A militarized target behavior analysis intelligent agent testing method and equipment

Abstract

The present invention belongs to the field of artificial intelligence analysis technology, and specifically relates to a method and device for testing an intelligent agent for analyzing the behavior of a militarized target. The method comprises the following steps: Step 1: obtaining multi-source data of a militarized target and performing data preprocessing on each category of data in the multi-source data; Step 2: extracting tensor features of the input tensor; defining a multi-layered intent hierarchy based on the tensor features; Step 3: defining the likelihood function of each intent layer; using a recursive neural network to model the prior probability; using the Bayesian rule to calculate the posterior probability; when the number of iterative executions reaches a set number, executing Step 4: Step 4: using an approximate posterior distribution to perform intent inference and calculating the uncertainty of the intent inference. The present invention significantly improves the accuracy, real-time performance, interpretability, and reliability of the behavior analysis of militarized targets.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence analysis technology, and specifically relates to a method and equipment for testing a militarized target behavior analysis intelligent agent. Background Art

[0002] In modern military operations, the application of intelligent and automated technologies is becoming increasingly widespread, particularly in the field of militarized target behavior analysis. By accurately predicting and analyzing enemy target behavior, military commanders can better make strategic decisions, optimize resource allocation, and improve operational effectiveness. However, existing militarized target behavior analysis technologies still face many problems and challenges in practical application. Currently, militarized target behavior analysis technologies primarily rely on data-driven approaches, including machine learning and deep learning. These technologies train predictive models on large amounts of historical data to identify and infer enemy target behavior intentions. For example, supervised learning methods allow researchers to train classifiers to identify specific target behavior patterns; unsupervised learning methods can uncover hidden behavioral patterns and associations; and deep learning methods, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can process and analyze complex spatiotemporal data for more refined target behavior predictions.

[0003] Militarized target behavior analysis relies on large amounts of high-quality training data. However, in practice, obtaining and processing this data presents numerous challenges. Military intelligence data often contains noise, incompleteness, and inconsistencies, which hinders model training and prediction accuracy. Furthermore, data formats and types vary significantly across different data sources, making effective integration of these data sources a significant challenge. Existing machine learning and deep learning models are prone to overfitting during training. This means that while the models perform well on the training data, they generally generalize poorly in real-world applications. This occurs because the models memorize specific data patterns during training and fail to effectively learn universally applicable features. This problem is particularly acute in militarized target behavior analysis, as actual combat environments are complex and ever-changing, and training data often lacks the capacity to cover all possible scenarios. Militarized target behavior analysis requires real-time processing and analysis of large amounts of data to produce accurate predictions. However, existing deep learning models often suffer from high computational complexity and slow processing speeds, making them difficult to meet real-time requirements. This can lead to delays and information lags in actual combat, impacting the timeliness and accuracy of decision-making. While deep learning models, especially neural network models, excel in complex pattern recognition and prediction, their internal structures are complex and lack interpretability. This makes it difficult for decision makers to understand the model's predictions and inference process, which in turn reduces their trust in and reliance on the model. Model interpretability is crucial in military decision-making, as decision makers must base their judgments on clear and reliable information. Existing technologies often neglect assessing the uncertainty of predictions when predicting target behavior. This prevents decision makers from effectively assessing the reliability and risk of predictions, potentially leading to erroneous decisions. In actual combat, uncertainty assessment can provide important information about the confidence level of predictions, helping decision makers make more prudent and rational decisions in uncertain environments. Summary of the Invention

[0004] In view of this, the main purpose of the present invention is to provide a method and equipment for testing a militarized target behavior analysis intelligent agent, which significantly improves the accuracy, real-time performance, interpretability and reliability of militarized target behavior analysis.

[0005] The technical solution adopted in the present invention is as follows:

[0006] A method for testing a militarized target behavior analysis agent, the method comprising:

[0007] Step 1: Obtain multi-source data of militarized targets, pre-process the data of each category in the multi-source data, align the timestamps and spatially register them, and then fuse them into an input tensor;

[0008] Step 2: Extract tensor features from the input tensor; based on the tensor features, define a multi-layer intent hierarchy, where each intent layer contains multiple possible intents, and use conditional probabilities to represent the dependencies between intent layers.

[0009] Step 3: Define the likelihood function of each intent layer; use a recursive neural network to model the prior probability; use Bayes' rule to calculate the posterior probability; iteratively perform the following process to complete the training: use variational inference to define an approximate posterior distribution, and then minimize the KL divergence; use importance sampling to approximate the expectation, and use stochastic gradient descent to optimize the parameters of variational inference; when the number of iterations reaches the set number, execute step 4:

[0010] Step 4: Use the approximate posterior distribution to perform intent inference and calculate the uncertainty of intent inference.

[0011] Furthermore, the multi-layered intention hierarchy in step 2 includes at least three intention layers, namely: a strategic intention layer, a tactical intention layer, and a specific action intention layer; the strategic intention layer includes at least the following intentions: attack, defense, retreat, reorganization, and deception; the tactical intention layer includes at least the following intentions: encirclement, breakthrough, reconnaissance, blocking, detour, and feint attack; the specific action intention layer includes at least the following intentions: deploying weapon systems, establishing fortifications, conducting air strikes, conducting electronic interference, launching ground attacks, conducting special operations, and establishing supply lines.

