fingerprint comparison, map closure evidence with expected feedback closure structure and sequentially fuse to update replay decision coefficient, output decision results such as discard, merge, complete or derive new node, then assemble closure evidence to generate node completion proof according to process token and write into process evidence chain, realize node-level traceable modeling.

Description

Technical Field

[0001] This invention relates to the field of equipment simulation training process management, and more specifically, to a method for modeling equipment simulation training processes across all combat positions. Background Technology

[0002] In equipment simulation training for all combat positions, LVC synthetic training and distributed simulation federation are often used to interconnect ground stations, virtual cockpits, construction simulation and evaluation systems. The training process is usually organized into several executable nodes, and the nodes are driven by the issuance of control commands, confirmation of receipts, situation or status updates and interface presentation.

[0003] In current implementations, to ensure link reliability, the command side often uses retransmission, sequence number comparison, or deduplication mechanisms based on single-channel confirmation, while the receipt side often summarizes the data using log timelines or simple completion markers. Under conditions of short-term link interruption, replay and out-of-order overlap after recovery, and asynchronous receipts from multiple channels, the number of times the command arrives and the number of times it is executed may be inconsistent for the same process node, and different positions may have different understandings of the node completion time and the basis for completion. This makes it difficult to maintain a consistent standard for the node evidence on which training and evaluation depend, and affects the retrieval and comparability of post-mortem attribution.

[0004] However, when link jitter causes control commands to be replayed or arrive out of order during the recovery phase, and there are gaps or conflicts in multi-channel receipts, the process nodes lack a unified and referable criterion that matches the semantic equivalence of the commands and the consistency of the receipt closures. This makes it difficult to distinguish whether the repeated arrival of commands is a link replay or a re-issuance by the post, which in turn leads to the node being executed repeatedly or suppressed incorrectly, inconsistencies in the proof of node completion between different positions, and breaks or ambiguities in the training evidence chain.

[0005] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an equipment simulation training modeling method for all combat positions. This method decomposes the training process into training process nodes and generates process tokens that bind node metadata to the nodes. The process tokens are used to organize instruction delivery and receipt collection. Semantic reduction is performed on inbound instructions to generate semantic skeletons and perform dual-domain fingerprint comparison. The expected receipt closure structure is combined to map closure evidence and sequentially fuse and update the replay adjudication coefficients. The adjudication results of discarding, merging and supplementing, or deriving new nodes are output. Then, closure evidence is assembled according to the process tokens to generate node completion proof and written into the process evidence chain, thereby realizing node-level traceable modeling to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: S1: Decompose the training process into training process nodes and configure node metadata, generate process tokens that bind node metadata, and the node metadata includes a set of state summary keys and an expected receipt closure structure. S2: The semantic skeleton is formed by semantic specification of inbound instructions and dual-domain fingerprinting is used to obtain semantic skeleton identity. The receipt is mapped to the process token and status digest key to obtain closure evidence net fit. The two are sequentially merged to update the replay adjudication coefficient and output the adjudication result. S3: When the adjudication result allows the node to advance or merge and complete the advancement, the multi-channel receipts are assembled according to the process token and the status summary key is aligned to form a closed evidence set. When the closed evidence set meets the expected receipt closure structure, a node completion proof is generated; otherwise, receipt completion is triggered. S4: Write the lifecycle of the process token into the process evidence chain entry. Calculate the entry summary in a chained manner according to the summaries of the preceding entries and merge them to generate the session root summary. Output the process evidence chain entry.

[0008] Furthermore, the training process node template is generated from the training subject script, the combat position password sequence, and the instruction interface contract. The event syntax is segmented according to the instruction type to trigger the synchronous password closure to form the training process node sequence. The training process node identifier is obtained by concatenating the script node number or the synchronous password identifier with the node sequence mark. The training process node intent label is determined by the task paragraph title and the type of the first instruction in the node.

[0009] Furthermore, the training process node metadata includes a set of instruction types, target entity identifiers, authorization domain constraints, a set of status summary keys, and an expected receipt closure structure. The target entity identifier is obtained by mapping the instruction's target to the equipment entity directory. The authorization domain constraints are obtained by encoding the authorization domain sequence using the battle station permission matrix. The set of status summary keys is extracted and reduced from the instruction type to status field mapping table. The expected receipt closure structure is constructed from the receipt category mapping and dependency rules and obtained through acyclic verification. The process token is generated by hashing the node metadata according to the standard encoding and is kept unique through collision detection and collision resolution.

[0010] Furthermore, the inbound instruction record extracts the instruction type, target entity identifier, authorization domain identifier, instruction field set, arrival channel identifier, and arrival time, and enters the pending judgment queue according to the process token. When the inbound instruction record is missing a process token, it locates the unique process token in the process token binding table according to the instruction type, target entity identifier, and authorization domain constraint. If the location fails, it outputs a resigned token derivation decision.

[0011] Furthermore, after the inbound instruction record enters the pending judgment queue, semantic reduction is performed to generate a semantic skeleton string. The semantic reduction performs reference frame reduction on the spatial position field and attitude field and unit domain reduction on the velocity field and distance field. The semantic reduction removes the retransmission identifier and message sequence number and retains the control field to form a semantic skeleton string. The semantic skeleton string is subjected to dual-domain fingerprinting to generate a semantic fingerprint and an execution fingerprint, and the semantic skeleton identity is obtained by comparing it with the historical received instruction fingerprint.

[0012] Furthermore, the set of arrived receipts under the same process token is mapped to the expected receipt closure structure as a closure evidence record according to the status summary key set. The closure evidence record calculates the net fit of the closure evidence and registers the conflict key and conflict source receipt category. Sequential evidence fusion is performed with the semantic skeleton identity and the net fit of the closure evidence to obtain the replay decision coefficient and output the discard decision, merge and complete decision, or resign token derived decision.

[0013] Furthermore, the assembly process token is determined according to the ruling result, and the expected receipt closure structure and status summary key set are read from the process token binding table. The set of arrived receipts corresponding to the assembly process token is normalized by the execution channel adaptation table, and a receipt index and a secondary index of status summary key are established according to the receipt category. When there are multiple receipt records for a receipt category, the main receipt record is determined according to the arrival time, and the remaining receipt records are written into the duplicate receipt list.

[0014] Furthermore, according to the order of receipt assembly in the expected receipt closure structure, the main receipt record is extracted from the receipt index to form a closure evidence set and the missing receipt category set is located. The closure evidence set is aggregated into a set of state summary values ​​according to the state summary key and the set of conflict keys and the set of conflict source receipt categories are located. When the set of missing receipt categories is empty and the set of conflict keys is empty, a node completion proof record is generated. When the set of missing receipt categories is not empty, a receipt completion request is generated. When the set of conflict keys is not empty, a comparison query request is generated and the retention or unresolved mark is executed according to the consistency of the comparison summary value.

