Neural network-based auxiliary simulation training scene command system and command method thereof

By using a neural network-based simulation training scenario command system, the instability problem of traditional decision support systems when faced with massive and rapidly changing data has been solved, achieving efficient and accurate decision support and situation prediction, and improving the scientific nature and efficiency of command and decision-making.

CN119005754APending Publication Date: 2024-11-22XIAN LINGKONG ELECTRONICS TECH CO LTD
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411263557.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Traditional decision support systems rely on the accumulated experience and intuitive judgment of experts. When faced with massive and rapidly changing scenario data, their processing capabilities become limited, leading to unstable and uncertain decision results.

Method used

The system employs a neural network-based simulation training scenario command system, which includes data acquisition, preprocessing, a neural network module, and a decision support module. It extracts scenario features and situations through a neural network model, and provides decision suggestions in conjunction with a user interaction module, supporting users in adjusting decision weights.

Benefits of technology

It improves the scientific and rational nature of decision-making, reduces the decision-making burden on commanders, and can provide early warning of potential threats in complex and ever-changing environments, offering rapid and accurate decision support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005754A_ABST
    Figure CN119005754A_ABST
Patent Text Reader

Abstract

The invention discloses an auxiliary simulation training scene command system based on a neural network and a command method thereof, and relates to the technical field of computer simulation training, a data acquisition module is configured to integrate and collect scene environment information from various sensors to form scene data; the data preprocessing module is configured to preprocess the scene data and provide input data for the neural network model; the neural network module is configured to extract scene features from the scene data through a neural network model, judge and predict a scene situation, and output the scene situation; the decision support module is configured to formulate decision suggestions in combination with the scene situation output by the neural network model; the user interaction module is configured to display scene situations and decision suggestions through an intuitive interface, provide an interaction interface between a user and the system, facilitate a commander to understand and execute decisions, and solve the problem that a traditional decision support system is affected by subjective cognitive deviation, resulting in instability and uncertainty of decision results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer simulation training technology, and in particular to a neural network-assisted simulation training scenario command system and a command method thereof. Background Art

[0002] In modern society and in all walks of life, decision support systems are key tools for improving work efficiency and coping with complex challenges. They are designed to assist decision makers in extracting valuable information from massive amounts of data and conducting fast and accurate decision analysis. Decision support systems provide decision makers with scientific decision-making basis and solution suggestions through automated processing, optimization algorithms and intelligent reasoning mechanisms, thereby improving decision efficiency and accuracy.

[0003] In today's complex and ever-changing scenarios, accurate and efficient decision support is crucial to improving action effectiveness and response capabilities. Traditional decision support systems rely heavily on the experience accumulation and intuitive judgment of experts. Although these valuable resources are indispensable, their processing capabilities are gradually limited when faced with massive and rapidly changing scenario data. They are also affected by subjective cognitive biases, resulting in instability and uncertainty in decision results. Summary of the invention

[0004] In the embodiments of the present application, by providing a neural network-assisted simulation training scenario command system and its command method, the problem of traditional decision support systems relying more on the experience accumulation and intuitive judgment of experts is solved. Although these valuable resources are indispensable, when faced with massive and rapidly changing scenario data, the limitations of their processing capabilities gradually become apparent, and they are affected by subjective cognitive biases, leading to the problem of instability and uncertainty in decision results.

[0005] In the first aspect, an embodiment of the present application provides a scenario command system based on a neural network-assisted simulation training, comprising: a data acquisition module, a data preprocessing module, a neural network module, a decision support module and a user interaction module; the data acquisition module is configured to integrate and collect scenario environment information from a variety of sensors and communication devices to form comprehensive scenario data; the data preprocessing module is configured to preprocess the scenario data to ensure data quality and consistency, and provide input data for the neural network model; the neural network module is configured to extract scenario features from the scenario data through the neural network model, judge and predict the scenario situation, and output it; the decision support module is configured to formulate decision recommendations based on the scenario situation output by the neural network model, and support users to adjust decision weights according to actual needs; the user interaction module is configured to display the scenario situation and decision recommendations through an intuitive interface, and provide an interactive interface between users and the system to facilitate commanders to understand and execute decisions.

[0006] On the second aspect, the embodiment of the present application provides a command method based on a neural network-assisted simulation training scenario command system, including executing an information collection step in a data acquisition module, specifically including: collecting and analyzing the deployment, resource distribution and dynamic changes of both parties in the scene; obtaining the scene weather conditions in real time through a meteorological monitoring and forecasting system, and predicting the scene weather environment within a preset time period in the future in advance; collecting the geographical environment information of the scene.

[0007] In one possible implementation, a preprocessing step is performed in a data preprocessing module, specifically including: screening the scene data to ensure that each scene data completely contains data information and fields, and ensuring that all scene data conform to preset specifications and standards; converting diverse scene data from different sensors into a unified format; denoising the scene data, completing missing scene data, and ensuring the continuity and integrity of the scene data on the timeline; performing dimensionality reduction on the scene data, and encrypting sensitive data.

[0008] In a possible implementation manner, the method further includes: recording and organizing the preprocessed scene data according to a time axis.

[0009] In a possible implementation, the prediction step is performed in a neural network module, specifically including: designing a network structure of a neural network model; wherein the network structure includes an input layer, a hidden layer, and an output layer; implementing an incremental learning strategy and adversarial neural network training in the neural network model; wherein the incremental learning strategy is determined based on the task type; the adversarial neural network training includes: setting the goal and strategy of adversarial training, and constructing a generator and a discriminator; by alternately training the generator and the discriminator, improving the authenticity of the generated synthetic data and the discrimination ability of the discriminator, and enhancing the generalization ability of the neural network model by using the generated synthetic data; using a training data set to iterate the neural network model for multiple times and monitor the performance of the neural network model; wherein the training data set includes the generated synthetic data; adjusting hyperparameters according to the performance of the neural network model; wherein the hyperparameters include a learning rate and a batch size; using a validation set to evaluate the performance of the neural network model to ensure that the neural network model maintains accurate and reliable prediction capabilities when encountering unknown or new data; using an independent test set to comprehensively evaluate the neural network model to verify its accuracy and usability; optimizing the neural network model according to the evaluation results, and deploying the optimized neural network model to the command system to achieve real-time scenario situation prediction and decision-making recommendations.