[0012] Furthermore, the multi-source data includes at least the following categories of data: position data, speed data, acceleration data, direction data, environmental data, altitude data, longitude data and latitude data; the process of performing data preprocessing on each category of the multi-source data in step 1 includes: performing data noise reduction and data normalization on the data of each category; in step 1, after performing timestamp alignment and spatial registration on the preprocessed data of each category, resampling and interpolation processing on the data are performed to fuse them into an input tensor; in step 2, using a deep neural network to extract tensor features of the input tensor .

[0013] Furthermore, in step 2, define Layer intention structure, Layer Intent Layer possible intentions; initialize the intent vector for each layer:

[0014] ;

[0015] in, , is the index number; Indicates the The first layer of the intention layer intention; each is a discrete variable and can be possible values;

[0016] Tensor features The nonlinear transformation is performed by the following formula to obtain the feature , to accommodate different levels of intent reasoning:

[0017] ;

[0018] in, represents element-wise multiplication, and They are the preset feature weight matrix and feature bias matrix respectively;

[0019] Use conditional probability to represent the dependencies between intent layers:

[0020] ;

[0021] in, Expressing the highest level of intention, Indicates the lowest level of intention; For a given feature The joint probability distribution of the entire intent hierarchy when ; For a given When considering layer intentions and features, the conditional probability of the lowest layer intention; For a given When layering intent and features, Conditional probability of layer intention; The conditional probability of the top-level intent given the features.

[0022] Furthermore, in step 3, the following formula is used to define Likelihood function of the layer intention layer , which contains a multi-layer perceptron with a main branch and a multi-layer sensor with a parallel branch. The main branch uses the tanh activation function, and the parallel branch uses the LeakyReLU activation function, which is expressed by the following formula:

[0023] ;

[0024] in, From the 1st to the The intention sequence of layers; is the embedding representation of the low-level intent, which is a function that maps a discrete sequence of intents to a continuous vector space. Let the dimension after embedding be ;set up The latitude is , is the concatenation of feature and low-level intent embedding, with dimension ; For the The weight matrix of the layer intention layer has the dimension ,in It is Dimensions of the layer intention layer; For the The bias vector of the layer intent layer has a dimension of ; For the The weight matrix of the layer intention layer has the dimension ,in is the dimension of the output; For the The bias vector of the layer intent layer has the dimension ; For the parallel branch The weight matrix of the layer intention layer has the dimension ; For the parallel branch The bias vector of the layer intent layer has a dimension of ; For the parallel branch The weight matrix of the layer intention layer has the dimension ; For the parallel branch The bias vector of the layer intent layer has a dimension of ; is the activation function of the rectified linear unit with leakage, is a function variable; is the Softmax function, which is used to convert the output into a probability distribution; is the Sigmoid activation function, which is used to compress the output to the (0,1) interval.

[0025] Furthermore, a recurrent neural network is used to model the first Prior probability of the intention layer , each intent layer corresponds to a hidden layer of the recurrent neural network:

[0026] ;

[0027] ;

[0028] in, represents a gated recurrent unit; The recurrent neural network corresponds to Hidden layer of the intent layer; The recurrent neural network corresponds to Hidden layer of the intent layer; is the weight matrix of the recurrent neural network; is the bias matrix of the recurrent neural network.

[0029] Furthermore, the Bayesian rule is used to calculate the Posterior probability of the intention layer :

[0030] .

[0031] in, is the subscript index; using variational inference, define an approximate posterior distribution using the following formula:

[0032] ;

[0033] in, is the parameterized approximate posterior distribution; are the parameters of variational inference; is the weight matrix for variational inference; is the bias vector for variational inference.

[0034] Furthermore, the KL divergence is minimized by the following formula:

[0035] ;

[0036] in, is the loss function; represents the KL divergence; importance sampling is used to approximate the expectation using the following formula:

[0037] +const;

[0038] in, is the number of sampling times; is the kth intention layer obtained by the sth sampling: const is a constant term; the parameters of variational inference are optimized using stochastic gradient descent using the following formula:

[0039] ;

[0040] in, For the Parameters of variational inference after iterative optimization; The parameters of the variational inference after optimization at the t+1th iteration; is the learning rate; is the gradient of the loss function with respect to the parameter θ.

[0041] Furthermore, in step 4, the approximate posterior distribution is used to perform intent inference using the following formula:

[0042] ;

[0043] in, is the result of intention reasoning; then use the following formula to calculate the uncertainty of intention reasoning:

[0044] ;

[0045] in, Reasoning about uncertainty in intentions.

[0046] A militarized target behavior analysis intelligent agent testing device includes: a data acquisition part, which is used to acquire multi-source data of militarized targets, perform data preprocessing on each category of data in the multi-source data, perform timestamp alignment and spatial registration, and then fuse them into an input tensor; an intention model construction part, which is used to extract tensor features of the input tensor; based on the tensor features, define a multi-layer intention hierarchy, each intention layer contains multiple possible intentions, and use conditional probability to represent the dependency relationship between the intention layers; an intention reasoning model construction part, which is used to define the likelihood function of each intention layer; use a recursive neural network to model the prior probability; use the Bayesian rule to calculate the posterior probability; iteratively execute the following process to complete training: use variational inference to define an approximate posterior distribution, and then minimize the KL divergence; use importance sampling to approximate expectations, and use stochastic gradient descent to optimize the parameters of variational inference; when the number of iterative executions reaches the set number, execute step 4: the intention reasoning part, which is used to perform intention reasoning using the approximate posterior distribution and calculate the uncertainty of intention reasoning

