A situational simulation method and device based on a large language model

Abstract

This invention discloses a situational simulation method and apparatus based on a large language model. The method includes: constructing a situational simulation information database; the database includes entity names, entity attributes, entity locations, and inter-entity communication relationships; training a large language model using the database to obtain an optimized large language model; acquiring user situational simulation requirement information; processing the user's situational simulation requirement information using the optimized large language model to obtain a situational simulation result; and rendering the situational simulation result to obtain a situational display result. This invention has significant theoretical and practical value for the presentation and interaction design of situational information in situational simulation.

Description

Technical Field

[0001] This invention relates to the field of situation simulation technology, and in particular to a situation simulation method and apparatus based on a large language model. Background Technology

[0002] In an information-driven environment, with increased dimensions and faster pace, a clear and accurate grasp of the situation and timely feedback of simulated situational information are key factors for achieving victory. Situational simulation systems are important platforms for commanders to conduct drills. While advancements in geographic information systems and graphics engine technology have led to some progress in the technical implementation of 2D and 3D situational simulation systems, a reasonable design system for the presentation of situational information and interactive design has yet to be established. Therefore, researching the display technology of situational simulation systems from the perspective of human-computer interaction is of great significance. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a situation simulation method and apparatus based on a large language model. This involves constructing a situation simulation information database; processing the database using a large language model to obtain situation simulation results; and rendering the simulation results to obtain the situation display results. This invention enables a reasonable presentation and interactive design of situation information, and has significant theoretical and practical value for the presentation and interactive design of situation information in situation simulation.

[0004] To address the aforementioned technical problems, the first aspect of this invention discloses a situational simulation method based on a large language model, the method comprising:

[0005] S1, Construct a situation simulation information database; the simulation information database includes entity names, entity attributes, entity locations, and communication relationships between entities;

[0006] S2, using the situational simulation information database, train the large language model to obtain an optimized large language model;

[0007] S3, obtain the user's situation simulation requirement information, and process the user's situation simulation requirement information using the optimized large language model to obtain the situation simulation result;

[0008] S4, render the situation simulation results to obtain the situation display results.

[0009] As an optional implementation, in the first aspect of the present invention, the step of training the large language model using the situational simulation information database to obtain an optimized large language model includes:

[0010] S21, Process the situation simulation information database to obtain situation simulation data information;

[0011] The situational simulation data information includes the entity's graph structure set G, the entity's screen coordinate information, and the entity's motion trajectory information;

[0012] S22, Using the situational simulation data, the large language model is trained to obtain an optimized large language model.

[0013] As an optional implementation, in the first aspect of the present invention, processing the situation simulation information database to obtain situation simulation data information includes:

[0014] S211, Process the situation simulation information database to obtain the graph structure set G of entities;

[0015] The graph structure set G of the entities is expressed as:

[0016] G = {V, E, Σ, L}

[0017] Where V is a non-empty set of vertices, and each vertex represents an entity. Let Σ be the set of edges, Σ be the set of vertex labels, and L be the label assignment function. There exists a vertex label L(v), where L(v)∈Σ;

[0018] S212, Process the graph structure set G to obtain entity screen coordinate information;

[0019] S213, process the communication relationship between the entities to obtain the entity motion trajectory information.

[0020] As an optional implementation, in the first aspect of the present invention, processing the communication relationship between the entities to obtain entity motion trajectory information includes:

[0021] S2131, Process the communication relationship between the entities to obtain the entity's environmental information, initial position information and target position information;

[0022] S2132, Process the environmental information, initial position information and target position information of the entity to obtain network training parameter information;

[0023] The network training parameter information includes the current action 'a' to be taken and the next state 's' obtained by executing action 'a'. t+1 Instant reward r and end signal;

[0024] S2133, Using the network training parameter information, the preset neural network model is trained to obtain an optimized neural network model;

[0025] S2134, using the optimized neural network model, the communication relationship between the entities to be processed is processed to obtain the entity motion trajectory information.