[0015] Furthermore, the process evidence chain uses the process token as the primary key to continuously write process evidence chain entries. Each process evidence chain entry uniformly carries the full lifecycle evidence corresponding to the training process node identifier and maintains the consistency of event timing. The writing process adopts fixed standard coding and synchronously associates nodes to complete the proof record, so that the same training process node forms a consistent evidence caliber under multi-channel playback conditions.

[0016] Furthermore, the current entry summary is generated by concatenating the previous entry summary and the standardized encoding of the process evidence chain entries. All current entry summaries in the training session are merged in a deterministic order to form the session root summary. During the debriefing phase, the adjudication trajectory and closure trajectory are traced back based on the training process node identifiers, and the node completion proof records are extracted.

[0017] The technical effects and advantages of the equipment simulation training modeling method for the entire combat station process of this invention are as follows: This invention transforms full-position training from a message-stacking recording method to a verifiable, progressive process centered on training flow nodes. It utilizes flow tokens to bind inbound commands, three-channel receipts, adjudication conclusions, and completion proofs to the same node context, freeing training progression from the "first come, first served" assumption of a single channel. When link jitter, out-of-order retransmissions, or duplicate commands are issued, semantic reduction and dual-domain fingerprints stably characterize the control intent of inbound commands, while receipt closure evidence structurally verifies the actual progress status of nodes. These two types of evidence, under a sequential fusion mechanism, form a consistent adjudication, thus distinguishing between "equivalent command replay" and "true intent change," preventing the training posture from being repeatedly driven by replayed commands or misjudged as executed, and ensuring that node progress and receipt closures are aligned on the same logical chain.

[0018] Meanwhile, the node completion proof and process evidence chain solidify each ruling basis, closure gap, conflict resolution and completion process into a reviewable evidence sequence. This allows for direct identification of training anomalies during review, whether they stem from instruction replay, missing receipts, or summary conflicts. Furthermore, it enables alignment of training progress and responsibility attribution using the same proof standard across different operational positions. Consequently, training evaluation shifts from subjective replay to evidence-driven, interpretable evaluation, reducing the risk of misattribution due to inconsistencies across multiple channels and enhancing the credibility and controllability of node progression in complex training processes. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an equipment simulation training modeling method for all combat positions according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: Figure 1 This invention presents a method for equipment simulation training modeling oriented towards the entire combat position process, comprising: S1, decompose the training process into training process nodes and configure node metadata, generate process tokens that are bound to node metadata, and the node metadata includes a set of state summary keys and an expected receipt closure structure. S2, the semantic skeleton is formed by semantic specification of inbound instruction and dual-domain fingerprinting to obtain semantic skeleton identity degree. The receipt is mapped by process token and status summary key to obtain closure evidence net fit degree. The two are sequentially merged to update replay adjudication coefficient and output adjudication result. S3, when the adjudication result allows the node to advance or merge and complete the advancement, the multi-channel receipt is assembled according to the process token and the status summary key is aligned to form a closed evidence set. When the closed evidence set meets the expected receipt closure structure, the node completion proof is generated; otherwise, receipt completion is triggered. S4. Write the lifecycle of the process token into the process evidence chain entry. Calculate the entry summary in a chained manner according to the summaries of the preceding entries and merge them to generate the session root summary. Output the process evidence chain entry.

[0022] This invention views "full-position training" as a process consisting of multiple training nodes linked together, rather than a collection of independent messages. First, the training process is broken down into several training process nodes. Each node is configured with allowed instruction types, target entity identifiers, authorization domain constraints, a set of status digest keys, and an expected acknowledgment closure structure. A process token is then generated for each node, ensuring that instruction delivery, acknowledgment collection, adjudication, and completion verification are all organized around the same process token.

[0023] When an inbound command arrives, instead of directly judging whether it is a duplicate based on the order of arrival, the command is first semantically reduced, unifying the reference frame and unit domain, and removing fluctuation fields that do not affect the control intent to obtain a semantic skeleton. Then, semantic fingerprints and execution fingerprints are generated and compared with the fingerprint of the most recently received command at the same training process node to obtain the semantic skeleton identity. At the same time, the flight control channel, status channel, and ground station presentation channel acknowledgments are mapped to closure evidence according to the process token and status summary key. The closure coverage and summary consistency are evaluated to obtain the net fit of the closure evidence. Then, the semantic skeleton identity and the net fit of the closure evidence are sequentially fused as two complementary pieces of evidence, and the decision result of discarding, merging and supplementing, or resigning tokens to derive new nodes is output, thereby distinguishing between "equivalent replay" and "true reissue".

[0024] When the decision allows progress, the multi-channel receipts are associated and assembled according to the process token, the state summary key is aligned, and the expected receipt closure structure is checked. If satisfied, a node completion proof is generated; otherwise, the missing receipt category and conflict key are located, and completion and resolution are triggered. Finally, the entire process of the process token from generation to consumption, decision, assembly, and completion proof is written into the process evidence chain, enabling direct tracing during review of why each node was judged to be replayed, why it entered the completion stage, where the conflict came from, and how it was ultimately completed, thus supporting a consistent approach to training evaluation and attribution.

[0025] Full-position equipment simulation training often proceeds simultaneously across positions, links, and channels. Command delivery and feedback are not naturally confined to the same context. Once the training process enters a state of out-of-order arrival, repeated delivery, or short-term channel disconnection, message-level management will find it difficult to maintain clear training node boundaries and consistent progress. Therefore, it is necessary to first break down the training process into manageable training process nodes and establish a set of node semantic boundaries for each node that can constrain command triggering and feedback closure.

[0026] S101 Training Process: Node Extraction and Node Boundary Determination

[0027] During full-station training, ground station commands are issued in parallel with flight control feedback, status feedback, and presentation feedback. Replay and out-of-order playback can occur when the link jitters. To ensure that the same training intent remains consistent from the perspective of different stations, the training process needs to be broken down into clearly defined training process nodes, and deterministic rules should be used to ensure that the start and end of the nodes do not drift with the log sequence.

[0028] The training process node sequence is generated using the sequence of subject flows given by the training subject script as the main timeline, the synchronization passwords given by the battle station password sequence as node boundary prompts, and the instruction type enumeration given by the instruction interface contract as the node trigger set, according to deterministic segmentation rules. The segmentation rules adopt finite state transition logic, and the state set includes three categories: initial state, node collection state, and closed waiting state.