[0047] The above technical solution achieves the following beneficial effects: By combining a multi-layer intent hierarchy with a recurrent neural network (RNN), the present invention effectively improves the model's generalization capabilities. Each intent layer uses a GRU (Gated Recurrent Unit) to model prior probabilities, enabling the model to handle long-term dependencies and capture complex patterns in target behavior. Variational inference defines an approximate posterior distribution to approximate the true posterior distribution. By minimizing the KL divergence, the variational inference parameters are optimized to ensure the approximate posterior distribution is as close as possible to the true posterior distribution. This approach effectively avoids model overfitting, improves generalization across diverse environments, and makes the model more robust in practical applications. The present invention also addresses significant work in enhancing model interpretability. Through a multi-layer intent hierarchy, target behavior is decomposed into strategic intent, tactical intent, and specific action intent. Each intent and computational process has clear physical meaning and explanation. This design makes the model's reasoning process more transparent, making it easier for decision makers to understand and trust. For each layer of intent, uncertainty is assessed by calculating entropy, allowing decision makers to more clearly understand the confidence and reliability of model predictions, thereby making more informed decisions in actual operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of a militarized target behavior analysis agent testing method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] All features disclosed in this specification, or all steps in the disclosed methods or processes, except mutually exclusive features and / or steps, can be combined in any manner.

[0050] Any feature disclosed in this specification (including any appended claims and abstract), unless otherwise stated, may be replaced by other equivalent or similar features. In other words, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

[0051] Example 1: Reference Figure 1 , a militarized target behavior analysis agent testing method, the method comprising:

[0052] Step 1: Obtain multi-source data of militarized targets, pre-process the data of each category in the multi-source data, align the timestamps and spatially register them, and then fuse them into an input tensor;

[0053] Multi-source data typically includes data from various sensors, such as imagery, radar data, and communication logs. This data provides rich information about the target across different temporal and spatial dimensions. However, due to the diversity and complexity of these data sources, directly analyzing the raw data often leads to accuracy and reliability issues. Therefore, data preprocessing is a critical step. First, each data type is normalized to ensure that different data types have the same dimensions and scale. This process typically involves operations such as noise removal, missing value filling, and data smoothing to improve data quality and consistency. After normalization, timestamp alignment is also required. Multi-source data is often collected asynchronously, so temporal alignment is necessary to ensure that analysis is performed at the same point in time. This step is achieved through methods such as interpolation or resampling, mapping data from different time points onto a unified timeline. Next, spatial registration is performed, which involves spatially aligning data collected by different sensors. For example, imagery and radar data may be collected in different spatial coordinate systems. Registration techniques can transform them into a common coordinate system, facilitating fusion and analysis. Spatial registration typically uses geometric transformations and image registration algorithms, such as feature-based or density-based registration methods. This process aims to ensure spatial consistency between different data sources and eliminate errors caused by varying sensor positions and viewpoints. After timestamp alignment and spatial registration, all preprocessed data is fused into a multidimensional input tensor. The input tensor is a high-dimensional data structure that contains information across multiple dimensions, including time, space, and data type. By fusing multi-source data into a single input tensor, we can leverage the spatiotemporal correlations between the various data sources to provide a complete and detailed description of target behavior. For example, image data can provide visual information about the target, radar data can provide distance and velocity information, and communication logs can provide information about the target's interactions and behaviors. The fusion of this information in the form of an input tensor makes subsequent feature extraction and behavior analysis more accurate and reliable. The construction of the input tensor is not simply a simple data overlay; it is based on a deep understanding of the data's characteristics and correlations. This high-dimensional tensor can capture the dynamic changes in target behavior across time and space, as well as the complex relationships between different data sources. The input tensor provides a solid foundation for tensor feature extraction and the definition of a multi-layered intent hierarchy in subsequent steps. Through this data preprocessing and fusion approach, the present invention is able to better process and analyze complex militarized target behavior data, improving the accuracy and reliability of agent testing. Overall, step 1 plays a crucial role in this invention. Through comprehensive preprocessing and fusion techniques, a complete and accurate input data foundation is constructed, making subsequent behavior analysis and intent inference more efficient and accurate.

[0054] Step 2: Extract tensor features from the input tensor; based on the tensor features, define a multi-layer intent hierarchy, where each intent layer contains multiple possible intents, and use conditional probabilities to represent the dependencies between intent layers.

[0055] First, deep learning methods are used to extract features from the input tensor. In this step, the input tensor, as a high-dimensional data structure, contains rich information from time, space, and multiple sources. Deep neural networks, particularly convolutional neural networks (CNNs), can effectively extract important features from the tensor. Through their hierarchical structure and convolution operations, CNNs can automatically learn and capture spatial and temporal patterns in the data, generating low-dimensional but rich feature vectors. These feature vectors serve as the basis for subsequent intent analysis. Next, a multi-layer intent hierarchy is defined based on the extracted tensor features. This multi-layer intent hierarchy models the target's behavioral intent hierarchically, with each layer representing a different level of intent, from low-level, specific behaviors to high-level, strategic intent. This hierarchical structure better captures the complexity and diversity of target behavior. For example, in a militarized scenario, low-level intent may involve basic target actions, such as movement or attack, while high-level intent may involve more complex tactical or strategic objectives. This hierarchical intent modeling enables a more comprehensive understanding and prediction of target behavior. To model the dependencies between intents at each level in the multi-layer intent hierarchy, conditional probability is used to represent these relationships. Conditional probability describes the probability distribution of an intent layer given the intent of the upper layer. This allows complex hierarchical relationships within intent to be clearly expressed using a mathematical model. The probability distribution of each intent layer depends not only on the tensor features of the current layer but also on the intent state of the previous layer, forming a recursive dependency relationship. This approach effectively captures the dynamic changes and hierarchical dependencies of target behavior, making the model's inference of behavioral intent more accurate. In a multi-layered intent hierarchy, each intent layer can be considered a latent variable. By defining a likelihood function for each intent layer, the distribution of observed data under different intents can be described. The likelihood function is based on the relationship between tensor features and intent. By maximizing the likelihood function, the most likely interpretation of the intent can be found. Based on this, a recurrent neural network (RNN) is used to model the prior probability of each intent layer. RNNs are capable of processing time series data and can capture the dynamic changes of intent over time. This prior modeling of intent using an RNN can better reflect the temporal patterns of target behavior. To infer the target's intent, a Bayesian inference method is used to calculate the posterior probability. Bayesian inference combines observed data and prior knowledge to produce a comprehensive probability distribution of intent. In practical calculations, since the posterior distribution is often complex and difficult to solve, variational inference can be used to approximate it. Variational inference finds an effective approximate solution by defining an approximate distribution and minimizing its KL divergence with the true posterior distribution. During the optimization process, importance sampling is used to approximate the expectation, combined with the stochastic gradient descent (SGD) algorithm to optimize the variational inference parameters, allowing the model to gradually approach the optimal solution during training.