[0010] In one possible implementation, when the task type is a new task, the incremental learning strategy is a regularization and / or replay mechanism; the regularization includes: identifying the differences between the new task and the old task, and adding a regularization term to the loss function to ensure that the neural network model does not forget the old knowledge while learning new knowledge; the replay mechanism includes: extracting samples from the data set of the old task, fusing them with the new data, and then training them in timeline order, to ensure that the neural network model reviews the old knowledge while learning new knowledge.

[0011] In one possible implementation, when the task type is an old task, the changed task parameters are determined, and a parameter isolation method is used as an incremental learning strategy. The parameter isolation method includes: dividing the neural network model parameters into multiple subsets, each subset is associated with a task type; during the training process, only the parameter subset related to the task type is updated, and the parameters of other task types are kept unchanged.

[0012] In a possible implementation, when the amount of scene data collected is less than a preset threshold, the neural network model is trained by semi-supervised learning and self-supervised learning, specifically including: organizing the scene data into a data set, and dividing the data set into a labeled data set and an unlabeled data set; in the early stage of neural network model training, semi-supervised learning is used for preliminary training, and when the preliminary training is completed, the training method is gradually transitioned from semi-supervised learning to self-supervised learning; the semi-supervised learning method includes: using a small amount of labeled data and a large amount of unlabeled data to train the neural network model, and gradually reducing the dependence on labeled data during the training process; the self-supervised learning method includes: using an unlabeled data set to train the model, and generating new training sample data based on the results predicted by the neural network model and the decision weights, for continuous optimization of the neural network model; after completing the training of the neural network model, the accuracy of the neural network model is verified by sampling verification and new data testing, and the results are analyzed to adjust the decision weights to optimize the neural network model.

[0013] In a possible implementation, the decision steps are executed in a decision support module, specifically including: selecting a method for constructing a decision tree according to the urgency of the current decision task; wherein the methods for constructing a decision tree include a dictionary construction method, a database construction method, and a decision model construction method; integrating the constructed decision tree into the decision support module; when a new decision task is received, calling a corresponding decision tree according to input scenario conditions and decision weights, and the decision tree outputs a decision result; performing statistical analysis on the output decision results; wherein the statistical analysis methods include frequency distribution statistics and normal distribution statistics, frequency distribution statistics are used to monitor and fine-tune the accuracy of the decision tree, and normal distribution statistics are used to detect whether the setting of the decision weight is reasonable; based on the statistical analysis, comparing the actual decision with the decision result, and using user feedback for supervised learning and real-time updating of the decision tree.

[0014] In a possible implementation, an interactive display step is performed in a user interaction module, specifically including: responding to information manually entered by the user to supplement or correct the collected scene data; using a map as a background to display the location, scale and movement of the forces of both parties; marking various symbols and charts on the map to intuitively display the overall scene situation; supporting users to adjust decision weights according to actual needs; displaying decision suggestions output by the decision support module to users; supporting the setting of different user roles, and allocating corresponding permissions and functions to user roles according to needs.

[0015] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects: The embodiment of the present application provides a command system based on a neural network-assisted simulation training scenario, which can efficiently integrate massive real-time data from a variety of sensors and communication devices, and ensure the integrity, accuracy and consistency of the data through a sophisticated data preprocessing process, so that the system can more accurately train the real situation of the scenario and provide strong support for command decision-making. Using the neural network model, the key features in the scene can be automatically extracted to achieve rapid and accurate judgment and prediction of the scene situation, which not only improves the commander's ability to grasp the scene dynamics, but also can warn potential threats in advance in a complex and changing environment, providing a valuable time window for decision-making. Combined with the situation judgment and prediction results output by the neural network, the decision support module can quickly generate a variety of feasible decision-making plans, adjust the decision weights according to actual needs, and provide the best decision-making suggestions. This function greatly reduces the decision-making burden of the commander, improves the scientificity and rationality of the decision, and helps to make correct judgments at critical moments. The user interaction module displays complex scene situations and decision suggestions to the commander in a graphical and visual way by designing an intuitive and easy-to-use operation interface, greatly reducing the threshold for information understanding. At the same time, it provides a convenient interactive interface, allowing commanders to easily exchange information with the system, realize the rapid issuance and execution of decisions, and improve the smoothness and efficiency of the command process. It solves the problem that traditional decision support systems rely more on the experience accumulation and intuitive judgment of experts. Although these valuable resources are indispensable, their processing capabilities are gradually limited when facing massive and rapidly changing scene data. They are affected by subjective cognitive biases, leading to instability and uncertainty in decision results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments of the present application or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A schematic diagram of the composition of a command system based on a neural network-assisted simulation training scenario provided in an embodiment of the present application; Figure 2 A specific flow chart of the information collection step performed in the data acquisition module provided in the embodiment of the present application; Figure 3 A specific flow chart of performing preprocessing steps in a data preprocessing module provided in an embodiment of the present application; Figure 4 A specific flow chart of executing the prediction step in the neural network module provided in the embodiment of the present application; Figure 5 A specific flow chart for training a neural network model by semi-supervised learning and self-supervised learning when the amount of scene data collected is less than a preset threshold provided in an embodiment of the present application; Figure 6 A specific flow chart of executing decision steps in a decision support module provided in an embodiment of the present application; Figure 7 A specific flow chart of performing interactive display steps in a user interaction module provided in an embodiment of the present application; Figure 8 A schematic diagram of a command server based on a neural network-assisted simulation training scenario provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] The following describes some of the techniques involved in the embodiments of the present application to facilitate understanding, and they should be considered as merely exemplary. Therefore, it should be appreciated by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, some descriptions of well-known functions and structures are omitted in the following description.

[0020] Figure 1 A schematic diagram of the composition of a command system based on a neural network-assisted simulation training scenario provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the neural network-assisted simulation training scenario command system includes a data acquisition module, a data preprocessing module, a neural network module, a decision support module and a user interaction module.

[0021] Specifically, the data acquisition module is configured to integrate and collect scene environment information from a variety of sensors and communication devices to form comprehensive scene data. The data preprocessing module is configured to preprocess the scene data, ensure data quality and consistency, and provide input data for the neural network model. The neural network module is configured to extract scene features from the scene data through the neural network model, judge and predict the scene situation, and output it. The decision support module is configured to formulate decision recommendations based on the scene situation output by the neural network model, and support users to adjust decision weights according to actual needs. The user interaction module is configured to display the scene situation and decision recommendations through an intuitive interface, and provide an interactive interface between the user and the system to facilitate commanders to understand and implement decisions.