[0029] When the initial state receives an instruction event belonging to the instruction type enumeration, it enters the node collection state and creates a training process node. The training process node identifier is taken from the node number in the training subject script. If the training subject script does not provide a node number, the training process node identifier is a unique identifier formed by concatenating the synchronization password identifier received for the first time in the node collection state with the node sequence mark. The node collection state continuously collects instruction events and synchronization password events related to the same training intent. The training intent label is determined by the task paragraph title in the training subject script and the first instruction type within the node. When the node collection state encounters a password event marked as the end of the node in the synchronization password sequence, it enters the closed waiting state and freezes the set of instruction types that can be triggered within the node. The closed waiting state ends the current training process node and returns to the initial state when all the closed receipt categories specified in the training subject script have appeared or when the next node start password event is encountered. If the closed receipts are incomplete but the next node start password appears, the list of incomplete receipt categories is written as node gap markers into the training process node metadata for use in subsequent steps. The range of values ​​for the training process node sequence is limited to the total number of nodes covered by the training subject script. The training process node identifier must be unique within the same training subject version and cannot be empty.

[0030] Example: Taking the UAV ground station route issuance subject as an example, the training subject script defines route issuance as a task segment. The combat station command sequence includes the start issuance command and the issuance completion command. After receiving the route issuance command event in the initial state, a training process node is created. The node collects the status and continuously records the route issuance command and confirmation command. In the closed waiting state, it waits for the flight control execution confirmation and status summary confirmation to appear. If the confirmation is missing due to link jitter, the confirmation is marked as a node gap and the node is terminated to avoid the node boundary being delayed by the missing receipt, which would cause confusion in the next node's affiliation.

[0031] S102 node metadata construction and value range constraints.

[0032] Replay adjudication and acknowledgment assembly must rely on stable node metadata. The node metadata needs to simultaneously constrain the semantic scope of instructions, the scope of target entities, the scope of permissions, and the scope of acknowledgment alignment in order to maintain consistency in node-level aggregation during link replay and out-of-order execution.

[0033] For each training process node output in step S101, node metadata is constructed. The node metadata consists of five parts: instruction type set, target entity identifier, authorization domain constraint, state summary key set, and expected acknowledgment closure structure. The instruction type set is taken from the instruction type enumeration of the instruction interface contract, and its value range is limited to a finite discrete set within the enumeration. The instruction type set is accumulated and deduplicated during the node collection state and frozen during the closed waiting state, disallowing cross-node drift. The target entity identifier is taken from the equipment entity directory, which is maintained by the training equipment list. Each entry includes three items: entity name, entity unique identifier, and entity category. The target entity identifier resolution rule is to extract the entity name from the target object field of each instruction within the node and map it to the entity unique identifier in the equipment entity directory. When multiple entity unique identifiers exist within a node, the target entity identifier is the one with the highest frequency and covered by the authorization domain constraint, avoiding mistaking a bypass observation object as a control object.

[0034] The authorization domain constraint is derived from the position authority matrix, which is maintained by the training organization. Rows represent position identifiers, and columns represent command types. The authorization domain constraint is represented as an authorization domain sequence, arranged in a fixed order according to the position identifiers. Each bit in the sequence is a binary label; the first value indicates triggering permission, and the second value indicates no triggering permission. The length of the authorization domain sequence equals the number of position identifiers. The status digest key set is derived from the command type to status field mapping table, which is generated by extracting status field paths from the command interface contract. Each status field path includes a naming domain and a field path. A digest key is calculated for each status field path. The process involves concatenating the naming domain string and the field path string using a fixed separator to form a digest input string. This digest input string is then fed into an anti-collision hash algorithm to calculate a fixed-length digest key. The digest key value is limited to a fixed-length set of bit strings with the dimension of an identifier. The status digest key set is sorted lexicographically within each node and deduplicated to ensure consistency in subsequent assembly.

[0035] S103 Expected Receipt Closure Structure Generation and Acyclic Validation.

[0036] Multichannel receipt assembly requires a clear definition of the receipt category set and dependencies. Missing dependencies will prevent the identification of gaps in out-of-order receipts, and loops in dependencies will prevent the closure condition from being met. The expected receipt closure structure must be generated from a searchable specification source and ensure assemblability through acyclic validation.

[0037] The expected acknowledgment closure structure is derived from the instruction type to acknowledgment category mapping table and the acknowledgment dependency rule set. The instruction type to acknowledgment category mapping table is jointly maintained by the training organization specification and the instruction interface contract. The mapping table provides a subset of required acknowledgment categories and a subset of optional acknowledgment categories for each instruction type. The range of acknowledgment category values ​​is limited to a finite discrete set, covering at least three categories: flight control execution confirmation, status summary confirmation, and presentation confirmation.

[0038] The set of receipt dependency rules is maintained by the training organization. Rules are given in the form of directed dependencies; for example, execution confirmation precedes state summary confirmation, and state summary confirmation precedes presentation confirmation. The rule set does not allow pairs of rules that are mutually prerequisites. The closure structure generation process involves first taking the union of the instruction type set in the node metadata with the mapping table to obtain the node-level receipt category set, and then generating a dependency relationship set according to the receipt dependency rule set. Each dependency relationship consists of a preceding receipt category and a following receipt category. Acyclic validation uses a topology sorting method. The input to topology sorting is the receipt category set and the dependency relationship set, and the output is the receipt assembly order. If topology sorting fails, a dependency resolution process is executed. This process first locates the dependencies introduced by optional receipt categories in the dependency relationship set and removes them according to rule priority, then re-executes topology sorting until topology sorting succeeds or all removable dependencies are exhausted. If the process still fails after exhausting all removable dependencies, the required receipt category set remains unchanged, and an anomaly marker for the closure structure is written into the node metadata. In the subsequent adjudication phase, this anomaly marker is used as one of the conditions for resigning tokens to derive nodes. The range of values ​​for the receipt assembly order is limited to a total order permutation of the receipt category set, with a length equal to the size of the receipt category set.

[0039] S104 Process Token Generation, Binding, and Collision Resolution.

[0040] Replay decisions and receipt assembly rely on the same anchor identifier. The anchor identifier needs to have a deterministic mapping to the node metadata and also needs to remain unique within the same training subject version. Relying solely on anti-collision hashes is insufficient to form an engineering-level uniqueness guarantee, and collision detection and resolution logic needs to be added.

[0041] The process token is obtained by inputting the node metadata after standardized encoding and anti-collision hash calculation. The node metadata standardized encoding process involves concatenating five types of encoded fragments in a fixed order: the first fragment is the training process node identifier text; the second fragment is the concatenated text of the instruction type set sorted lexicographically; the third fragment is the target entity identifier text; the fourth fragment is the concatenated text of the authorization domain sequence sorted in a fixed job order; and the fifth fragment is the concatenated text of the status digest key set sorted lexicographically and the concatenated text of the receipt assembly order. These five types of encoded fragments are connected using a fixed delimiter to obtain the standardized encoded input string. The anti-collision hash calculation maps the standardized encoded input string to a fixed-length bit string to obtain the process token, whose dimension is the identifier value. The process token binding process involves establishing a process token binding table, with the training process node identifier as the key and the process token and node metadata as the value. The collision detection process involves comparing the newly generated process token with each existing process token in the process token binding table. If the process tokens are the same but the training process node identifiers are different, a collision is determined. The collision resolution process involves generating a collision resolution marker and recalculating the process token. The collision resolution marker generation process involves concatenating the training process node identifier text and the node sequence marker text with a fixed delimiter, inputting them into the same anti-collision hash to calculate a fixed-length bit string, appending the collision resolution marker to the end of the canonical encoded input string, and performing anti-collision hash calculation again to generate a new process token. This collision detection is repeated until the process token is unique in the process token binding table.