[0056] Step 3: Define the likelihood function of each intent layer; use a recursive neural network to model the prior probability; use Bayes' rule to calculate the posterior probability; iteratively perform the following process to complete the training: use variational inference to define an approximate posterior distribution, and then minimize the KL divergence; use importance sampling to approximate the expectation, and use stochastic gradient descent to optimize the parameters of variational inference; when the number of iterations reaches the set number, execute step 4:

[0057] The core of step 3 is modeling and inferring the multi-layer intent hierarchy using recurrent neural networks (RNNs) and Bayesian inference. This process involves defining a likelihood function for each intent layer, modeling prior probabilities using RNNs, and calculating posterior probabilities using Bayes' rule. Finally, optimization and approximation are performed through variational inference and importance sampling to complete the training and inference of the target behavior intent. First, in the multi-layer intent hierarchy, each intent layer can be considered a latent variable. By defining a likelihood function for each intent layer, we can describe the distribution of observed data under different intents. The likelihood function is the probability distribution of observed data given an intent. By maximizing the likelihood function, we can find the most likely explanation for the intent. This process is based on the relationship between tensor features and intent. The definition of the likelihood function links observed data with intent, providing a foundation for subsequent inference. Next, a recurrent neural network (RNN) is used to model the prior probabilities of each intent layer. RNNs are capable of processing time series data and can capture the dynamic changes of intent over time. Specifically, RNNs, through their internal loop structure, are able to memorize and process long-range dependencies in sequential data, which is crucial for modeling the temporal dynamics of target behavior. During the RNN modeling process, the intent state at each moment depends not only on the current observed data but also on the intent state at the previous moment, forming a recursive dependency. Using RNNs to model intent a priori can better reflect the temporal patterns of target behavior, making the model more accurate and robust when processing complex temporal data. To infer the target's intent, Bayesian inference methods are used to calculate the posterior probability. Bayesian inference combines observed data and prior knowledge to produce a comprehensive probability distribution of intent. Specifically, Bayesian inference uses the Bayesian formula to combine the prior probability and the likelihood function to calculate the posterior probability. The posterior probability represents the probability distribution that the target has a certain intent, given the observed data. Because posterior distributions are often complex and difficult to solve, and direct calculation is very difficult, approximate inference methods are often used in practical applications to simplify the calculation process. In this invention, variational inference methods are used to approximate the posterior distribution. Variational inference is an optimization technique that finds an effective approximate solution by defining an approximate distribution and minimizing the KL divergence between it and the true posterior distribution. Specifically, variational inference transforms complex integral calculations into an optimization problem, searching for a parameterized approximate distribution that is as close as possible to the true posterior distribution. During the optimization process, importance sampling is used to approximate the expectation. Importance sampling is a Monte Carlo method that estimates the expected value by reweighting samples. This method can achieve higher estimation accuracy with a smaller number of samples, improving computational efficiency. In addition, the stochastic gradient descent (SGD) algorithm is used when optimizing the parameters of variational inference.SGD is a highly efficient optimization algorithm that uses the gradient information of the current parameters at each iteration to gradually adjust the parameters, allowing the objective function to converge to the optimal solution. In this paper, SGD is used to minimize the KL divergence in variational inference, allowing the approximate distribution to better approximate the true posterior distribution. Through this optimization process, the model can gradually approach the optimal solution during training, thereby improving the accuracy and robustness of intent inference.

[0058] When the set number of iterations is reached, the training process is complete, and the model has achieved a high level of inference capability. Based on this, intent inference is performed using an approximate posterior distribution, which represents the probability distribution of the target's intent given the observed data. This probability distribution allows the most likely intent state of the target to be inferred and the uncertainty of intent inference to be calculated. Uncertainty represents the confidence level of the model in its inference of the target's intent given the observed data. Specifically, uncertainty can be measured using the entropy of the posterior distribution. Higher entropy values ​​indicate greater uncertainty in the model's inferred intent; conversely, lower entropy values ​​indicate greater confidence in the model's inferred intent.

[0059] Step 4: Use the approximate posterior distribution to perform intent inference and calculate the uncertainty of intent inference.