[0026] As an optional implementation, in the first aspect of the present invention, the step of training a large language model using the situational simulation data information to obtain an optimized large language model includes:

[0027] S221, The situation simulation data information is divided to obtain a labeled sample set and an unlabeled sample set;

[0028] S222, The first large language model is obtained by training the large language model using the labeled sample set.

[0029] S223, using the first large language model, process the unlabeled sample set to obtain a pseudo-labeled sample set;

[0030] S224, Integrate the labeled sample set and the pseudo-labeled sample set to obtain the training sample set;

[0031] S225, Using the training sample set, train the large language model to obtain an optimized large language model.

[0032] As an optional implementation, in the first aspect of the present invention, the step of processing the unlabeled sample set using the first large language model to obtain a pseudo-labeled sample set includes:

[0033] S2231, Extract the unlabeled sample set to obtain the first unlabeled sample set;

[0034] S2232, perform confidence processing on the first unlabeled sample set to obtain the second unlabeled sample set;

[0035] S2233, Perform a difference test on the second unlabeled sample set to obtain a pseudo-labeled sample set.

[0036] As an optional implementation, in the first aspect of the present invention, rendering the situation simulation result to obtain the situation display result includes:

[0037] S41, Visualize the situation simulation results to obtain a situation simulation diagram;

[0038] S42, Denoise the situation simulation image to obtain a denoised situation simulation image;

[0039] S43, Visual enhancement is performed on the denoised situation simulation image to obtain the situation display result.

[0040] A second aspect of this invention discloses a situational simulation device based on a large language model, the device comprising:

[0041] The information database construction module is used to construct a situation simulation information database; the simulation information database includes entity names, entity attributes, entity locations, and communication relationships between entities;

[0042] The model training module is used to train the large language model using the situation simulation information database to obtain an optimized large language model.

[0043] The situation simulation module is used to acquire the user's situation simulation requirement information, and process the user's situation simulation requirement information using the optimized large language model to obtain the situation simulation result.

[0044] The situation display module is used to render the situation simulation results to obtain the situation display results.

[0045] As an optional implementation, in the second aspect of the present invention, the step of training the large language model using the situational simulation information database to obtain an optimized large language model includes:

[0046] S21, Process the situation simulation information database to obtain situation simulation data information;

[0047] The situational simulation data information includes the entity's graph structure set G, the entity's screen coordinate information, and the entity's motion trajectory information;

[0048] S22, Using the situational simulation data, the large language model is trained to obtain an optimized large language model.

[0049] As an optional implementation, in the second aspect of the present invention, processing the situation simulation information database to obtain situation simulation data information includes:

[0050] S211, Process the situation simulation information database to obtain the graph structure set G of entities;

[0051] The graph structure set G of the entities is expressed as:

[0052] G = {V, E, Σ, L}

[0053] Where V is a non-empty set of vertices, and each vertex represents an entity. Let Σ be the set of edges, Σ be the set of vertex labels, and L be the label assignment function. There exists a vertex label L(v), where L(v)∈Σ;

[0054] S212, Process the graph structure set G to obtain entity screen coordinate information;

[0055] S213, process the communication relationship between the entities to obtain the entity motion trajectory information.

[0056] As an optional implementation, in the second aspect of the present invention, processing the inter-entity communication relationship to obtain entity motion trajectory information includes:

[0057] S2131, Process the communication relationship between the entities to obtain the entity's environmental information, initial position information and target position information;

[0058] S2132, Process the environmental information, initial position information and target position information of the entity to obtain network training parameter information;

[0059] The network training parameter information includes the current action 'a' to be taken and the next state 's' obtained by executing action 'a'. t+1 Instant reward r and end signal;

[0060] S2133, Using the network training parameter information, the preset neural network model is trained to obtain an optimized neural network model;

[0061] S2134, using the optimized neural network model, the communication relationship between the entities to be processed is processed to obtain the entity motion trajectory information.

[0062] As an optional implementation, in the second aspect of the present invention, the step of training the large language model using the situational simulation data information to obtain an optimized large language model includes:

[0063] S221, The situation simulation data information is divided to obtain a labeled sample set and an unlabeled sample set;

[0064] S222, The first large language model is obtained by training the large language model using the labeled sample set.