[0042] The training process node template uses node metadata to constrain instruction types, target entity identifiers, authorization domain constraints, state summary key sets, and expected receipt closure structures. It also generates a process token for each training process node to complete the binding, thus fixing the training process within the scope of the node identified by the process token. The training data organization is transformed from a message set to a unified organization of node entities and node evidence.

[0043] Even after process tokens establish node boundaries, inbound commands may still be retransmitted out of order due to link jitter or issued repeatedly due to semantically equivalent operations. Simply relying on message sequence numbers and arrival times cannot distinguish control intent. Figure 1 The replay of instructions and the re-issuance of instructions due to changes in control intent necessitate the reduction of the instruction content to stable semantics and the combination of node closure evidence to form a definable adjudicative caliber.

[0044] Step S2 selects semantic skeleton identity and closure evidence net fit as two complementary parameters for comprehensive analysis to obtain the interference confidence coefficient. The fundamental reason is that the "interference" in full-position training is not a single source, but a coupling distortion between the command side and the receipt side: semantic skeleton identity only answers whether the inbound command is equivalent to the historical received command in terms of control intent, and can filter out non-essential differences in out-of-order retransmission, field fluctuations and repeated issuance of commands, but cannot explain whether the node has progressed in the actual execution link; closure evidence net fit only answers whether the receipt closure under the same process token is fully covered and whether the status summary key is consistent, and can reflect the uncertainty of progress caused by multi-channel receipt gaps and conflicts, but cannot distinguish the essential difference between "the same command being replayed" and "different intents causing receipt forks". By combining the two for determining the interference confidence coefficient, the discrimination simultaneously satisfies both the intent equivalence and closure consistency constraints. This avoids misjudging genuine re-issues as interference based solely on instruction similarity in engineering, and also avoids misjudging link jitter as execution anomalies based solely on receipt gaps. Ultimately, this makes the three types of decisions—drop, merge and complete, and re-signed derivation—more stable, node progress more verifiable, and post-mortem attribution more explainable.

[0045] S201 Inbound Instruction Entry and Process Token Ownership Determination.

[0046] Replay decisions must be limited to the training process nodes; otherwise, similar instructions between different training process nodes will interfere with each other. Process token ownership determination is used to ensure that inbound instructions fall into a unique training process node.

[0047] Upon receiving an inbound instruction, the instruction type, target entity identifier, authorization domain identifier, instruction field set, arrival channel identifier, and arrival time are extracted to form an inbound instruction record. When the inbound instruction record contains a process token, the value range of the process token is limited to the process token set generated in step S1, and the inbound instruction record enters the pending judgment queue according to the process token. When the inbound instruction record does not contain a process token, a default attribution judgment is performed. The default attribution judgment retrieves the candidate training process node metadata set from the process token binding table. The candidate training process node metadata satisfies three consistency conditions: the first consistency condition is that the instruction type belongs to the instruction type set in the training process node metadata; the second consistency condition is that the target entity identifier is equal to the target entity identifier in the training process node metadata; and the third consistency condition is that the authorization domain identifier belongs to the authorization domain constraint coverage set in the training process node metadata. When the number of candidate training process node metadata is one, the corresponding process token is written into the inbound instruction record and added to the pending judgment queue. When the number of candidate training process node metadata is not one, a resigned token derivation decision is output and the inbound instruction record is transferred to the exception queue. After the process token is enqueued, its consumption status is read. The consumption status of the process token is limited to two values: unconsumed and consumed. Subsequent decision branches are executed based on the consumption status of the process token.

[0048] S202 semantic reduction generates a semantic skeleton string.

[0049] The link recovery phase is often accompanied by changes in channel identifiers, retransmission identifiers, and message sequence numbers. Semantic specifications are used to remove fluctuation fields that do not affect control intent and compress inbound instructions into semantic skeleton strings with stable control intent, so as to avoid equivalence judgment being misled by fluctuation fields.

[0050] The semantic reduction input is the set of instruction fields from the inbound instruction record, and the semantic reduction output is a semantic skeleton string, with the dimension of the semantic skeleton string being a sequence of identifier bytes. Semantic reduction includes two types of processing: reference frame reduction and unit domain reduction. Reference frame reduction transforms spatial position and attitude fields to the standard reference frame specified in the instruction interface contract. The transformation order is: first, obtain the homogeneous coordinate vector in the source reference frame, then multiply the homogeneous coordinate vector in the source reference frame by the homogeneous transformation matrix to obtain the homogeneous coordinate vector in the standard reference frame. The spatial position field retains its length dimension, and the attitude field retains its dimensionless dimension. Unit domain reduction transforms velocity, angular velocity, distance, and time window fields to the standard unit set specified in the instruction interface contract. The transformation order is: first, identify the original unit enumeration of the field, then convert it to the standard unit according to the unit mapping table. Field filtering removes the channel transient identifier, retransmission identifier, message sequence number, and check field, while retaining the target entity identifier, instruction type, and reduced control fields. Fields are sorted lexicographically according to the field paths in the instruction interface contract and concatenated to form the semantic skeleton string.

[0051] Example: Taking the UAV ground station route issuance as an example, during the recovery phase of the inbound command, two route issuance commands with the same content but different message sequence numbers appear. The semantic specification removes the message sequence number and retransmission identifier, retains the route point sequence field and the target entity identifier field, and unifies the coordinate reference system, while keeping the semantic skeleton string consistent.

[0052] S203 dual-domain fingerprinting and semantic skeleton identity calculation.

[0053] The semantic skeleton string field-by-field comparison is sensitive to field missing and field order perturbation. Dual-domain fingerprinting maps the semantic skeleton string into two stable representations: the task intent domain and the execution control domain. Semantic skeleton identity is used to characterize the semantic equivalence between inbound instructions and historical received instructions.

[0054] The dual-domain fingerprinting input is a semantic skeleton string, and the dual-domain fingerprinting output consists of a semantic fingerprint and an execution fingerprint. The values ​​of the semantic fingerprint and execution fingerprint are limited to a fixed-length set of bit strings, with the dimension being the identifier quantity. The semantic fingerprint is constructed by first building an intent tag sequence based on the metadata of the training process nodes. The intent tag sequence consists of intent tags derived from the target entity identifier, instruction type, and task paragraph title. Then, a Locality Sensitive Hash (LSH) signature algorithm is applied to the intent tag sequence to obtain the semantic fingerprint. The LSH signature algorithm is constructed by calculating a multi-path hash bit string for each intent tag, accumulating positive and negative counts for each bit, and determining the value of the corresponding bit in the output fingerprint by the count sign after accumulation. The execution fingerprint is constructed by first building a control plane tag sequence based on the instruction interface contract. The control plane tag sequence consists of control plane tags derived from the path to the control field and the control field specification value. Then, the same LSH signature algorithm is applied to the control plane tag sequence to obtain the execution fingerprint.