[0060] The approximate posterior distribution represents the probability distribution of a target having a certain intent, given observed data. By maximizing this probability distribution, the target's most likely behavioral intention can be inferred. This inference process is based on Bayesian inference, combining prior probabilities with the likelihood of observed data. This ensures that the inference results not only consider the influence of historical data but also incorporate current observations, resulting in high reliability and accuracy. In practice, when using the approximate posterior distribution for intent inference, all possible intent states are traversed, the posterior probability of each state is calculated, and the state with the highest probability is selected as the target's predicted intent. While this process may appear simple, it involves high-dimensional data and complex probability calculations, requiring efficient computational methods and powerful computing power. Using variational inference and recurrent neural networks can effectively simplify the computational process and improve efficiency, thus enabling real-time intent inference. Calculating uncertainty in intent inference is another key step. Uncertainty assessment is crucial in practical applications, as it provides the model with a measure of confidence in the current inference result. When the model's uncertainty is high, it means that the current observations are insufficient to support a definitive inference. In this case, additional observations or further analysis may be needed to improve the reliability of the inference. Conversely, when the model's uncertainty is low, it means the model has high confidence in the current inference result, allowing appropriate decisions to be made based on the inference result. Uncertainty calculations are typically based on the entropy of the posterior distribution. Entropy is a measure of the uncertainty of a probability distribution. Higher entropy values ​​indicate a more dispersed distribution and greater uncertainty, while lower entropy values ​​indicate a more concentrated distribution and less uncertainty. In this method, by calculating the entropy of the posterior distribution, the model's confidence in each intent inference result can be quantified. This process not only helps understand the model's behavior but also provides important information support for practical decision-making. Combining the above steps, the method of the present invention enables efficient and accurate intent inference and uncertainty assessment when analyzing militarized target behavior. By preprocessing and fusing multi-source data, defining a multi-layer intent hierarchy, combining recursive neural networks with Bayesian inference, and optimizing variational inference and importance sampling, a powerful analytical model is constructed. This model not only handles complex time series data and multi-dimensional information, but also provides reasonable assessments in the face of uncertainty, thereby improving the reliability and practicality of analytical results.

[0061] Example 2: The multi-layer intention hierarchy in step 2 includes at least three intention layers, namely: a strategic intention layer, a tactical intention layer and a specific action intention layer; the strategic intention layer includes at least the following intentions: attack, defense, retreat, reorganization and deception; the tactical intention layer includes at least the following intentions: encirclement, breakthrough, reconnaissance, blocking, detour and feint; the specific action intention layer includes at least the following intentions: deploying weapon systems, establishing fortifications, carrying out air strikes, conducting electronic interference, launching ground attacks, conducting special operations, and establishing supply lines.

[0062] Specifically, the strategic intent layer is the highest level of intent, primarily used to describe the macro-strategic decisions of militarized targets. Intent at this level includes attack, defense, retreat, reorganization, and deception. Offensive intent indicates the target's primary goal is to launch a proactive attack to suppress enemy forces. Defensive intent is intended to protect its own forces and resist enemy attacks. Retreat intent indicates the target chooses to retreat in the current situation to avoid unnecessary losses or to make strategic adjustments. Reorganization intent indicates the target is reorganizing its forces, perhaps to fight more effectively or adjust its strategic deployment. Deception intent aims to achieve strategic advantage through disguise or confusion, causing the enemy to misjudge its true intentions. These strategic intents provide the macro-level direction and framework for overall military operations. Secondly, the tactical intent layer is the middle level of intent, primarily used to describe the tactical actions of militarized targets in specific combat situations. Intent at this level includes encirclement, breakthrough, reconnaissance, blocking, detour, and feint. Encirclement intent indicates the target attempts to encircle the enemy, preventing them from escaping or launching an effective counterattack. The intent to penetrate involves launching a powerful assault at a weak point in the enemy's defenses to break them. The intent to reconnaissance involves gathering intelligence to understand enemy movements and deployments. The intent to block an enemy attack or advance is to establish defensive lines at key locations. The intent to bypass the enemy's main force and attack its flanks or rear, achieving an effect of surprise. The intent to feint involves faking an offensive posture to attract the enemy's attention and misjudge one's true actions. These tactical intents provide tactical guidance for specific actions in actual combat. Finally, the specific action intent layer, the lowest level of intent, primarily describes the specific operations to be carried out by militarized objectives. Intentions at this level include deploying weapons systems, establishing fortifications, conducting air strikes, conducting electronic jamming, launching ground offensives, conducting special operations, and establishing supply lines. The intent to deploy weapons systems involves placing weapons at specific locations to enhance firepower coverage. The intent to establish fortifications involves building defensive facilities to enhance one's own defensive capabilities. The intent to conduct an air strike means the goal is to strike the enemy through air power. The intent to conduct electronic jamming involves disrupting enemy communications and radar systems through electronic warfare. The intent to launch a ground offensive means the goal is to attack enemy positions through ground forces. The intent to conduct special operations means to conduct high-precision, high-risk military operations using special forces. The intent to establish supply lines means the goal is to ensure logistical support to support frontline operations. These specific operational intents provide detailed guidance for each step in actual operations.

[0063] For each layer, a probability distribution is output, indicating the probability of each possible intent. For example:

[0064] (Layer 1);

[0065] (Layer 2);

[0066] (Layer 3).

[0067] The most likely intent combination: By selecting the intent with the highest probability in each layer, the most likely intent combination can be obtained: ; For example: (attack, break through, launch a ground attack).

[0068] Confidence of intent: For each layer of intent, its entropy can be calculated to represent the uncertainty of the reasoning: ; Lower entropy values ​​indicate higher confidence.