[0022] The present application embodiment provides a command method based on a neural network-assisted simulation training scenario command system. Figure 2 A specific flow chart of the information collection step performed in the data acquisition module provided in the embodiment of the present application, such as Figure 2 As shown, it includes steps S101 to S103.

[0023] S101: Collect and analyze the deployment, resource distribution and dynamic changes of both parties in the scenario.

[0024] Specifically, the scenario in the present application is a simulated training scenario, and the two parties in the present application can be any two opposing entities or teams.

[0025] Deployment refers to the initial position arrangement or strategic layout of both parties in the simulation scenario, including the specific locations of the entities or team members of both parties in the simulation scenario, such as coordinate points or area divisions on a map. Defensive facilities that may be established by either party, such as bunkers, obstacles, bunkers, etc., to enhance defense capabilities or control key areas. Pre-planned or potential offensive routes show how one party attempts to approach or infiltrate the other party's area. According to the characteristics of the scenario and team strategy, the two parties arrange personnel configuration, equipment selection, etc. to support their overall strategic intentions. Resource distribution refers to the spatial distribution of various resources in the simulation scenario, which are critical for both parties to achieve their goals. Resources can include material resources, energy or score points, intelligence information, and terrain advantages.

[0026] S102: Obtain scene weather conditions in real time through a meteorological monitoring and forecasting system, and predict scene weather conditions within a preset time period in the future in advance.

[0027] Specifically, the meteorological monitoring and forecasting system may include meteorological sensors, meteorological satellites, and meteorological forecasting models. The preset time period may be within the next 24 hours.

[0028] S103: Collecting geographical environment information of the scene.

[0029] Specifically, the geographic environment information of the scene can be collected through geographic information technology tools. Geographic information technology tools include maps, satellite images, map data and geographic detectors, and the geographic environment information of the scene includes topography, roads, mountains, ditches, rivers, lakes and buildings.

[0030] Furthermore, all the data collected above can be recorded and organized according to the timeline, that is, all the collected data can be arranged in the order of their occurrence to construct a clear scene event timeline.

[0031] Figure 3 A specific flow chart of performing preprocessing steps in a data preprocessing module provided in an embodiment of the present application, such as Figure 3 As shown, it includes steps S201 to S204.

[0032] S201: Filter the scene data to ensure that each piece of scene data completely contains data information and fields, and ensure that all scene data conform to preset specification standards.

[0033] Specifically, the scenario data is screened to ensure that each scenario data contains all necessary data information and fields, including checking the integrity of the data fields, that is, each data record should have all the preset fields, and these fields should not be empty. At the same time, it is also ensured that all scenario data conforms to the preset specifications, including the uniqueness verification of the data primary key, the correctness of the data type, the legitimacy of the data value (such as length, size, value range, precision, etc.) and a comprehensive assessment of the scope integrity.

[0034] S202: Convert diverse scene data from different sensors into a unified format.

[0035] Diverse scene data from different sensors may exist in multiple formats such as JSON, XML, plain text or numbers. In order to ensure the consistency and efficiency of subsequent data processing, these diverse scene data need to be converted into a unified format. This application selects JSON (JavaScript Object Notation) as a unified data format because it is lightweight, easy to read and write. In the process of data formatting, special attention is paid to the importance of the timeline. Since scene data is usually closely related to time, when converting the data to JSON format, the timestamp or time range of each data is specially marked to ensure the accuracy and continuity of the data on the timeline. In this way, in subsequent data analysis or model training, the data can be easily sorted, filtered or aggregated according to the time order, so as to more accurately reflect the evolution of the scene situation.

[0036] S203: De-noising the scene data, and completing the missing scene data to ensure the continuity and integrity of the scene data on the time axis.

[0037] Specifically, denoising includes data cleaning and data filtering. The main purpose is to remove noise and outliers in the formatted data. The Kalman filter algorithm can be used to denoise the scene data. The algorithm optimizes the estimated value of the data through two steps of prediction and update, and effectively filters out random noise. In addition, for some obvious erroneous data, it can be directly identified and replaced by calculating the mean, median or other statistical methods. Due to the complexity and uncertainty of sensors, radars and other equipment in the scene environment, data missing or faults are inevitable during the data collection process. In order to ensure the continuity of the data on the time axis, the missing scene data needs to be supplemented. The key to supplementing data is to find a reasonable law of data change. In many cases, the changes in scene data (such as position, speed, direction, etc.) may follow linear or nonlinear trends. When the data change law conforms to the linear change rule, linear regression analysis (such as univariate linear regression or binary linear regression) can be used to predict and supplement the missing data points. This method establishes a mathematical model based on existing data points, and then uses the model to estimate the missing values. It should be noted that when choosing a completion method, the actual characteristics and application scenarios of the data should be fully considered to ensure the accuracy and reliability of the completed data.

[0038] S204: Perform dimensionality reduction processing on the scene data and encrypt the sensitive data.

[0039] Specifically, dimensionality reduction processing of scene data can improve computing efficiency. Dimensionality reduction aims to reduce the number of features in the data set while retaining important information of the original data as much as possible to facilitate subsequent data analysis and model training. According to the characteristics and requirements of the data, select dimensionality reduction techniques such as principal component analysis (PCA), linear discriminant analysis (LDA), factor analysis (FA) or singular value decomposition (SVD). Apply the selected dimensionality reduction method to process the data, remove redundant and noisy data, and reduce the data dimension. Formulate encryption strategies based on the sensitivity and transmission requirements of the scene data. Use a variety of encryption algorithms (such as symmetric encryption algorithms DES, AES and asymmetric encryption algorithms RSA, DSA) in rotation to increase the difficulty of cracking and ensure the security of data transmission and storage. Encrypt sensitive data that needs to be protected (such as location information, communication content, etc.). According to the encryption strategy, select the appropriate encryption algorithm and key length to encrypt the data to ensure that the data cannot be easily stolen or tampered with during transmission and storage.

[0040] After executing the preprocessing step, the process also includes: recording and arranging the preprocessed scene data according to the time axis. It should be noted that the appropriate time accuracy can be selected according to the requirements of the application scenario. Too high time accuracy will increase the difficulty of data storage and processing, while too low time accuracy may not accurately reflect the dynamic changes of the data.