[0065] S223, using the first large language model, process the unlabeled sample set to obtain a pseudo-labeled sample set;

[0066] S224, Integrate the labeled sample set and the pseudo-labeled sample set to obtain the training sample set;

[0067] S225, Using the training sample set, train the large language model to obtain an optimized large language model.

[0068] As an optional implementation, in a second aspect of the present invention, the step of processing the unlabeled sample set using the first large language model to obtain a pseudo-labeled sample set includes:

[0069] S2231, Extract the unlabeled sample set to obtain the first unlabeled sample set;

[0070] S2232, perform confidence processing on the first unlabeled sample set to obtain the second unlabeled sample set;

[0071] S2233, Perform a difference test on the second unlabeled sample set to obtain a pseudo-labeled sample set.

[0072] As an optional implementation, in the second aspect of the present invention, rendering the situation simulation result to obtain the situation display result includes:

[0073] S41, Visualize the situation simulation results to obtain a situation simulation diagram;

[0074] S42, Denoise the situation simulation image to obtain a denoised situation simulation image;

[0075] S43, Visual enhancement is performed on the denoised situation simulation image to obtain the situation display result.

[0076] A third aspect of the present invention discloses another situational simulation device based on a large language model, the device comprising:

[0077] Memory containing executable program code;

[0078] A processor coupled to the memory;

[0079] The processor calls the executable program code stored in the memory to execute some or all of the steps in the situational simulation method based on a large language model disclosed in the first aspect of the present invention.

[0080] The fourth aspect of the present invention discloses a computer-storable medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the situational simulation method based on a large language model disclosed in the first aspect of the present invention.

[0081] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0082] This invention discloses a situation simulation method and apparatus based on a large language model. By constructing a situation simulation information database, processing the database using a large language model, and obtaining the situation simulation results, the method performs noise reduction and visual enhancement on these results to obtain the situation display results. This invention has significant theoretical and practical value for the presentation and interaction design of situation information in situation simulation. Attached Figure Description

[0083] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0084] Figure 1 This is a flowchart illustrating a situational simulation method based on a large language model disclosed in an embodiment of the present invention;

[0085] Figure 2 This is a schematic diagram of the structure of a situation simulation device based on a large language model disclosed in an embodiment of the present invention;

[0086] Figure 3 This is a schematic diagram of another situational simulation device based on a large language model disclosed in an embodiment of the present invention. Detailed Implementation

[0087] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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.

[0088] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0089] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0090] This invention discloses a situational simulation method and apparatus based on a large language model. The method includes: constructing a situational simulation information database; the database includes entity names, entity attributes, entity locations, and inter-entity communication relationships; training a large language model using the database to obtain an optimized large language model; acquiring user situational simulation requirement information; processing the user's situational simulation requirement information using the optimized large language model to obtain a situational simulation result; and rendering the situational simulation result to obtain a situational display result. This invention has significant theoretical and practical value for the presentation and interaction design of situational information in situational simulation. These will be described in detail below.

[0091] Example 1

[0092] Please see Figure 1 , Figure 1 This is a flowchart illustrating a situational simulation method based on a large language model disclosed in an embodiment of the present invention. Wherein, Figure 1 The situation simulation method based on a large language model described herein is applied in the field of situation simulation technology, and the embodiments of this invention are not limited thereto. Figure 1 As shown, this situational simulation method based on a large language model can include the following operations:

[0093] S1, Construct a situation simulation information database; the simulation information database includes entity names, entity attributes, entity locations, and communication relationships between entities;

[0094] S2, using the situational simulation information database, train the large language model to obtain an optimized large language model;

[0095] S3, obtain the user's situation simulation requirement information, and process the user's situation simulation requirement information using the optimized large language model to obtain the situation simulation result;

[0096] S4, render the situation simulation results to obtain the situation display results.

[0097] Optionally, the step of training the large language model using the situational simulation information database to obtain an optimized large language model includes:

[0098] S21, Process the situation simulation information database to obtain situation simulation data information;

[0099] The situational simulation data information includes the entity's graph structure set G, the entity's screen coordinate information, and the entity's motion trajectory information;

[0100] S22, Using the situational simulation data, the large language model is trained to obtain an optimized large language model.