[0069] Temporal Intention Reasoning: Considering the dynamic nature of military operations, intention reasoning can be extended in the temporal dimension: ; This allows tracking changes in intent over time.

[0070] Conditional intent prediction: Based on the current inferred intent, possible future intent changes can be predicted: .

[0071] This multi-level, dynamic intention reasoning provides rich information for military decision-making, including not only the most likely current intention combination, but also the probability distribution of various possibilities and the trend of intention changes over time. This information can be used to formulate corresponding strategies, tactics, and specific action plans.

[0072] Example 3: The multi-source data includes at least the following categories of data: position data, speed data, acceleration data, direction data, environmental data, altitude data, longitude data and latitude data; the process of performing data preprocessing on each category of the multi-source data in step 1 includes: performing data noise reduction and data normalization processing on the data of each category; in step 1, after performing timestamp alignment and spatial registration on the preprocessed data of each category, resampling and interpolation processing on the data are performed to fuse them into an input tensor; in step 2, using a deep neural network to extract tensor features of the input tensor .

[0073] Specifically, multi-source data comes from a wide range of sources, and each category of data has different physical quantities and units. Therefore, preprocessing this data is essential. Data denoising is the first step in preprocessing. Using filters or other noise reduction techniques, noise is removed from the data, preserving the useful signal. Common noise reduction methods include low-pass filtering, high-pass filtering, and band-pass filtering. These methods can effectively remove high-frequency noise or low-frequency interference, thereby improving the signal-to-noise ratio of the data. Data normalization is the second step in preprocessing. Through normalization, data of different categories are converted to the same scale, facilitating subsequent processing and analysis. Normalization methods include min-max normalization and z-score standardization. These methods can narrow the data value range to a standard interval, such as [0, 1] or [-1, 1], thereby eliminating dimensional differences between different data categories. After preprocessing, data needs to be timestamp aligned and spatially registered to ensure consistency across different data sources at the same time and spatial location. Timestamp alignment involves aligning data from different data sources by timestamp to ensure that data of all categories is captured at the same time. This step is typically accomplished through interpolation or resampling. Interpolation methods include linear interpolation and spline interpolation, while resampling methods include downsampling and upsampling. These methods can map data of different temporal resolutions onto a unified timeline. Spatial registration refers to the spatial alignment of data from different data sources to ensure that all categories of data are captured at the same spatial location. This step is typically accomplished through geometric and coordinate transformations, such as using camera calibration techniques to align image data and lidar data, or using GPS coordinate conversion to align geographic data. After timestamp alignment and spatial registration, the data is resampled and interpolated to form a single input tensor. The input tensor is a high-dimensional data structure that contains multidimensional information about time, space, and multiple sources. Resampling and interpolation can further improve the consistency and integrity of the data. Through these processes, the data is fused into a unified high-dimensional representation, making subsequent feature extraction and behavioral analysis more efficient and accurate. In step 2, a deep neural network is used to extract tensor features H from the input tensor. Deep neural networks, particularly convolutional neural networks (CNNs), have strong capabilities for extracting features from high-dimensional data. Through its hierarchical structure and convolution operations, CNNs can automatically learn and capture spatial and temporal patterns in data, generating a low-dimensional but information-rich feature vector. Specifically, the input tensor passes through multiple convolutional, pooling, and fully connected layers, gradually extracting features at different levels. These features not only contain the spatiotemporal information of the original data, but also contain deeper semantic information. Convolutional layers effectively extract spatial features through local receptive fields and weight sharing mechanisms; pooling layers reduce the dimensionality of the data through downsampling operations to prevent overfitting; and fully connected layers integrate different features through linear transformations to generate the final tensor features.

[0074] Example 4: In step 2, define Layer intention structure, Layer Intent Layer possible intentions; initialize the intent vector for each layer:

[0075] ;

[0076] in, , is the index number; Indicates the The first layer of the intention layer intention; each is a discrete variable and can be possible values;

[0077] Tensor features The nonlinear transformation is performed by the following formula to obtain the feature , to accommodate different levels of intent reasoning:

[0078] ;

[0079] in, represents element-wise multiplication, and They are the preset feature weight matrix and feature bias matrix respectively;

[0080] Use conditional probability to represent the dependencies between intent layers:

[0081] ;

[0082] in, Expressing the highest level of intention, Indicates the lowest level of intention; For a given feature The joint probability distribution of the entire intent hierarchy when ; For a given When considering layer intentions and features, the conditional probability of the lowest layer intention; For a given When layering intent and features, Conditional probability of layer intention; The conditional probability of the top-level intent given the features.