[0055] The semantic skeleton identity score is limited to a closed interval between zero and one, and its dimensionless value. The calculation order for semantic skeleton identity score is as follows: First, read the semantic fingerprint and execution fingerprint of the most recent historical accepted instruction from the process token consumption record. Then, calculate the Hamming distance between the semantic fingerprint and the execution fingerprint, respectively. The Hamming distance is defined as the number of bits of the same length that differ in corresponding positions. Next, divide the Hamming distance of the semantic fingerprint by the fingerprint length to obtain the semantic normalized distance, and divide the Hamming distance of the execution fingerprint by the fingerprint length to obtain the execution normalized distance. Then, subtract the semantic normalized distance from a constant to obtain the semantic similarity, and subtract the execution normalized distance from a constant to obtain the execution similarity. Finally, use harmonic synthesis to obtain the semantic skeleton identity score. The harmonic synthesis calculation order is as follows: first, calculate the reciprocal of the semantic similarity and the reciprocal of the execution similarity, then average the two reciprocals, and finally take the reciprocal of the average.

[0056] When semantic similarity or execution similarity falls outside a preset open interval, boundary convergence processing is performed first. This process restricts the similarity to within the preset open interval, preventing the reciprocal and natural logarithm from becoming infinite. If no historical received instructions exist, the semantic fingerprint and execution fingerprint are registered as the initial fingerprint, and the semantic skeleton identity is set to the maximum value.

[0057] S204 Closure Evidence Mapping and Closure Evidence Net Fit Calculation.

[0058] When inbound instructions have similar semantics, closure evidence coverage and closure evidence consistency can better reflect whether the training process nodes have been advanced; closure evidence net fit quantifies the receipt gap and receipt conflict in a unified way, providing a stable decision for sequential evidence fusion.

[0059] The input to the closure evidence mapping is a set of arrived receipts, a set of status digest keys, and a desired receipt closure structure for the same process token. The output is a node-level closure evidence record. Each receipt in the arrived receipt set contains a receipt category, a process token, a subset of status digest keys, and a set of status digest values. The range of receipt category values ​​is limited to the set of receipt categories in the desired receipt closure structure. The status digest value is formed by extracting a canonical encoded string from the status field value carried in the receipt, and then performing anti-collision hashing on the canonical encoded string to obtain a fixed-length bit string, with the dimension being the identifier.

[0060] The net fit of closure evidence is limited to a closed interval of zero to one, and its dimension is dimensionless. The calculation of net fit of closure evidence consists of two parts: coverage assessment and conflict deduction. The coverage assessment is calculated as follows: first, the set of necessary receipt categories is read from the expected receipt closure structure; then, the number of necessary receipt categories covered in the received receipt set is counted; finally, the coverage ratio is obtained by dividing the number of covered necessary receipt categories by the total number of necessary receipt categories. If the total number of necessary receipt categories is empty, the coverage ratio is set to its maximum value. The conflict deduction is calculated as follows: first, the set of state summary keys appearing in the received receipt set is counted; then, the set of corresponding state summary values ​​is collected for each state summary key. If the set of state summary values ​​contains more than one value, a conflict key is determined, and the conflict source receipt category is recorded. Then, the conflict ratio is obtained by dividing the number of conflict keys by the total number of state summary keys that have appeared; finally, the consistency ratio is obtained by subtracting the conflict ratio from a constant; if the total number of state summary keys that have appeared is empty, the consistency ratio is set to its minimum value. The net fit of closure evidence is calculated by multiplying the coverage ratio by the consistency ratio to obtain the net fit of closure evidence, and writing the conflict key set and conflict source receipt category into the closure evidence record for step S3 to locate the conflict source.

[0061] S205 Sequential evidence fusion updates the replay decision coefficients and outputs the decision result.

[0062] During the link recovery phase, replay commands often arrive in clusters, and single-time discrimination is easily interfered with by out-of-order and gap receipts. Sequential evidence fusion utilizes the continuous arrival process of process tokens to accumulate evidence and stably output the adjudication result.

[0063] The replay decision coefficient is limited to a closed interval between zero and one, and its dimension is dimensionless. Sequential evidence fusion maintains the evidence score, with an initial value of zero. Upon arrival of each inbound instruction record, the evidence score is updated sequentially. The update order is: first, calculate the logarithmic ratio of semantic skeleton identity; then, calculate the logarithmic ratio of net fit of closure evidence; finally, add the two logarithmic ratios to the evidence score from the previous time step to obtain the new evidence score.

[0064] The logarithmic ratio is calculated by first dividing the parameter by a constant and then subtracting the parameter to obtain the ratio, and then taking the natural logarithm of the ratio. Before the semantic skeleton identity and net fit of closure evidence are included in the logarithmic ratio calculation, boundary convergence processing is performed to restrict the parameter to within a preset open interval, preventing the natural logarithm from becoming infinite. The replay adjudication coefficient is obtained from the evidence score through logical mapping. The logical mapping calculation order is as follows: first, take the negative of the evidence score; then, take the exponent of the negative; then, add the exponent result to the constant; finally, take the reciprocal to obtain the replay adjudication coefficient.

[0065] The adjudication boundaries adopt a pre-defined three-part adjudication boundary set, which is given by the training organization specification or subject configuration file. The three-part adjudication boundary set satisfies the following condition: the discard adjudication boundary is greater than the merged and supplemented adjudication boundary, and the merged and supplemented adjudication boundary is greater than the derived adjudication boundary. The adjudication rule order is as follows: first determine the consumption status of the process token, and then determine the interval assignment of the replay adjudication coefficient and the net fit of the closure evidence.

[0066] When the process token consumption status is consumed and the replay decision coefficient falls within the discard decision range, the discard decision is output and the inbound instruction record is marked as link replay; When the semantic skeleton is in the equivalent instruction range and the net fit of the closure evidence is in the unclosed range, the merge and complete decision is output and the inbound instruction record is added to the complete sub-queue. The complete sub-queue is only allowed to trigger receipt completion and is not allowed to trigger repeated execution. When the net fit of the closure evidence falls into the conflict range or the replay decision coefficient falls into the derived decision range, a resigned token derived decision is output. The resigned token derived decision generates a derived flag and submits it to the process token generation logic. The process token generation logic generates a new process token based on the derived flag and writes it into the process token binding table. Step S3 selects the process token or the new process token for receipt association and assembly based on the decision result.