[0101] Optionally, processing the situation simulation information database to obtain situation simulation data information includes:

[0102] S211, Process the situation simulation information database to obtain the graph structure set G of entities;

[0103] The graph structure set G of the entities is expressed as:

[0104] G = {V, E, Σ, L}

[0105] Where V is a non-empty set of vertices, and each vertex represents an entity. Let Σ be the set of edges, Σ be the set of vertex labels, and L be the label assignment function. There exists a vertex label L(v), where L(v)∈Σ;

[0106] S212, Process the graph structure set G to obtain entity screen coordinate information;

[0107] Obtain the geographic coordinates of the entity. The rules for defining geographic coordinates are as follows:

[0108] The X-axis (representing longitude) increases to the right, and the Y-axis (representing latitude) increases upwards;

[0109] The rules for defining screen coordinates are as follows:

[0110] The X-axis increases to the right, and the Y-axis increases downwards;

[0111] The only difference between geographic coordinates and screen coordinates is that the Y-axis increases in the opposite direction.

[0112] Given the screen's height (y) and width (h), and the geographic coordinate range (maxlon, minlon, maxlat, minlat), the longitude and latitude represented by each pixel can be calculated using the following formula:

[0113] scaleX = h / ((maxlon - minlon) * 3600) --- Longitude in seconds represented by each pixel on the X-axis

[0114] scaleY = y / ((maxlat - minlat) * 3600) --- latitude in seconds represented by each pixel on the Y-axis

[0115] The coordinates of any point (lon, lat) on the geographic coordinate region on the screen are:

[0116] screenX = lon * 3600 / scaleX --- Screen coordinates (X-axis coordinate)

[0117] screenY = lat * 3600 / scaleY --- Screen Y-axis coordinate

[0118] minx = minlon * 3600 / scaleX --- Set the leftmost edge of the region.

[0119] minY = minlat * 3600 / scaleY --- Sets the top of the region.

[0120] When the geographic area fills the entire screen, the formula for calculating any point within that area using the screenX and screenY parameters is as follows:

[0121] X=screenX-minX=(lon-minlon)*3600 / scaleX

[0122] Because the direction of latitude is opposite to the Y-axis of the screen.

[0123] screenMaxlat=(mxlat-minlat)*3600 / scaleY

[0124] screenlat=(lat-minlat)*3600 / scaleY

[0125] Y=screenMaxlat-screenlat=(maxlat-lat)scaleY

[0126] The formula for converting latitude and longitude to screen coordinates is:

[0127] X = (lon - minlon) * 3600 / scaleX

[0128] Y = (maxlat - lat) * 3600 / scaleY

[0129] The formula for converting screen coordinates to latitude and longitude coordinates is:

[0130] lon = X * scaleX / 3600 + minlon

[0131] lat = maxlat - y * scaleY / 3600

[0132] S213, process the communication relationship between the entities to obtain the entity motion trajectory information.

[0133] Optionally, processing the communication relationship between the entities to obtain entity motion trajectory information includes:

[0134] S2131, Process the communication relationship between the entities to obtain the entity's environmental information, initial position information and target position information;

[0135] S2132, Process the environmental information, initial position information and target position information of the entity to obtain network training parameter information;

[0136] The network training parameter information includes the current action 'a' to be taken and the next state 's' obtained by executing action 'a'. t+1 Instant reward r and end signal;

[0137] S2133, Using the network training parameter information, the preset neural network model is trained to obtain an optimized neural network model;

[0138] S2134, using the optimized neural network model, the communication relationship between the entities to be processed is processed to obtain the entity motion trajectory information.