[0083] Specifically, in this multi-layer intent hierarchy, each intent layer is defined as a A vector of possible intentions . This definition allows each intent layer to express the specific behavioral intentions of the target at different levels. For example, the highest-level intent can represent macro-strategic decisions, such as attack, defense, retreat, etc.; the middle-level intent can represent specific tactical actions, such as encirclement, breakthrough, reconnaissance, etc.; and the lowest-level intent can be refined to specific operations, such as deploying weapon systems, building fortifications, etc. Such a hierarchical design can not only capture the diversity and complexity of target behavior, but also provide a clearer hierarchical structure during analysis, making the relationship between each intent layer clearer. In order to adapt to the intention reasoning at different levels, the preprocessed tensor features are processed. Perform nonlinear transformation to obtain features This process is achieved through a transformation formula with a nonlinear activation function, which can deeply process the original features and make them better suited for subsequent intention reasoning. Through this transformation, the model can capture more complex relationships and patterns between features, thereby providing richer information support in the reasoning process. Nonlinear transformation plays a vital role in this process, making the features It not only contains the basic information of the original data but also reflects higher-level abstract features. The key to establishing a multi-layered intent hierarchy is the use of conditional probability to represent the dependencies between intent layers. By defining the conditional probability of each layer's intent, the joint probability distribution of each layer's intent can be derived layer by layer. This conditional probability representation method effectively captures the dependencies between intents at different levels, ensuring that each layer's intent depends not only on the characteristics of the current layer but also on the influence of the intent of the previous layer. For example, tactical intent depends not only on current observation data but is also guided by strategic intent; specific action intent is influenced by both the current tactical intent and the constraints of the overall strategy. Through layer-by-layer reasoning and calculation, the probability distribution of each layer's intent, given the given characteristics and the previous layer's intent, can be accurately determined, enabling comprehensive analysis and inference of target behavior. This combination of a multi-layered intent hierarchy and conditional probability makes the method of the present invention highly accurate and robust when processing complex militarized target behavior. In practical applications, through layer-by-layer derivation and reasoning, the probability distribution of each layer's intent can be updated and adjusted in real time, dynamically reflecting changes in target behavior. Whether at the strategic level or at the tactical and operational levels, it can provide accurate intention analysis and prediction, which not only improves the intelligent agent's ability to understand target behavior, but also provides strong technical support for military decision-making.

[0084] Example 5: In step 3, the following formula is used to define Likelihood function of the layer intention layer , which contains a multi-layer perceptron with a main branch and a multi-layer sensor with a parallel branch. The main branch uses the tanh activation function, and the parallel branch uses the LeakyReLU activation function, which is expressed by the following formula:

[0085] ;

[0086] in, From the 1st to the The intention sequence of layers; is the embedding representation of the low-level intent, which is a function that maps a discrete sequence of intents to a continuous vector space. Let the dimension after embedding be ;set up The latitude is , is the concatenation of feature and low-level intent embedding, with dimension ; For the The weight matrix of the layer intention layer has the dimension ,in It is Dimensions of the layer intention layer; For the The bias vector of the layer intent layer has the dimension ; For the The weight matrix of the layer intention layer has the dimension ,in is the dimension of the output; For the The bias vector of the layer intent layer has a dimension of ; For the parallel branch The weight matrix of the layer intention layer has the dimension ; For the parallel branch The bias vector of the layer intent layer has a dimension of ; For the parallel branch The weight matrix of the layer intention layer has the dimension ; For the parallel branch The bias vector of the layer intent layer has a dimension of ; is the activation function of the rectified linear unit with leakage, is a function variable; is the Softmax function, which is used to convert the output into a probability distribution; is the Sigmoid activation function, which is used to compress the output to the (0,1) interval.

[0087] Specifically, the likelihood function The calculation is done through the joint action of the main branch and the parallel branch. The core of the main branch is the tanh activation function, which can introduce nonlinear factors to make the feature transformation more complex and diverse. The output of the tanh function is This makes it work better when processing signed data. The parallel branch uses the LeakyReLU activation function, which is a leaky rectified linear unit activation function that can retain a certain gradient when the input is negative, avoiding the problem of the traditional ReLU gradient being zero in the negative area. The form of LeakyReLU is , through this activation function, the model can better handle sparse and negative features. In the specific implementation, the input feature The embedding representation of the lower-level intent is concatenated to form a new feature vector This operation integrates information from different levels, allowing the model to simultaneously consider the influence of the current feature and the previous layer's intention. After this splicing operation, the weight matrix and the bias vector Perform linear transformation to generate an intermediate feature vector. The intermediate features in the main branch are processed by the tanh activation function to output a feature vector after nonlinear transformation; the intermediate features in the parallel branch are processed by the LeakyReLU activation function to output another feature vector after nonlinear transformation. The outputs of these two branches are normalized by the softmax function and the sigmoid function respectively. The softmax function is used to convert the output into a probability distribution, while the sigmoid function is used to compress the output to interval. The final result is the element-by-element product of the two. This design makes the model more flexible and efficient in processing and fusing multiple features. By combining different activation functions and normalization methods, the model can capture both linear and nonlinear feature relationships, thereby making more accurate inferences in complex environments. In militarized target behavior analysis, this combination of multi-layer intention structure and conditional probability enables the calculation of each layer of intention to take into account the intention of the previous layer and the current features. For example, in militarized scenarios, low-level intentions such as specific action intentions (such as deploying weapon systems, building fortifications, etc.) are affected by mid-level intentions (such as tactical intentions such as encirclement and breakthrough) and high-level intentions (such as strategic intentions such as attack and defense). Through this layer-by-layer processing method, the model can more accurately infer the actual intention of the target. In addition, the use of embedding representation to map discrete intention sequences to continuous vector space enables the model to better handle discrete variables. Dimensions of embedding representation and features Dimension The combined feature vectors undergo linear transformations using weight matrices and bias vectors, and then undergo processing using nonlinear activation functions to generate intermediate features. These intermediate features are then further processed to produce the final output. This multi-level, multi-branch structural design allows the model to excel in analyzing complex militarized target behavior. In step 3, the likelihood function of each layer of intent not only considers the features of the current layer, but also considers the likelihood of the target being targeted. , and also combines the first to the Layer Intent Sequence . By embedding , these low-level intentions are mapped into a continuous vector space so that they can be concatenated and processed with the current features. Weight matrix and Linear transformation for the main branch and parallel branch, bias vector and The output of the main branch is processed by the tanh activation function and converted to a probability distribution by the softmax function; the output of the parallel branch is processed by the LeakyReLU activation function and compressed to This multi-layered structure and activation function combination ensures that the calculation of each layer's intent takes into account the intent of the previous layer and the current features. This hierarchical approach can capture the complex dependencies within the target's behavior. For example, in a military scenario, low-level intent, such as specific action intentions, can be influenced by mid-level and high-level intents. Through this layered approach, the model can more accurately infer the target's actual intent.

