CN-121981569-A - Intelligent prediction and decision method for military material guarantee

CN121981569ACN 121981569 ACN121981569 ACN 121981569ACN-121981569-A

Abstract

The embodiment of the invention provides an intelligent prediction and decision-making method for a munition material guarantee, which comprises the steps of constructing a multi-source munition data fusion frame system, integrating multidimensional material demand characteristic information, defining and quantifying characteristic vectors of different combat scenes based on military rules and business knowledge, constructing a scene classification model, constructing a munition large model based on scene dynamic replenishment, fusing scene information and material demand characteristic information by adopting a layered attention mechanism, implementing a multi-stage training strategy, completing the training of the munition large model, constructing a dynamic replenishment decision-making module based on reinforcement learning based on the prediction result of the munition large model, and realizing real-time optimization of a replenishment scheme. The embodiment of the invention has the advantages of high prediction precision, strong adaptability, intelligent decision, high resource utilization rate and good expandability, and realizes intelligent prediction, dynamic planning and accurate replenishment of military demand materials.

Inventors

YAN YUSHENG
ZHANG CHUANMING
WANG YU
WEI ZIXIANG
LI LU
WANG FUMIN
WANG WENJING
LIU XIAOPENG

Assignees

中国船舶集团有限公司综合技术经济研究院
北京中科灵犀智能科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (10)

1. An intelligent prediction and decision method for army material guarantee is characterized by comprising the following steps: Step S1, constructing a multi-source military data fusion frame system, and integrating multi-dimensional material demand characteristic information; Step S2, defining and quantifying feature vectors of different combat scenes based on military rules and business knowledge, and establishing a scene classification model; step S3, constructing a large military requirement model based on scene dynamic replenishment, and fusing scene information and material demand characteristic information by adopting a layered attention mechanism; Step S4, implementing a multi-stage training strategy to finish training of the large military requirement model; and S5, constructing a dynamic replenishment decision module based on reinforcement learning based on a prediction result of the Legion needs large model, and realizing real-time optimization of a replenishment scheme.
2. The intelligent prediction and decision-making method for the supply of military supplies according to claim 1, further comprising, And S6, deploying an online learning mechanism to enable the model to have dynamic self-adaptive learning capability.
3. The intelligent prediction and decision-making method for military supplies guarantee according to claim 1, wherein the multi-dimensional supplies demand characteristic information comprises battlefield information data information, meteorological data information, equipment state data information, historical consumption data information, weapon force deployment data information and logistics resource data information.
4. The intelligent prediction and decision-making method for military supplies guarantee according to claim 1, wherein in step S2, scene characteristics are defined from dimensions including combat type, geographical environment, climate conditions, friend-foe situation and mission stage.
5. The intelligent prediction and decision-making method for military supplies guarantee according to claim 4, wherein in step S2, the encoding of scene feature vectors is realized by adopting a multi-granularity encoding strategy comprising discrete feature single-hot encoding, continuous feature segmentation quantization and text type feature semantic embedding.
6. The intelligent prediction and decision method for military supplies guarantee according to claim 5, wherein in step S2, the scene classification model establishes a scene association map based on feature vector cosine similarity to realize scene similarity calculation.
7. The intelligent prediction and decision-making method for municipality material security according to claim 1, wherein in step S3, said large municipality model based on scenerization dynamic replenishment comprises, The input layer is used for splicing multi-source data feature vectors, and the dimension is N multiplied by D, wherein N is the feature quantity, and D is the feature dimension; The scene attention layer calculates the weight of each feature in the current scene and realizes scene self-adaptive feature selection; The space-time coding layer introduces time convolution and space diagram convolution and captures the space-time dependency relationship of material consumption; A cross-modal fusion layer for fusing multi-modal information including numerical data, text report and image information through a multi-head attention mechanism; And the prediction output layer outputs various material demand predictions at a plurality of time points in the future and confidence assessment.
8. The intelligent prediction and decision-making method for municipality assurance according to claim 1, wherein in step S4, the multi-stage training strategy comprises, Training a basic language model of a large military requirement model on text data comprising desensitized public military documents, historical combat cases and equipment manuals, so that the large military requirement model learns the military field knowledge, and completing the pre-training of the large military requirement model; And (3) performing scene fine tuning, namely performing supervision fine tuning on the pre-trained army large model by using the corresponding data of the marked scene and material consumption, so that the model is suitable for a specific replenishment task.
9. The intelligent prediction and decision-making method for military supplies guarantee according to claim 1, wherein in step S5, the construction process of the reinforcement learning-based dynamic replenishment decision-making module is as follows, State space design, namely setting state parameters including current inventory, on-road materials, predicted requirements, transportation capacity and battlefield environment; the design of action space, namely, setting decision variables including replenishment time, replenishment category, replenishment quantity, transportation path and priority allocation; Designing a plurality of optimization targets including balance guarantee timeliness, resource utilization rate and risk control, and correspondingly setting a multi-target rewarding function; The training algorithm is selected by adopting a near-end strategy optimization algorithm to train the decision-making agent in the simulation environment.
10. The intelligent prediction and decision-making method for the supply of military supplies according to claim 2, wherein in step S6, model parameters are continuously optimized through actual combat feedback to form a closed-loop optimization system.