[0139] The specific steps include:

[0140] 1. Environment Setup: Set up the map for inter-entity communication, with the goal of finding the optimal path to the target point. An entity's state is defined as a triple (x, y, α), where x and y are the entity's real-time coordinates on the simulation map, and α is the yaw angle between the entity and the target point. Let the entity's initial coordinates be (x0, y0), then the actual heading angle is initialized as θ = arctan(x0, y0). Let the current coordinates be (x, y), and the target coordinates be (p, q), then the formula for calculating the desired heading β is:

[0141] β=arctan(xp,yq)-arctan(x,y)

[0142] The actual heading angle θ of the entity changes with each action performed. Let θ be the direction angle corresponding to action a. θ Then the update formula for θ is: θ new =θ old +a θ

[0143] The yaw angle α is defined as: α = β - θ

[0144] The steering angle Δθ is defined as: Δθ = new -θ old

[0145] The value of the turning angle Δθ at time t is the angle of action taken by the entity at that time, which is α. θ .

[0146] 2. Set the reward function;

[0147] 1) Initialize the neural network parameters ω and θ, and initialize the state sequence buffer M and the training loop count counter C.

[0148] 2) Initialize the input sequence input and the initial state s0.

[0149] 3) Use the Actor value network to obtain the action 'a' to be taken, execute action 'a', and obtain environmental feedback: the next state 's'. t+1 Instant reward r, end signal Do, collision signal Co.

[0150] 4) Determine if Do is True. If yes, end the current training and jump to step 2).

[0151] 5) s t a, r, s t+1 The state-action sequence is stored in the state sequence cache M so that it can be sampled to train the neural network.

[0152] 6) Sample the current state sequence input_current and the next state sequence input_next from the state sequence buffer M, and input them into the Critic value network and the Critic target network respectively to obtain the current state value V(s). t ) and the next state value V(s) t+1 ).

[0153] 7) Calculate the first difference information and update the parameters ω of the Critic value network using it. The loss function for the Critic value network is the mean squared error function. The expression for the first difference information is as follows, where i is the sample index:

[0154]

[0155] The loss function of the Critic network is defined as the square of the error of the first difference information, that is:

[0156]

[0157] The update formula for the Critic network parameters is:

[0158]

[0159] The current state sequence (input) is fed into the Actor value network to obtain the probability distribution of the action output corresponding to the current state sequence. The parameters θ of the Actor value network are updated using the first difference information. The loss function expression of the Actor network is as follows:

[0160]

[0161] The formula for updating the Actor network parameters is:

[0162]

[0163] Every 30 steps the agent performs, it synchronizes the parameters of the Critic value network and the Critic target network.

[0164] Check if Co is True; if so, return to the previous state and jump to step 3).

[0165] Determine if C has reached its maximum value; otherwise, end the training or jump to step 2.

[0166] Optionally, the step of using the situational simulation data information to train the large language model to obtain an optimized large language model includes:

[0167] S221, The situation simulation data information is divided to obtain a labeled sample set and an unlabeled sample set;

[0168] The division ratio was set by the experiment. In this embodiment, the labeled sample set was 30% and the unlabeled sample set was 50%.

[0169] S222, The large language model is trained using the labeled sample set to obtain the first large language model;

[0170] Large language models are deep learning models trained on massive amounts of text data. They can generate natural language text or understand the meaning of language text. The formula used is the text2vec model, which primarily vectorizes the text. The text2vec model can represent the entire text as a sequence of vectors, thus preserving the semantic information of the text. The question is then converted into a vector, and vector similarity is compared.

[0171] S223, using the first large language model, process the unlabeled sample set to obtain a pseudo-labeled sample set;

[0172] S224, Integrate the labeled sample set and the pseudo-labeled sample set to obtain the training sample set;

[0173] Specifically, this involves merging the labeled sample set and the pseudo-labeled sample set together;

[0174] S225, Using the training sample set, train the large language model to obtain an optimized large language model.

[0175] Optionally, the step of processing the unlabeled sample set using the first large language model to obtain a pseudo-labeled sample set includes:

[0176] S2231, Extract the unlabeled sample set to obtain the first unlabeled sample set;

[0177] A 10% sample is extracted from the unlabeled sample set to obtain the first unlabeled sample set;

[0178] S2232, perform confidence processing on the first unlabeled sample set to obtain the second unlabeled sample set;

[0179] Specifically, the first language model is used to select high-confidence samples S from the first unlabeled sample set and assign them pseudo-labels to obtain the second unlabeled sample set.