[0067] Example: Taking the UAV ground station route issuance as an example, during the recovery phase, two inbound commands with semantic skeletons close to the maximum value appear. The flight control execution confirmation has arrived, but the confirmation is missing. The net fit of the closure evidence falls into the unclosed interval. The sequential evidence fusion output merges and completes the decision. The inbound command record enters the completion sub-queue. The ground station no longer issues routes repeatedly, but only triggers the confirmation completion. After the closure is completed, the process proceeds to step S3, the assembly node, to complete the proof.

[0068] After the inbound instruction enters the pending queue at the process token level, it generates a semantic skeleton and forms a semantic fingerprint and an execution fingerprint. The semantic skeleton identity degree describes the equivalence relationship of the instruction intent. The net fit of the closure evidence describes the consistency between the closure coverage of the receipt and the summary. The two types of complementary evidence are fused sequentially to update the replay adjudication coefficient and output the new node adjudication result derived from discarding, merging and supplementing or resigning tokens.

[0069] The ruling indicates whether to advance the node and whether a new node needs to be derived. However, the credibility of the training advancement depends on whether the acknowledgments from the flight control channel, status channel, and ground station presentation channel form a consistent closure under the same process token. If acknowledgment gaps and summary conflicts are not structured and located, the node completion status will diverge from the perspectives of different positions and reduce the interpretability of the post-mortem.

[0070] S301 Assembly Process Token Determination and Node Constraint Loading.

[0071] The assembly of receipt associations revolves around the training process nodes. The boundaries of the training process nodes are defined by the process token. When the adjudication result is derived, the process token is switched. The assembly of the process token needs to be determined first to avoid receipts from being mixed into the same closure evidence set across nodes.

[0072] Step S2 outputs the adjudication result, process token, and derived process token. The assembly process token is determined according to the adjudication result. When the adjudication result is to re-sign the token and derive the adjudication, the assembly process token takes the derived process token. When the adjudication result is to allow the node to advance the adjudication or merge and complete the adjudication, the assembly process token takes the process token. After the assembly process token is determined, the training process node metadata, expected receipt closure structure, and status summary key set are read from the process token binding table. If the read fails, the assembly status is set to abnormal assembly and written to the exception record. The exception record contains the assembly process token and the training process node identifier.

[0073] S302 Multi-channel receipt normalization and master receipt determination.

[0074] The message structures of flight control channel receipts, status channel receipts, and presentation channel receipts are different, and the receipt category naming and status field paths are different. Normalization needs to converge the channel differences to a unified receipt category and a unified status summary key expression. The determination of the main receipt needs to converge duplicate receipts into a single assembly entry point.

[0075] The system reads the set of arrived receipts corresponding to the assembly process token. Each receipt in the set contains at least a receipt category identifier, an assembly process token, a subset of status digest keys, a set of status digest values, an arrival channel identifier, and an arrival time. Receipt normalization is performed based on the channel adaptation table, which is generated from the receipt field paths and message header identifiers of each channel given by the instruction interface contract. The channel adaptation table maps the channel message header identifiers to a unified receipt category and maps the channel field paths to key entries in the status digest key set. After normalization, a receipt index is created by receipt category, and a secondary index is created by status digest key. The secondary index records the list of receipt records corresponding to the status digest key. The main receipt determination rule is performed on a per-receipt-category basis. First, receipt records with empty status digest key subsets are filtered out. Then, the receipt record with the earliest arrival time is selected as the main receipt record from the remaining receipt records. The remaining receipt records are written to the duplicate receipt list and the arrival channel identifier and arrival time are retained for subsequent conflict backtracking.

[0076] S303 Closure Evidence Assembly and Consistency Verification.

[0077] The completion of node proof requires the closure evidence set to satisfy the required receipt category closure of the expected receipt closure structure, and the state summary key to be consistent. The closure evidence assembly and consistency verification transform gaps and conflicts into a clear set of missing receipt categories and a set of conflict keys.

[0078] The required receipt category set and receipt assembly order are read from the expected receipt closure structure. The required receipt categories are traversed according to the receipt assembly order, and the corresponding master receipt record is extracted from the receipt index to form the closure evidence set. If no master receipt record is found for a required receipt category, the required receipt category is written to the missing receipt category set and the traversal continues. After the traversal, the closure evidence set and the missing receipt category set are obtained. Consistency verification is performed at the granularity of state digest key. First, the set of state digest keys that have appeared is gathered from the closure evidence set. Then, the set of state digest values ​​corresponding to each state digest key that has appeared is collected. If the set of state digest values ​​has more than two values, the state digest key is written to the conflict key set, and the receipt category that produces different values ​​is written to the conflict source receipt category set. The node closure completion judgment adopts a two-condition approach: the closure is judged to be complete when the missing receipt category set is empty and the conflict key set is empty; the closure is judged to be incomplete when the missing receipt category set is not empty or the conflict key set is not empty.

[0079] Example: In the UAV route distribution node, the flight control channel receipt and the status channel receipt arrive, the presentation channel receipt does not arrive, the closure evidence set includes flight control confirmation and status summary confirmation, the missing receipt category set includes presentation confirmation, the conflict key set is empty, the node closure completion is determined to be incomplete and enters the completion process.

[0080] Node S304 completes the generation of proofs and the completion and resolution tasks.

[0081] Training debriefing requires a single node completion proof to align multiple combat position perspectives; incomplete closures require executable completion requests and verifiable conflict resolution requests; node completion proofs and task generation require the same assembly process token constraints.

[0082] When a node closure is deemed complete, a node completion proof is generated. This generation process follows canonical coding and anti-collision hashing. The canonical coding input is formed by concatenating the assembly process token, the assembly order of the receipts, the receipt category of each master receipt record in the closure evidence set, the subset of state digest keys, and the set of state digest values ​​in lexicographical order. Anti-collision hashing is then performed on the canonical coding input to obtain the node completion proof, which is written to the node completion proof record. When a node closure is deemed incomplete, a completion and resolution task is generated. This task generates receipt completion requests for each item in the missing receipt category set. Each receipt completion request includes the assembly process token, the missing receipt category, and the subset of keys associated with the missing receipt category in the state digest key set. The task generates a comparison query request for each item in the conflict key set. The comparison query request includes the assembly process token, the conflict key, and the set of conflict source receipt categories. The comparison query data source is limited to the training situation bus or the device status snapshot storage. The data source selection order is as follows: first, locate the status snapshot of the node time period according to the assembly process token; then, extract the status field standard encoding string according to the field path corresponding to the conflict key; and then perform anti-collision hashing on the status field standard encoding string to obtain the comparison summary value. When the comparison summary value is consistent with the status summary value corresponding to the same conflict key in the closure evidence set, retain the master receipt record and mark the conflict key as resolved. When the comparison summary value is inconsistent with the status summary value, mark the conflict source receipt category as unresolved and keep the assembly status as pending.