[0041] Figure 4 A specific flow chart of executing the prediction step in the neural network module provided in the embodiment of the present application is as follows: Figure 4 As shown, it includes steps S301 to S307.

[0042] S301: Design a network structure of a neural network model, wherein the network structure includes an input layer, a hidden layer, and an output layer.

[0043] Specifically, it is necessary to select a suitable neural network model according to the specific requirements of the task (such as image recognition, time series prediction, etc.). For example, if the task is to process and analyze image data in the scene (such as scene images taken by drones), a convolutional neural network (CNN) can be selected to take advantage of its powerful image feature extraction capabilities. If the task involves predicting time series data of situations (such as action trajectory prediction), a recurrent neural network (RNN) or its variants, long short-term memory networks (LSTMs) and deep neural networks (DNNs) can be selected to process such sequence data. When designing the network structure, it is necessary to determine the specific configuration of the input layer, hidden layer, and output layer. The input layer is responsible for receiving the original scene data (such as image pixel values, time series data, etc.) from the data acquisition module. The hidden layer contains multiple layers, each of which is composed of a certain number of neurons. These neurons gradually abstract and extract features from the input data through complex interconnections and calculations. The output layer generates the final prediction results or decision recommendations based on the task requirements.

[0044] In this application, a deep neural network (DNN) is used as the basic architecture, that is, the neural network model in this application is a deep neural network. By stacking multiple nonlinear processing layers, the deep neural network can learn high-level abstract features in the data, thereby more accurately capturing the changing patterns of the scene situation.

[0045] S302: Implementing incremental learning strategy and adversarial neural network training in the neural network model, wherein the incremental learning strategy is determined based on the task type.

[0046] When the task type is a new task, the incremental learning strategy is regularization and / or replay mechanism. Regularization includes: identifying the difference between the new task and the old task, and adding regularization terms to the loss function to ensure that the neural network model does not forget the old knowledge while learning new knowledge. The replay mechanism includes: extracting samples from the old task dataset, fusing them with the new data, and then training them in timeline, ensuring that the neural network model reviews the old knowledge while learning new knowledge.

[0047] When the task type is an old task, the changed task parameters are determined, and the parameter isolation method is used as the incremental learning strategy. The parameter isolation method includes: dividing the neural network model parameters into multiple subsets, each subset is associated with a task type. During the training process, only the parameter subset related to the task type is updated, and the parameters of other task types are kept unchanged.

[0048] Specifically, incremental learning allows the neural network model to continuously learn new tasks while maintaining the performance of old tasks. This is particularly important for scene data analysis because the scene environment is constantly changing and new situations are constantly emerging. Regularization methods are limited to feature alignment or spatial geometric structure matching between the features learned by individual samples in the new task and the response features of the original network, lacking a global vision. In order to consider the correlation between old and new knowledge from a holistic perspective, this scheme further adopts a replay mechanism based on regularization. The combination of the two also ensures that the model will not overfit and be simplistic. Regularization, as a key constraint, plays a vital role in the model update process. By adding regularization terms to the loss function, the model is guided to maintain sensitivity to old tasks while learning new knowledge, thereby avoiding forgetting old knowledge. Specifically, when the model faces new training data, the regularization term prompts the model to consider the performance of old tasks while optimizing the loss of new tasks, ensuring a balance between new and old knowledge. In addition, it is emphasized that all training data must be sorted and trained in timeline order to strengthen the model's understanding of the temporal and logical nature of scene situations, so that it can output corresponding results according to changes in scene situations. In order to further enhance the model's ability to retain old knowledge, a replay mechanism is introduced. When the model learns a new task, this mechanism randomly extracts samples from the old task data set, fuses these samples with the new data, and trains them after sorting them in chronological order. In this way, the model can review and consolidate old knowledge while being exposed to new data, thereby effectively preventing the loss of old knowledge. The replay mechanism not only helps reduce the model's forgetting of old data, but also alleviates the problem of model overfitting to a certain extent and improves the model's generalization ability.

[0049] Furthermore, in order to significantly improve the adaptability and flexibility of the neural network model in complex and changing task environments, a parameter isolation method is introduced. This method cleverly divides the huge parameter system of the neural network model into multiple independent but interrelated subsets based on factors such as user roles, task types, and specific scenarios. Each subset is carefully designed to specifically serve a set of tasks or roles associated with the task type, thereby achieving a one-to-one mapping relationship between parameters and tasks.

[0050] Specifically, during the implementation process, a detailed task and role analysis was first conducted to clarify which tasks or roles are relatively independent and which may share certain common features. Based on this analysis, the neural network model is divided into several subsets in terms of parameters, each of which carries specific information required to process the task type or user role. This division not only helps to reduce parameter redundancy, but also lays the foundation for efficient parameter management. When the neural network model is trained on a new task type, only the parameter subset related to the task type is updated, while the parameters of the task type remain unchanged. This approach reduces the possible mutual interference between different tasks and ensures that each task can be fully optimized without being affected by other tasks.

[0051] Adversarial neural network training includes: setting the goal and strategy of adversarial training, building the generator and discriminator. By alternately training the generator and the discriminator, the authenticity of the generated synthetic data and the discriminator's ability to distinguish are improved, and the generated synthetic data is used to enhance the generalization ability of the neural network model.

[0052] Specifically, in adversarial neural networks, the goal of the generator is to generate fake data that is as close to the real data as possible, while the goal of the discriminator is to distinguish between the real data and the fake data generated by the generator.