Description

Intelligent prediction and decision method for military material guarantee Technical Field The invention relates to the technical field of army material guarantee, in particular to an intelligent prediction and decision method for army material guarantee. Background In modern army logistics support systems, accurate supply of army supplies is a key factor in maintaining the combat power of the army. Traditional supply modes of military supplies mainly depend on historical experience, fixed period and static planning, and are difficult to adapt to the rapid change and highly uncertain battlefield environment of modern warfare. The prior art mainly has the following problems: 1. Static planning limitation that a traditional replenishment model cannot respond to battlefield situation changes in real time based on a fixed period and a preset scheme; 2. The data island problem is that related data of army needs to be ensured to be dispersed in different systems, and effective multi-source data fusion and association analysis are lacked; 3. The prediction precision is insufficient, the existing prediction model is inaccurate in grasping the material consumption law in the complex battlefield environment, and the emergency situation is difficult to predict; 4. The adaptability is poor, the trained model is difficult to quickly adapt to new combat scenes and task types, and a large number of retraining is needed; 5. the decision support is limited, the existing system provides data display functions, and intelligent decision suggestion and optimization schemes are lacked. In recent years, large model technology has made breakthrough progress in the fields of natural language processing, computer vision, etc., but application in the field of military logistics is still in the primary stage. How to deeply integrate large model technology with military need supply business and construct a dynamic supply system which can adapt to diversified battlefield scenes becomes a technical problem to be solved urgently. Disclosure of Invention In view of the above problems in the prior art, the embodiment of the invention provides an intelligent prediction and decision method for military material guarantee, which aims to solve the technical problems of static planning limitation, data island, insufficient prediction precision, poor adaptability, limited decision support and the like in the prior art. The embodiment of the invention provides an intelligent prediction and decision method for army material guarantee, which comprises the following steps: Step S1, constructing a multi-source military data fusion frame system, and integrating multi-dimensional material demand characteristic information; Step S2, defining and quantifying feature vectors of different combat scenes based on military rules and business knowledge, and establishing a scene classification model; step S3, constructing a large military requirement model based on scene dynamic replenishment, and fusing scene information and material demand characteristic information by adopting a layered attention mechanism; Step S4, implementing a multi-stage training strategy to finish training of the large military requirement model; and S5, constructing a dynamic replenishment decision module based on reinforcement learning based on a prediction result of the Legion needs large model, and realizing real-time optimization of a replenishment scheme. In one embodiment, the method further comprises the steps of, And S6, deploying an online learning mechanism to enable the model to have dynamic self-adaptive learning capability. In one embodiment, the multi-dimensional material demand characteristic information includes battlefield information data information, meteorological data information, equipment status data information, historical consumption data information, force deployment data information, and logistical resource data information. In one embodiment, in step S2, scene features are defined from dimensions including type of battle, geographic environment, climate conditions, friend or foe situation, and mission phase. In one embodiment, in step S2, the encoding of the scene feature vector is implemented using a multi-granularity encoding strategy including discrete feature single-hot encoding, continuous feature segment quantization, and text-type feature semantic embedding. In an embodiment, in step S2, the scene classification model establishes a scene association map based on the feature vector cosine similarity to implement scene similarity calculation. In one embodiment, in step S3, the scenerization-based dynamic replenishment-based large municipality model includes, The input layer is used for splicing multi-source data feature vectors, and the dimension is N multiplied by D, wherein N is the feature quantity, and D is the feature dimension; The scene attention layer calculates the weight of each feature in the current scene and realizes scene self-adaptive fea