[0083] Example: The ground station presents two duplicate receipts. The duplicate receipt list retains two arrival times. The main receipt record selects the receipt with the earliest arrival time. Subsequently, a status summary conflict key appears. The conflict key field is extracted from the status snapshot by the comparison query, and a comparison summary value is formed to complete the retention and removal. The conflict key is marked as resolved or not resolved and falls into the corresponding task queue.

[0084] Step S3 outputs the closure evidence set, missing receipt category set, conflict key set, conflict source receipt category set, node completion proof record, receipt completion request, and comparison query request at the assembly process token dimension. The closure evidence assembly and consistency verification are determined under the constraints of the expected receipt closure structure and state summary key set. Step S4 records the node completion proof and completion resolution trajectory to form process evidence chain entries based on this.

[0085] After the assembly process token is determined, normalization and index convergence are performed on the multi-channel receipts. The closure evidence set is assembled according to the expected receipt closure structure, and consistency verification is performed on the state summary key. When the closure is completed, a node completion proof record is generated. When the closure is not completed, the missing receipt category set and conflict key set are output, and receipt completion request and comparison query request are triggered to keep the node in the pending state.

[0086] The completion of node certification provides verifiable evidence for node advancement. However, the evaluation and attribution of training at all combat positions require connecting the entire process of token generation, adjudication update, receipt assembly, completion resolution, and completion certification into a continuous sequence of evidence. Otherwise, the review can only show the results and cannot locate the source of evidence, the path of conflict evolution, and the boundary of responsibility.

[0087] S401 Process Evidence Chain Entry Field Definitions and Value Range Constraints.

[0088] A full-position process review requires placing both command-side evidence and receipt-side evidence into the same item structure. The item structure needs to be organized around the process token and maintain a consistent standard in the training process node identifier dimension.

[0089] For each process token, a process evidence chain entry sequence is established. Each process evidence chain entry includes at least the training process node identifier, process token, event category, event time, evidence payload, summary of the preceding entry, and summary of the current entry. The training process node identifier is limited to the set of training process node identifiers output in step S1. The process token is limited to the set of process tokens output in step S1 and the set of derived process tokens output in step S2. The event category is limited to token generation event, token consumption event, adjudication update event, receipt assembly event, completion proof generation event, receipt completion request generation event, and comparison query request generation event. The event time is limited to the timestamp sequence generated by the unified time source of the training session and satisfies monotonicity and non-reversal within the same process token. The evidence payload is limited to the standardized encoding result that corresponds one-to-one with the event category and includes the semantic skeleton identity degree, closure evidence net fit degree, replay adjudication coefficient, adjudication result output in step S2, and the summary of the closure evidence set, the summary of the missing receipt category set, the summary of the conflict key set, and the node completion proof record output in step S3.

[0090] S402 entry triggers the acquisition and evidence payload assembly rules.

[0091] The process evidence chain needs to cover the key turning points from token generation to the completion of proof. The trigger points and payload assembly rules need to correspond one-to-one with the output objects of steps S1 to S3 to avoid missing pages of evidence during review.

[0092] The token generation event is triggered in step S1 when the process token is written to the process token binding table, and writes the training process node identifier, process token, node metadata specification encoding summary, and collision resolution marker summary. The adjudication update event is triggered in step S2 each time an adjudication result is output, and writes the semantic skeleton string summary, semantic fingerprint summary, execution fingerprint summary, semantic skeleton identity, closure evidence net fit, replay adjudication coefficient, adjudication result, and specification basis identifier set. The specification basis identifier set is limited to the following values: reference system mapping version identifier, unit domain mapping version identifier, stripped field path set summary, retained field path set summary, and channel adaptation table version identifier. The three-part adjudication boundary version identifier is provided by the training organization specification or subject configuration file; the receipt assembly event is triggered after the closure evidence assembly and consistency verification are completed in step S3 and the assembly process token, closure evidence set summary, missing receipt category set summary, conflict key set summary, and conflict source receipt category set summary are written; the proof completion generation event is triggered after the proof completion record is generated in step S3 and the proof completion record and proof specification code summary are written; the receipt completion request generation event and the comparison query request generation event are triggered when the receipt completion request and the comparison query request are generated in step S3 and the request summary and the request target set summary are written.

[0093] Example: In the training process node of UAV route issuance, after the process token is written into the binding table, a token generation event entry is formed. During the link recovery phase, two inbound commands trigger two adjudication update event entries and record the adjudication results as merged and completed. When the confirmation is missing, the receipt assembly event entry and the receipt completion request event entry are generated. When the confirmation is completed and the closure completion condition is met, the completion proof event entry is triggered and written into the node completion proof record.

[0094] S403 Current entry summary calculation, chain constraints, and session root summary anchoring.

[0095] The process evidence chain needs to resist entry insertion and entry tampering. The chain constraint relies on the previous entry summary to lock the writing order. The session root summary is used to anchor the entire sequence of entries into a single summary for the evaluation side to refer to.

[0096] The rules for determining the value of the preceding entry summary are as follows: the first entry under the same process token uses a preset zero summary, which is limited to a fixed-length string of all zero bits with the same length as the anti-collision hash output. For entries other than the first entry, the current entry summary of the previous entry is used. The calculation order of the current entry summary is as follows: first, perform canonical encoding on the process evidence chain entries, then concatenate it with the preceding entry summary, and finally perform anti-collision hashing. The canonical encoding field order is fixed as training process node identifier, process token, event category, event time, evidence payload, and preceding entry summary, and adopts a deterministic serialization format. The anti-collision hash output is used as the current entry summary and written into the process evidence chain entries. The session root summary anchor aggregates the current entry summaries of all process tokens according to the training session identifier. The aggregation order is determined by the joint sorting key of the training process node identifier and the event time and written into the sorting key version identifier. Then, Merkle merging is performed on the sorted current entry summary sequence. The Merkle merging calculation order is as follows: adjacent two summaries are first concatenated and then hashed to obtain the upper-level summary sequence, and this process is iterated until a single session root summary is obtained. The session root summary is written into the session anchor entry and saved together with the training session identifier.

[0097] Merkle merge algorithm is the process of constructing and merging Merkle trees. It works by concatenating the bottom-level digests in pairs in a fixed order, then hashing them to obtain the next-level digest. This "pairwise merge + hash" process is repeated until only one root digest remains. This root digest is the integrity fingerprint of the entire batch of data.

[0098] S404 Retrospective Query and Training Evaluation Attribution Input Construction.

[0099] The review side focuses on the adjudication trajectory, closure trajectory, completion resolution trajectory, and node completion proof record under the node identifiers in the training process. The evaluation side needs node-level attribution input records that can be directly referenced and aligned with the session root summary.