[0180] Among them, high-confidence samples are unlabeled samples with a comparison membership degree greater than the membership threshold;

[0181] The membership degree is compared as follows:

[0182] N ij =max(P(C) j |x i ;φ)) / max(P(C j′ |x i ;φ))

[0183] Where: N ij For sample x i With sample x j The comparative membership degree between them, j′∈{1,2,…,c|j′≠j}, i∈{1,2,…,m}, j∈{1,2,…,c}, P(C j |x i ;φ) represents the sample x i Belongs to category C j The probability, where c is the number of categories, P(C j′ |x i ;φ) is the sample x i Belongs to category C j′ The probability, m is the number of unlabeled samples, and φ is the membership function;

[0184] S2233, Perform a difference test on the second unlabeled sample set to obtain a pseudo-labeled sample set.

[0185] The difference test involves removing sample points whose local outlier exceeds a threshold. The local outlier is defined as:

[0186] τ(x i )=1 / lrd k (x i )

[0187] Where τ(x) i ) is the sample x i Local outlier

[0188]

[0189] N k (x i (excluding sample x) i A set of.

[0190] Optionally, rendering the situation simulation results to obtain the situation display results includes:

[0191] S41, Visualize the situation simulation results to obtain a situation simulation diagram;

[0192] S42, Denoise the situation simulation image to obtain a denoised situation simulation image;

[0193] The noise reduction method is as follows:

[0194] The situation simulation diagram is processed to obtain real and imaginary components;

[0195] Perform wavelet transform on the real components to obtain the first wavelet parameter information;

[0196] Perform wavelet transform on the imaginary component to obtain the second wavelet parameter information;

[0197] The first wavelet parameter information is filtered to obtain the third wavelet parameter information.

[0198] The second wavelet parameter information is filtered to obtain the fourth wavelet parameter information;

[0199] The filtering method is as follows:

[0200] Set the initial threshold δ th The target interval is [α,β], where 0≤α<β≤max{S(m,n)}, and S(m,n) is the wavelet parameter information;

[0201] The target interval [α,β] is divided into L equal parts;

[0202] The initial threshold δ th Set to α, according to The threshold is obtained by iteratively increasing the step size.

[0203] Using threshold Calculate the wavelet parameter information after filtering in each iteration.

[0204]

[0205] Where S(m,n) represents the original wavelet parameter information, This represents the wavelet parameter information after filtering, where a and b represent the time and frequency parameters in the threshold.

[0206] Calculate the difference between the wavelet parameter information obtained in each iteration and the previous iteration. This represents the number of iterations.

[0207] Difference Curve fitting is performed to obtain curve C. When curve C reaches an inflection point, the corresponding threshold is the optimal threshold.

[0208] Using the aforementioned optimal threshold, the wavelet parameter information S(m,n) is subjected to optimal threshold filtering to obtain the filtered wavelet parameter information;

[0209] The third wavelet parameter information and the fourth wavelet parameter information are reconstructed to obtain a reconstructed time-frequency image;

[0210] Perform an inverse time-frequency transform on the reconstructed time-frequency image to obtain a denoised situation simulation image;

[0211] S43, Visual enhancement is performed on the denoised situation simulation image to obtain the situation display result.

[0212] The methods for visual enhancement are:

[0213] S431, Process the denoised situation simulation image to obtain a reference image;

[0214] Specifically, the denoised situational simulation image is downsampled to obtain a reference image;

[0215] S432, Standardize the denoised situation simulation image and the reference image to obtain a standardized denoised situation simulation image and a standardized reference image;

[0216] S433, the denoised situation simulation image is segmented to obtain N visible light image blocks;

[0217] S434, The standardized reference image is segmented to obtain N reference image blocks;

[0218] The N visible light image blocks and the N reference image blocks are of equal size;

[0219] S435, perform similarity calculation on the N visible light image blocks and the N reference image blocks to obtain similarity values;

[0220] S436, Based on the similarity value, perform migration processing on the N visible light image blocks and the N reference image blocks to obtain a visually enhanced image, which is the situation display result.