[0053] The expression of the generator's objective function is: .in, is the expected symbol, which means taking the average of all possible values ​​of the random variable z, where z is the noise randomly sampled from a prior distribution pz (such as normal distribution) and used as the input of the generator. It represents the fake data generated by the generator G that maps the noise z to the data space. The fake data generated by the discriminator D for the generator The logarithm of the predicted probability of . In the training process of the adversarial neural network, from the perspective of the generator, it hopes that the discriminator will be The predicted probability of is as high as possible, so it tries to maximize , but since we are actually minimizing the loss function when training the generator, a minus sign is added in front of it. is a hyperparameter that balances the relative importance of the two parts of the loss. The value of needs to be adjusted according to the specific task and dataset. is a custom loss function designed to introduce additional constraints or regularization terms to ensure that the generated fake data and a real data point that precedes it in time and a real data point that comes after it in time The dynamic consistency between them. The expression of the custom loss function is: .in, The dynamic feature extraction function is used to extract dynamic features from data points. These features include position, velocity, acceleration, momentum, and any other attributes that can describe the dynamic behavior of data points. represents the dynamic features extracted from the fake data points, represents the dynamic features extracted from the previous real data point. is the dynamic feature change, indicating the change from arrive The dynamic characteristics of the changes. Calculate the dynamic characteristics of the fake data points Dynamic characteristics of fake data that are predicted based on real data points before and after them The smaller this distance is, the more consistent the dynamic characteristics of the generated fake data are with the real data, thus achieving the evaluation of the dynamic consistency of the generated data. This loss value will be used to guide the training of the generator so that it can generate fake data that is more consistent with the dynamic characteristics of the real data.

[0054] The expression of the objective function of the discriminator is: .in, is the expected symbol, which means the weighted average of all possible values ​​of the real data x. The real data x is randomly drawn from the real data distribution pdata, and z is the noise randomly sampled from a prior distribution pz (such as normal distribution). It represents the fake data generated by the generator G that maps the noise z to the data space. is the logarithm of the predicted probability of the discriminator D for the real data x generated by the generator, This is fake data generated by the discriminator D The goal of the discriminator is to try to make As small as possible to distinguish fake data from real data.

[0055] S303: Perform multiple iterations of training on the neural network model using the training data set to monitor the performance of the neural network model, wherein the training data set includes the generated synthetic data.

[0056] Specifically, the neural network model is trained multiple times using a training dataset containing real data and synthetic data generated by an adversarial neural network. During this process, the generator continuously tries to generate more realistic data to deceive the discriminator, while the discriminator strives to improve its ability to distinguish between real data and synthetic data. This alternating training method can not only improve the authenticity of the generated data, but also enhance the robustness of the discriminator, and then use these generated synthetic data to enrich the training set, significantly improving the generalization ability of the neural network model.

[0057] S304: Adjusting hyperparameters according to the performance of the neural network model, wherein the hyperparameters include learning rate and batch size.

[0058] Specifically, the optimization of hyperparameters has an important impact on the convergence speed, training stability and final performance of the model. By fine-tuning these parameters, we aim to find the best configuration combination to further improve the prediction accuracy and efficiency of the model.

[0059] S305: Use the validation set to evaluate the performance of the neural network model to ensure that the neural network model maintains accurate and reliable prediction capabilities when encountering unknown or new data.

[0060] Specifically, the validation set consists of data that is different from the training set but has similar characteristics, and is used to test the predictive ability of the neural network model when faced with unknown or new data.

[0061] S306: Use an independent test set to fully evaluate the neural network model to verify its accuracy and usability.

[0062] Specifically, the test set contains a variety of possible scenarios to simulate the challenges in actual applications. Through this step, we can gain a comprehensive understanding of the model performance and discover possible problems or deficiencies.

[0063] S307: Optimize the neural network model according to the evaluation results, and deploy the optimized neural network model to the command system to achieve real-time scenario situation prediction and decision-making recommendations.

[0064] Figure 5 The specific flow chart of training the neural network model by semi-supervised learning and self-supervised learning is provided in the embodiment of the present application when the amount of scene data collected is less than a preset threshold, such as Figure 5 As shown, it includes steps S401 to S403.

[0065] Specifically, if the scene data is of high quality, low noise, and very useful for training a neural network model, the preset threshold can be appropriately lowered, because a small amount of high-quality data may be sufficient to start the training process. If the data quality is not high or there is a lot of noise, more data is needed to ensure the stability and accuracy of the model training, so a higher preset threshold should be set.

[0066] S401: Arrange the scene data into data sets, and divide the data sets into labeled data sets and unlabeled data sets.

[0067] Specifically, the collected scene data are organized into data sets, and are clearly divided into labeled data sets and unlabeled data sets according to the label conditions of the data.

[0068] S402: In the initial stage of neural network model training, a semi-supervised learning method is used for preliminary training. When the preliminary training is completed, the training method is gradually transitioned from the semi-supervised learning method to the self-supervised learning method.

[0069] The semi-supervised learning method includes: using a small amount of labeled data and a large amount of unlabeled data to train the neural network model, and gradually reducing the dependence on labeled data during the training process.

[0070] Specifically, due to the scarcity of labeled data, semi-supervised learning is used for preliminary training. This stage makes full use of a small amount of labeled data and a large amount of unlabeled data, gradually reducing the dependence on labeled data while using the information in unlabeled data to enhance the generalization ability of the model. As the training progresses, the model gradually shifts from relying on labeled data to using patterns and information in unlabeled data.

[0071] The self-supervised learning method includes: using unlabeled data sets to train the model, and generating new training sample data based on the predicted results and decision weights of the neural network model to continuously optimize the neural network model.

[0072] Specifically, when the training of the semi-supervised learning stage reaches a certain level and the model has been able to learn certain feature representations from unlabeled data, the training method gradually transitions to self-supervised learning. In the self-supervised learning stage, the model relies entirely on unlabeled data sets for training. By predicting a certain structure or pattern within the data (such as the rotation angle in the image, the contextual relationship in the text, etc.), the model can self-generate supervision signals, thereby achieving continuous learning under unsupervised conditions. At the same time, based on the results predicted by the neural network model and the preset decision weights, the system will generate new training sample data, which are further used to optimize and consolidate the performance of the model.

[0073] S403: After completing the training of the neural network model, verify the accuracy of the neural network model through sampling verification and data testing, and analyze the results to adjust the decision weights to optimize the neural network model.

[0074] Specifically, after completing the training of the neural network model, the accuracy of the model is comprehensively evaluated through sampling verification and data testing. This step is to test the performance of the model under different data conditions to ensure that it can complete the task stably and accurately in practical applications. Based on the verification results, the system will adjust and optimize the decision weights to further improve the predictive performance and adaptability of the model. Through continuous iterative training and verification, a neural network model that can both efficiently utilize limited data resources and have excellent predictive performance is finally obtained.

[0075] Figure 6 A specific flow chart of executing decision steps in a decision support module provided in an embodiment of the present application, such as Figure 6 As shown, it includes steps S501 to S505.