[0100] The recap query input includes training process node identifiers and process tokens or assembly process tokens. The query process starts with the current entry summary and backtracks backward according to the previous entry summaries to obtain the entry sequence, which is then rearranged according to the event time to form a node-level timeline. In the node-level timeline, the adjudication update event entry outputs the replay of the adjudication coefficient update sequence and the adjudication result change point to form an adjudication trajectory summary. The receipt assembly event entry outputs the missing receipt category set summary and the conflict key set summary to form a closure trajectory summary. The receipt completion request generation event entry and the comparison query request generation event entry output the request target set summary and the completion mark to form a completion resolution trajectory summary. The completion proof generation event entry outputs the node completion proof record and the proof specification code summary to form a completion proof summary. The training evaluation attribution input record is constructed with the training process node identifier as the primary key and includes process tokens, assembly process tokens, final adjudication results, adjudication trajectory summaries, closure trajectory summaries, completion resolution trajectory summaries, node completion proof records, and session root summaries. The training evaluation side performs cross-position alignment and attribution explanation output based on this.

[0101] The process evidence chain records all lifecycle event entries with process tokens as the primary key and constrains the order consistency of entries with chained summaries. The training session evidence set is anchored with the session root summary. During review, the adjudication trajectory, closure trajectory, completion resolution trajectory and node completion proof record are traced back according to the training process node identifier, so that training evaluation and attribution are aligned based on the same evidence standard.

[0102] Specifically, the above are merely preferred embodiments of this application and are not intended to limit this application.

[0103] The preset thresholds or parameters can be pre-calibrated through offline simulation testing or set to fixed values ​​according to the on-site operating procedures.

[0104] In the description of this specification, references to terms such as "an embodiment," "example," and "specific example" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0105] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for equipment simulation training modeling oriented towards the entire combat station process, characterized in that, Including the following steps: S1: Decompose the training process into training process nodes and configure node metadata, generate process tokens that bind node metadata, and the node metadata includes a set of state summary keys and an expected receipt closure structure. Step S1 further includes: training process node metadata includes a set of instruction types, target entity identifiers, authorization domain constraints, a set of status summary keys, and an expected receipt closure structure. The target entity identifier is obtained by mapping the instruction action object to the equipment entity directory. The authorization domain constraints are obtained by encoding the authorization domain sequence using the battle station permission matrix. The set of status summary keys is extracted and reduced from the instruction type to status field mapping table. The expected receipt closure structure is constructed by the receipt category mapping and dependency rules and obtained through acyclic verification. The process token is generated by hashing the node metadata in a standardized manner and is kept unique through collision detection and collision resolution. S2: The semantic skeleton is formed by semantic specification of inbound instructions and dual-domain fingerprinting is used to obtain semantic skeleton identity. The receipt is mapped to the process token and status digest key to obtain closure evidence net fit. The two are sequentially merged to update the replay adjudication coefficient and output the adjudication result. Step S2 further includes: after the inbound instruction record enters the queue to be judged, semantic reduction is performed to generate a semantic skeleton string. The semantic reduction performs reference frame reduction on the spatial position field and attitude field and unit domain reduction on the velocity field and distance field. The semantic reduction removes the retransmission identifier and message sequence number and retains the control field to form a semantic skeleton string. The semantic skeleton string is subjected to dual-domain fingerprinting to generate a semantic fingerprint and an execution fingerprint, and the semantic skeleton identity is obtained by comparing it with the historical received instruction fingerprint. The set of received receipts under the same process token is mapped to the expected receipt closure structure as a closure evidence record according to the status summary key set. The closure evidence record calculates the net fit of the closure evidence and registers the conflict key and conflict source receipt category. Sequential evidence fusion is performed with the semantic skeleton identity and the net fit of the closure evidence to obtain the replay decision coefficient and output the discard decision, merge and complete decision, or resign token derived decision. S3: When the adjudication result allows the node to advance or merge and complete the advancement, the multi-channel receipts are assembled according to the process token and the status summary key is aligned to form a closed evidence set. When the closed evidence set meets the expected receipt closure structure, a node completion proof is generated; otherwise, receipt completion is triggered. S4: Write the lifecycle of the process token into the process evidence chain entry. Calculate the entry summary in a chained manner according to the summaries of the preceding entries and merge them to generate the session root summary. Output the process evidence chain entry.

2. The equipment simulation training modeling method for all combat positions according to claim 1, characterized in that, Step S1 includes: The training process node template is generated from the training subject script, the combat position password sequence, and the instruction interface contract. The event syntax is segmented according to the instruction type to trigger the synchronous password closure to form the training process node sequence. The training process node identifier is taken from the script node number or the result of concatenating the synchronous password identifier and the node sequence mark. The training process node intent label is determined by the task paragraph title and the type of the first instruction in the node.

3. The equipment simulation training modeling method for all combat positions according to claim 2, characterized in that, Step S2 includes: The inbound instruction record extracts the instruction type, target entity identifier, authorization domain identifier, instruction field set, arrival channel identifier, and arrival time, and enters the pending judgment queue according to the process token. When the inbound instruction record is missing a process token, it locates the unique process token in the process token binding table according to the instruction type, target entity identifier, and authorization domain constraint. If the location fails, it outputs a resigned token derivation decision.

4. The equipment simulation training modeling method for all combat positions according to claim 3, characterized in that, Step S3 includes: The assembly process token is determined according to the ruling result, and the expected receipt closure structure and status summary key set are read from the process token binding table. The set of arrived receipts corresponding to the assembly process token is normalized by the execution channel adaptation table, and a receipt index and a secondary index of status summary key are established according to the receipt category. When there are multiple receipt records for a receipt category, the main receipt record is determined according to the arrival time, and the remaining receipt records are written into the duplicate receipt list.

5. The equipment simulation training modeling method for all combat positions according to claim 4, characterized in that, Step S3 also includes: According to the expected closure structure of the closure, the main closure record is extracted from the closure index to form a closure evidence set and the missing closure category set is located. The closure evidence set is aggregated into a state summary value set according to the state summary key and the conflict key set and conflict source closure category set are located. When the missing closure category set is empty and the conflict key set is empty, a node completion proof record is generated. When the missing closure category set is not empty, a closure completion request is generated. When the conflict key set is not empty, a comparison query request is generated and the retention or unresolved mark is executed according to the consistency of the comparison summary value.

6. The equipment simulation training modeling method for all combat positions according to claim 5, characterized in that, Step S4 includes: The process evidence chain uses the process token as the primary key to continuously write process evidence chain entries. Each process evidence chain entry uniformly carries the full lifecycle evidence corresponding to the training process node identifier and maintains the consistency of event timing. The writing process adopts fixed standard encoding and synchronously associates nodes to complete the proof record, so that the same training process node forms a consistent evidence caliber under multi-channel playback conditions.

7. The equipment simulation training modeling method for all combat positions according to claim 6, characterized in that, Step S4 also includes: The current entry summary is generated by concatenating the previous entry summary and the standardized encoding of the process evidence chain entries. All current entry summaries in the training session are merged in a deterministic order to form the session root summary. During the debriefing phase, the adjudication trajectory and closure trajectory are traced back based on the training process node identifiers, and the node completion proof records are extracted.

Images