[0221] As can be seen, this invention discloses a situation simulation method and apparatus based on a large language model. By constructing a situation simulation information database, processing the database using a large language model, and obtaining the situation simulation results, the method further performs noise reduction and visual enhancement on these results to obtain the situation display results. This invention has significant theoretical and practical value for the presentation and interaction design of situation information in situation simulation.

[0222] Example 2

[0223] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a situation simulation device based on a large language model, as disclosed in an embodiment of the present invention. Figure 2 The situation simulation device based on a large language model described herein is applied in the field of situation simulation technology, and the embodiments of this invention are not limited thereto. Figure 2 As shown, the situational simulation device based on a large language model can include the following operations:

[0224] S301, Information Database Construction Module, used to construct a situation simulation information database; the simulation information database includes entity names, entity attributes, entity locations, and communication relationships between entities;

[0225] S302, Model Training Module, used to train the large language model using the situation simulation information database to obtain an optimized large language model;

[0226] S303, Situation simulation module, used to acquire user situation simulation requirement information, process the user situation simulation requirement information using the optimized large language model, and obtain situation simulation results;

[0227] S304, Situation display module, is used to render the situation simulation results and obtain the situation display results.

[0228] Example 3

[0229] Please see Figure 3 , Figure 3 This is a schematic diagram of another situational simulation device based on a large language model disclosed in an embodiment of the present invention. Figure 3 The situation simulation device based on a large language model described herein is applied in the field of situation simulation technology, and the embodiments of this invention are not limited thereto. Figure 3 As shown, the situational simulation device based on a large language model can include the following operations:

[0230] Memory 401 storing executable program code;

[0231] Processor 402 coupled to memory 401;

[0232] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the situational simulation method based on a large language model described in Embodiment 1.

[0233] Example 4

[0234] This invention discloses a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program enables a computer to perform the steps in the situational simulation method based on a large language model described in Embodiment 1.

[0235] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0236] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0237] Finally, it should be noted that the situational simulation method and apparatus based on a large language model disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A situational simulation method based on a large language model, characterized in that, The method includes: S1, Construct a situation simulation information database; the simulation information database includes entity names, entity attributes, entity locations, and communication relationships between entities; S2, using the aforementioned situational simulation information database, train the large language model to obtain an optimized large language model, including: S21, Process the situation simulation information database to obtain situation simulation data information; The situational simulation data information includes a graph structure set of entities. G The entity's screen coordinates and motion trajectory information include: S211, Process the situation simulation information database to obtain a graph structure set of entities. G ; The graph structure set G of the entities is expressed as: in, V It is a non-empty set of vertices, where each vertex represents an entity. It is a set of edges. It is a collection of vertex labels. L It is a tag assignment function, for All L ( v ) represents a vertex v Tags, among which, ; S212, for the graph structure set G The process is performed to obtain the physical screen coordinate information; S213, Process the communication relationship between the entities to obtain entity motion trajectory information, including: S2131, Process the communication relationship between the entities to obtain the entity's environmental information, initial position information and target position information; S2132, Process the environmental information, initial position information and target position information of the entity to obtain network training parameter information; The network training parameter information includes the action to be taken. Execute actions The next state obtained Instant rewards r and end signal; S2133, Using the network training parameter information, the preset neural network model is trained to obtain an optimized neural network model; S2134, Using the optimized neural network model, the communication relationship between the entities to be processed is processed to obtain the entity motion trajectory information; The specific steps include: Environment setup: Configure the map for inter-entity communication; the state of an entity is defined as a triple. ,in and This refers to the real-time coordinates of the entity on the simulation map. Let be the yaw angle between the entity and the target point; let the initial coordinates of the entity be... The actual heading angle is then initialized as follows: Let the current coordinates be... The target coordinates are Then the expected course The calculation formula is: Actual heading angle of the entity The action changes with each execution. The corresponding direction angle is ,but The update formula is: Yaw angle Defined as: Steering angle Defined as: The turning angle at time t The value is the angle of action taken by the entity at this moment. ; S22, Using the situational simulation data information, the large language model is trained to obtain an optimized large language model; S3, obtain the user's situation simulation requirement information, and process the user's situation simulation requirement information using the optimized large language model to obtain the situation simulation result; S4, render the situation simulation results to obtain the situation display results.