[0076] S501: Select a method for constructing a decision tree according to the urgency of the current decision task, wherein the method for constructing a decision tree includes a dictionary construction method, a database construction method, and a decision model construction method.

[0077] Specifically, when faced with extremely urgent decision-making tasks, such as sudden scene crises, major natural disasters, etc., immediate response is required to reduce losses. In such emergency situations, time is the most precious resource. Therefore, the dictionary construction method should be selected. The dictionary construction method is simple and fast, and is suitable for processing small-scale and fixed data sets. Since super-urgent tasks often involve known or highly predictable scenes, the corresponding decision paths can be quickly retrieved through predefined dictionaries. Advanced emergency tasks usually involve more complex scene situations or emergencies, and multiple factors need to be considered comprehensively to make decisions. Although time is tight, the decision-making process requires a certain degree of flexibility and adaptability. In this case, the decision model construction method can be selected, especially the classification decision tree. The classification decision tree can quickly classify complex scenes and provide decision suggestions based on historical data and expert knowledge. This method can not only ensure the efficiency of decision-making, but also take into account the accuracy and flexibility of decision-making. Ordinary decision-making tasks may involve daily scene management, resource allocation, etc. These tasks do not require immediate response, but require the decision results to have high accuracy and stability. For ordinary tasks, the database construction method can be selected. The database can store a large amount of historical data and real-time information, and support complex queries and relationship management. By building a relational database, we can systematically manage scenario data and build a decision tree based on this data. Although this method is relatively complex to build, it can ensure the comprehensiveness and accuracy of the decision results.

[0078] S502: Integrate the constructed decision tree into the decision support module.

[0079] S503: When a new decision task is received, the corresponding decision tree is called according to the input scenario conditions and decision weights, and the decision tree outputs the decision result.

[0080] Specifically, when the decision support module receives a new decision task, it will call the corresponding decision tree for calculation based on the input scenario conditions (such as scenario situation, resource allocation, etc.) and the preset decision weights. After comprehensively considering various factors, the decision tree outputs the optimal decision result for the commander's reference.

[0081] Furthermore, the decision support module receives a new decision task, which is about how to allocate limited air support resources in a complex environment to maximize the strike effect. This task requires the system to make decisions based on multiple scenario conditions such as the current scenario situation, enemy activity mode, our personnel location and status, and preset decision weights (such as mission urgency, target value, our loss risk, etc.). The scenario condition input can be as follows. 1. Scenario situation. The enemy activities in the current area are frequent, and multiple important target points (such as command posts) are threatened. 2. Resource allocation. Air support resources are limited, including drones. 3. Our personnel location and status. We are scattered in multiple areas and urgently need air support. The preset decision weights can be as follows. 1. The mission urgency is high. 2. Target value. Sort each target point according to its strategic significance to us. 3. Our loss risk. Consider the risk of our personnel exposure that may be caused by sending air support. The decision process includes: The decision support module first parses the input scenario conditions and decision weights and converts them into a format that the decision tree can understand. According to the task type and input conditions, the module selects and calls a pre-built decision tree specifically designed to handle such resource allocation problems. This decision tree can be trained based on historical data and expert knowledge, and can comprehensively consider various factors and give the optimal allocation plan. After receiving the input, the decision tree begins to traverse its nodes level by level. At each node, it will evaluate and compare according to the current conditions (such as scenario situation, resource status, etc.) and preset weights (such as task urgency, target value, etc.), and finally find the optimal decision path. After a series of calculations and reasoning, the decision tree outputs one or more optimal decision results. In this example, the decision result can be a specific suggestion on how to allocate air support resources, including which type of air support to send, which target points to support first, and how to adjust the support strategy to minimize our losses. Finally, these decision results are presented to the commander for reference. The commander can adjust or confirm the execution of the decision results based on the actual situation and his own judgment.

[0082] S504: Perform statistical analysis on the output decision results. The statistical analysis methods include frequency distribution statistics and normal distribution statistics. Frequency distribution statistics are used to monitor and fine-tune the accuracy of the decision tree, and normal distribution statistics are used to detect whether the setting of decision weights is reasonable.

[0083] Specifically, frequency distribution statistics can identify in which cases the decision tree performs well and in which cases there may be deviations or deficiencies by analyzing the frequency of occurrence of decision results under different scenario conditions. Based on the results of frequency distribution, the accuracy of the decision tree can be evaluated and the poorly performing parts can be fine-tuned, such as adjusting the node division of the decision tree, modifying the decision rules, or adding new decision paths. Normal distribution statistics assumes that the decision results or a key indicator (such as decision benefits, risk levels, etc.) should follow a normal distribution, and the system can analyze whether the actual results meet this assumption. The characteristics of normal distribution (such as mean, standard deviation) can reflect the balance and stability of decision weights. If the actual results deviate from the expectations of normal distribution, it means that the setting of decision weights is not reasonable or there are deviations. Based on the results of normal distribution statistics, the distribution of decision weights can be re-evaluated and adjusted to ensure a more balanced and effective decision process.

[0084] S505: Based on the statistical analysis, the actual decision is compared with the decision result, and the user feedback is used for supervised learning and real-time updating of the decision tree.

[0085] Specifically, the output decision results are compared with the actual decision, that is, the decision that is actually implemented, and the differences and reasons between the two are analyzed. This comparison helps to evaluate the effectiveness and reliability of the decision support module in practical applications.

[0086] Figure 7 A specific flow chart of performing interactive display steps in a user interaction module provided in an embodiment of the present application, such as Figure 7 As shown, it includes steps S601 to S606.

[0087] S601: In response to the user manually entering information, to supplement or correct the collected scene data.

[0088] S602: Use the map as a background to show the location, size and movement of the forces of both sides.

[0089] Specifically, users can clearly see the overall layout and detailed changes of the scene through operations such as zooming and panning the map. At the same time, the system will also mark the key targets, routes and other information of both parties on the map to help users quickly understand the situation of the scene.

[0090] S603: Draw various symbols and charts on the map to intuitively display the overall scene situation.

[0091] Specifically, in order to intuitively display the overall situation of the two sides (assuming that they are the blue team and the red team, representing any two opposing entities or teams) in the simulated training scenario, the system will draw various symbols and charts on the map, as shown in the following examples. The initial positions of the blue team and the red team are marked on the map by flags or icons of different colors, such as blue flags for the blue team and red flags for the red team. Next to each flag is the approximate coordinate point or area name of the team on the map. The defensive facilities established by both sides, such as bunkers, obstacles, and imaginary bunkers, are marked on the map with corresponding symbols (such as semicircles for bunkers, rectangles with slashes for obstacles, and small fortress shapes for bunkers) and colors (corresponding to the team colors) to show their distribution and coverage. Lines with arrows or paths of different widths are used to represent the pre-planned or potential attack routes of both sides. The attack route of the blue team is marked with blue arrows or paths, and that of the red team is marked with red. These lines can show the general direction and strategy of both sides in trying to approach or infiltrate each other's areas. Key defensive areas are highlighted with special marks (such as a circle with an asterisk), indicating that these areas are the focus of contention between the two sides or require key defense. Although specific personnel and equipment models may not be marked directly on the map, the density of icons or marks can imply the deployment of both sides in specific areas. Material resources, energy or score points, intelligence information, etc. are represented on the map with specific icons, such as treasure chest icons for material resources, lightning icons for energy points, and eye icons for intelligence information. The locations of these resource points are clearly marked to show their spatial distribution. Terrain advantages are shown through a combination of natural features of the map (such as highlands, rivers, forests, etc.) and man-made facilities (such as bridges and roads). For example, highlands are represented by color gradients of varying shades, and bridges and roads are marked with specific lines to reflect their impact on the actions of both sides.

[0092] S604: Support users to adjust decision weights according to actual needs.

[0093] Specifically, the decision weight is a preset decision weight, which includes but is not limited to the following aspects: 1. Task urgency. The system presets a higher weight for "task urgency", which means that when evaluating decision plans, those options that can respond quickly and complete the task will be given priority. Users can adjust this weight according to actual conditions, such as appropriately reducing its weight when the task time window is expanded; otherwise, it will be increased. 2. Target value. The system will preliminarily sort the importance of each target point to our strategic significance and assign corresponding weights to them. Users can readjust the weights of these target points according to the change in the priority of strategic goals. For example, if a target point that originally seemed to be secondary suddenly becomes a key breakthrough, the user can increase its weight accordingly to ensure that the decision plan can fully consider this. 3. Our risk of loss. When evaluating decision plans, the system will also consider the risk of our personnel being exposed due to dispatching air support or other actions, and set a certain weight for it. Users can dynamically adjust this weight according to changes in the scenario situation, such as the enhancement of the enemy's air defense capabilities or the improvement of our concealment measures. If the user believes that the current loss risk is high, the weight can be appropriately increased to prompt the system to generate a more conservative or safe decision-making plan.

[0094] S605: Display the decision suggestions output by the decision support module to the user.

[0095] Specifically, the decision suggestions output by the decision support module and the results of its detailed statistical analysis can be presented to users clearly and intuitively. In the decision suggestion display part, the system will present the optimal or recommended decision-making scheme in a way that is easy for users to understand (such as charts, text descriptions, etc.). These schemes will elaborate on the action steps, expected results, and possible risks and challenges to help users quickly grasp the key points of the decision. At the same time, the system will also conduct in-depth statistical analysis of the decision suggestions, including but not limited to the benefit evaluation, success rate prediction, resource consumption comparison, etc. of each decision scheme. These statistical data are presented in the form of charts, curves, etc., which not only allows users to intuitively see the differences between different schemes, but also helps them better understand the logic and basis behind the decision. Through this display method, users can not only obtain specific decision suggestions, but also have a more comprehensive and in-depth understanding of the decision-making process, so as to make more wise and scientific decisions.

[0096] S606: Support the setting of different user roles, and assign corresponding permissions and functions to the user roles according to requirements.

[0097] Specifically, in the user role management part, the system allows administrators to define multiple user roles, such as commanders, staff, executives, etc., according to organizational structure and business needs. Each role represents different responsibilities and authority scopes, ensuring the standardization and orderliness of system use. In terms of authority allocation, the system will allocate corresponding functional modules and operation permissions to different user roles. For example, the commander can have the authority to view all decision results, adjust decision weights, and assign tasks; while the staff can be mainly responsible for auxiliary work such as data analysis and program formulation; and the executive can only have the authority to view and execute specific tasks. Through such a user role management and authority allocation mechanism, the system not only ensures the accurate transmission and efficient use of information, but also effectively avoids the risk of authority abuse and sensitive information leakage. At the same time, it also provides users with a more personalized and flexible user experience, so that users of different roles can find a working method and operation interface that suits them in the system.

[0098] The devices or modules described in the above application embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described in various modules according to their functions. When implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware. Of course, the module that implements a certain function can also be implemented by combining multiple sub-modules or sub-units.

[0099] The methods, devices or modules described in this application can be implemented in the form of computer-readable program codes. The controller can be implemented in any appropriate manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program codes (such as software or firmware) that can be executed by the (micro)processor, logic gates, switches, application-specific integrated circuits (English: Application Specific Integrated Circuit, referred to as: ASIC), programmable logic controllers and embedded microcontrollers. Examples of controllers include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program codes, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included in it for implementing various functions can also be regarded as structures within the hardware component. Or even, the means for realizing various functions may be regarded as both a software module for realizing the method and a structure within a hardware component.

[0100] like Figure 8 As shown, an embodiment of the present application also provides a command server based on a neural network-assisted simulation training scenario, including a memory 801 and a processor 802; the memory 801 is used to store computer-executable instructions; the processor 802 is used to execute computer-executable instructions to implement the command method based on the neural network-assisted simulation training scenario command system described above in the embodiment of the present application.

[0101] An embodiment of the present application also provides a computer-readable storage medium, which stores executable instructions. When a computer executes the executable instructions, it can implement the command method based on the neural network-assisted simulation training scenario command system described above in the embodiment of the present application.

[0102] It can be seen from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, or it can be reflected in the implementation process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the method described in the embodiment of the present application.

[0103] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. All or part of this application can be used in many general or special computer system environments or configurations.

[0104] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present application.