2. The situational simulation method based on a large language model according to claim 1, characterized in that, The step of training the large language model using the situational simulation data to obtain an optimized large language model includes: S221, The situation simulation data information is divided to obtain a labeled sample set and an unlabeled sample set; S222, The first large language model is obtained by training the large language model using the labeled sample set. S223, using the first large language model, process the unlabeled sample set to obtain a pseudo-labeled sample set; S224, Integrate the labeled sample set and the pseudo-labeled sample set to obtain the training sample set; S225, Using the training sample set, train the large language model to obtain an optimized large language model.

3. The situational simulation method based on a large language model according to claim 2, characterized in that, The process of using the first large language model to process the unlabeled sample set to obtain a pseudo-labeled sample set includes: S2231, Extract the unlabeled sample set to obtain the first unlabeled sample set; S2232, perform confidence processing on the first unlabeled sample set to obtain the second unlabeled sample set; S2233, Perform a difference test on the second unlabeled sample set to obtain a pseudo-labeled sample set.

4. The situational simulation method based on a large language model according to claim 1, characterized in that, The process of rendering the situation simulation results to obtain the situation display results includes: S41, Visualize the situation simulation results to obtain a situation simulation diagram; S42, Denoise the situation simulation image to obtain a denoised situation simulation image; S43, Visual enhancement is performed on the denoised situation simulation image to obtain the situation display result.

5. A situational simulation device based on a large language model, characterized in that, The device includes: The information database construction module is used to construct a situation simulation information database; the simulation information database includes entity names, entity attributes, entity locations, and communication relationships between entities; The model training module, used to train the large language model using the situation simulation information database to obtain an optimized large language model, includes: S21, Process the situation simulation information database to obtain situation simulation data information; The situational simulation data information includes a graph structure set of entities. G The entity's screen coordinates and motion trajectory information include: S211, Process the situation simulation information database to obtain a graph structure set of entities. G ; The graph structure set G of the entities is expressed as: in, V It is a non-empty set of vertices, where each vertex represents an entity. It is a set of edges. It is a collection of vertex labels. L It is a tag assignment function, for All L ( v ) represents a vertex v Tags, among which, ; S212, for the graph structure set G The process is performed to obtain the physical screen coordinate information; S213, Process the communication relationship between the entities to obtain entity motion trajectory information, including: S2131, Process the communication relationship between the entities to obtain the entity's environmental information, initial position information and target position information; S2132, Process the environmental information, initial position information and target position information of the entity to obtain network training parameter information; The network training parameter information includes the action to be taken. Execute actions The next state obtained Instant rewards r and end signal; S2133, Using the network training parameter information, the preset neural network model is trained to obtain an optimized neural network model; S2134, Using the optimized neural network model, the communication relationship between the entities to be processed is processed to obtain the entity motion trajectory information; The specific steps include: Environment setup: Configure the map for inter-entity communication; the state of an entity is defined as a triple. ,in and This refers to the real-time coordinates of the entity on the simulation map. Let be the yaw angle between the entity and the target point; let the initial coordinates of the entity be... The actual heading angle is then initialized as follows: Let the current coordinates be... The target coordinates are Then the expected course The calculation formula is: Actual heading angle of the entity The action changes with each execution. The corresponding direction angle is ,but The update formula is: Yaw angle Defined as: Steering angle Defined as: The turning angle at time t The value is the angle of action taken by the entity at this moment. ; S22, Using the situational simulation data information, the large language model is trained to obtain an optimized large language model; The situation simulation module is used to acquire the user's situation simulation requirement information, and process the user's situation simulation requirement information using the optimized large language model to obtain the situation simulation result. The situation display module is used to render the situation simulation results to obtain the situation display results.

6. A situational simulation device based on a large language model, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the situation simulation method based on a large language model as described in claims 1 to 4.

7. A computer-storable medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked by the processor, are used to execute the situation simulation method based on a large language model as described in claims 1-4.

Images