Claims

1. A command system based on neural network-assisted simulation training scenario, characterized in that: include: Data acquisition module, data preprocessing module, neural network module, decision support module and user interaction module; The data acquisition module is configured to integrate and collect scene environment information from a variety of sensors and communication devices to form comprehensive scene data; The data preprocessing module is configured to preprocess the scene data to ensure data quality and consistency and provide input data for the neural network model; The neural network module is configured to extract scene features from scene data through a neural network model, judge and predict scene situations, and output them; The decision support module is configured to formulate decision suggestions based on the scenario situation output by the neural network model, and support users to adjust decision weights according to actual needs; The user interaction module is configured to display scene situations and decision suggestions through an intuitive interface, and provide an interactive interface between the user and the system to facilitate commanders to understand and implement decisions.

2. The command method based on the neural network assisted simulation training scenario command system according to claim 1, characterized in that: The step of collecting information is included in the data acquisition module, specifically including: Collect and analyze the deployment, resource distribution and dynamic changes of both parties in the scenario; Obtain scene weather conditions in real time through the meteorological monitoring and forecasting system, and predict scene weather conditions within a preset time period in the future in advance; Collect the geographic environment information of the scene.

3. The command method based on the neural network assisted simulation training scenario command system according to claim 1, characterized in that: The preprocessing steps are performed in the data preprocessing module, including: Screen the scene data to ensure that each scene data contains complete data information and fields, and ensure that all scene data conform to the preset specification standards; Convert diverse scene data from different sensors into a unified format; De-noise the scene data and complete the missing scene data to ensure the continuity and integrity of the scene data on the timeline; Perform dimensionality reduction on scene data and encrypt sensitive data.

4. The command method based on the neural network assisted simulation training scenario command system according to claim 3 is characterized in that: Also includes: The preprocessed scene data is recorded and organized according to the timeline.

5. The command method based on the neural network assisted simulation training scenario command system according to claim 1, characterized in that: This involves performing the prediction step in the neural network module, specifically: Designing a network structure of a neural network model; wherein the network structure includes an input layer, a hidden layer, and an output layer; Implementing an incremental learning strategy and adversarial neural network training in a neural network model; wherein the incremental learning strategy is determined based on the task type; The adversarial neural network training includes: setting the goal and strategy of adversarial training, constructing a generator and a discriminator; improving the authenticity of the generated synthetic data and the distinguishing ability of the discriminator by alternately training the generator and the discriminator, and enhancing the generalization ability of the neural network model by using the generated synthetic data; Perform multiple iterations of training on the neural network model using a training data set, and monitor the performance of the neural network model; wherein the training data set includes generated synthetic data; Adjusting hyperparameters according to the performance of the neural network model; wherein the hyperparameters include learning rate and batch size; Use the validation set to evaluate the performance of the neural network model to ensure that the neural network model maintains accurate and reliable prediction capabilities when encountering unknown or new data; Comprehensively evaluate the neural network model using an independent test set to verify its accuracy and usability; The neural network model is optimized according to the evaluation results, and the optimized neural network model is deployed to the command system to achieve real-time scenario situation prediction and decision-making recommendations.

6. The command method based on the neural network assisted simulation training scenario command system according to claim 5, characterized in that: When the task type is a new task, the incremental learning strategy is regularization and / or replay mechanism; The regularization includes: identifying the difference between the new task and the old task, and ensuring that the neural network model does not forget the old knowledge while learning new knowledge by adding a regularization term in the loss function; The replay mechanism includes: extracting samples from the data set of the old task, fusing them with the new data and then training them in timeline, ensuring that the neural network model reviews old knowledge while learning new knowledge.

7. The command method based on the neural network assisted simulation training scenario command system according to claim 5, characterized in that: When the task type is an old task, determine the changed task parameters, and use a parameter isolation method as an incremental learning strategy. The parameter isolation method includes: Divide the neural network model parameters into multiple subsets, each subset is associated with a task type; During training, only the subset of parameters relevant to the task type is updated, keeping the parameters of other task types unchanged.

8. The command method based on the neural network assisted simulation training scenario command system according to claim 7, characterized in that: It also includes training the neural network model through semi-supervised learning and self-supervised learning when the amount of scene data collected is less than a preset threshold, specifically including: Organize the scene data into data sets, and divide the data sets into labeled data sets and unlabeled data sets; In the early stage of neural network model training, semi-supervised learning is used for preliminary training. When the preliminary training is completed, the training method is gradually transitioned from semi-supervised learning to self-supervised learning. Semi-supervised learning methods include: using a small amount of labeled data and a large amount of unlabeled data to train the neural network model, and gradually reducing the dependence on labeled data during the training process; Self-supervised learning methods include: using unlabeled data sets to train models, and generating new training sample data based on the results predicted by the neural network model and decision weights to continuously optimize the neural network model; After completing the training of the neural network model, the accuracy of the neural network model is verified through sampling verification and data testing, and the results are analyzed to adjust the decision weights to optimize the neural network model.

9. The command method based on the neural network assisted simulation training scenario command system according to claim 1, characterized in that: This includes executing the decision steps in the decision support module, including: According to the urgency of the current decision-making task, a method of constructing a decision tree is selected; among which, the methods of constructing a decision tree include a dictionary construction method, a database construction method, and a decision model construction method; Integrate the constructed decision tree into the decision support module; When a new decision task is received, the corresponding decision tree is called according to the input scenario conditions and decision weights, and the decision tree outputs the decision result; Perform statistical analysis on the output decision results; the statistical analysis methods include frequency distribution statistics and normal distribution statistics. Frequency distribution statistics are used to monitor and fine-tune the accuracy of the decision tree, and normal distribution statistics are used to detect whether the setting of decision weights is reasonable; Based on statistical analysis, the actual decision is compared with the decision results, and the user's feedback is used for supervised learning and real-time updating of the decision tree.

10. The command method based on the neural network assisted simulation training scenario command system according to claim 1, characterized in that: The interactive display step is performed in the user interaction module, specifically including: In response to manual user input of information, to supplement or amend the collected scene data; Use the map as a background to show the location, size and movement of the forces on both sides; Draw various symbols and charts on the map to intuitively display the overall scene situation; Support users to adjust decision weights according to actual needs; Display the decision suggestions output by the decision support module to the user; Supports the setting of different user roles, and assigns corresponding permissions and functions to user roles according to needs.

Citation Information

Cited By

  • Simulation acquisition method and system for multiple board cards

    CN120370827A

  • Action model learning system and method based on embedded AI

    CN120745740A

  • Simulation training guidance control method and system based on artificial intelligence

    CN120746238A