[0088] Example 6: Using the following formula, a recursive neural network is used to model the Prior probability of the intention layer , each intent layer corresponds to a hidden layer of the recurrent neural network:

[0089] ;

[0090] ;

[0091] in, represents a gated recurrent unit; The recurrent neural network corresponds to Hidden layer of the intent layer; The recurrent neural network corresponds to Hidden layer of the intent layer; is the weight matrix of the recurrent neural network; is the bias matrix of the recurrent neural network.

[0092] Specifically, the core of recurrent neural networks lies in their ability to process sequence data. By combining the hidden state of the previous moment with the current input and updating the current hidden state, RNN can effectively capture the dependencies in the time series. In this invention, each intent layer corresponds to a hidden layer of the recurrent neural network. , updated through the gated recurrent unit (GRU). GRU solves the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequence data by introducing update gates and reset gates. In the operation of GRU, the update gate determines how to combine the hidden state of the previous moment with the current input to generate a new hidden state; the reset gate controls how to forget part of the hidden state of the previous moment and focus on the influence of the current input. Through this mechanism, GRU can dynamically adjust the degree of dependence on historical information and current information to capture information over a longer time range. For the prior probability modeling of the intent layer in the present invention, the intent embedding representation of each layer and the hidden state of the previous layer Together as the input of GRU, update the hidden state of the current layer This processing method can transmit and integrate the intention information of the previous layer layer by layer to form a comprehensive judgment of the intention of the current layer. In order to convert the hidden state into the intention probability distribution, the present invention uses the softmax function. By performing linear transformation and combining it with the softmax activation function, we can get the conditional probability distribution of the current layer intent. This method can effectively map continuous hidden states to a discrete intent space and generate the probability of each intent. In specific applications, this recursive modeling method has significant advantages. First, through the GRU's gating mechanism, the model can flexibly handle long-term dependencies, ensuring accurate capture of target behavior. Second, through layer-by-layer recursive modeling, the model can effectively integrate the complex relationships of multiple layers of intent, thereby improving the accuracy of inference. In military target behavior analysis, this method can accurately model complex multi-layer intent relationships, such as intent associations at the strategic, tactical, and specific action levels, enabling intelligent agents to more accurately predict and analyze target behavior. In addition, using embedding representation to map discrete intent sequences to a continuous vector space enables the model to process discrete variables. The dimensions and features of the embedding representation are combined to form a new feature vector, which is linearly transformed by the weight matrix and bias vector, and then processed by the GRU to generate new hidden states. These hidden states are then converted into probability distributions through the softmax function to generate the prior probability of each layer of intent.

[0093] Example 7: Using the following formula, Bayesian rule is used to calculate Posterior probability of the intention layer :

[0094] .

[0095] in, is the subscript index; using variational inference, define an approximate posterior distribution using the following formula:

[0096] ;

[0097] in, is the parameterized approximate posterior distribution; are the parameters of variational inference; is the weight matrix for variational inference; is the bias vector for variational inference.

[0098] Specifically, the core idea of ​​Bayes' rule is to combine the currently observed data (i.e., features ) and prior knowledge (i.e., previous intention ), to update the probability distribution of the target intention. Specifically, the likelihood function Describes the given intention Under the conditions of its previous layer intention, the current feature is observed The prior probability It reflects the fact that before observing the current feature, By combining these two, we can get a comprehensive posterior probability , which is used to more accurately infer the actual intention of the target. In order to effectively calculate the posterior probability, the present invention introduces variational inference. Variational inference defines a parameterized approximate posterior distribution , to approximate the true posterior distribution. This approximate posterior distribution is formed by concatenating the features and the embedded representation of the previous layer's intent, undergoing a weighted linear transformation, and then converting it into a probability distribution through the softmax function. Through this parameterized method, variational inference can transform the calculation problem of the posterior probability into a parameter optimization problem, thereby simplifying the calculation process. In the specific implementation of variational inference, the goal is to optimize the parameters by minimizing the KL divergence between the approximate posterior distribution and the true posterior distribution. . KL divergence is a measure of the difference between two probability distributions. By minimizing the KL divergence, the approximate posterior distribution can be made as close as possible to the true posterior distribution. During the optimization process, optimization algorithms such as stochastic gradient descent (SGD) are used to iteratively update the parameters of the variational inference and gradually approach the optimal solution. The advantage of this method is that it can efficiently handle complex probability distribution calculation problems, especially when involving large amounts of data and high-dimensional space. By introducing a parameterized approximate posterior distribution, the present invention effectively reduces the computational complexity, making the model more efficient and accurate when processing complex multi-layer intent inference. In militarized target behavior analysis, this method is particularly suitable for processing large amounts of complex data and highly uncertain scenarios. For example, in a complex military scenario, it may be necessary to analyze the intentions of the target at different levels, including strategic-level intentions such as attack and defense, and tactical-level intentions such as encirclement and assault. Through the method of the present invention, the observed feature data and the intention information of the previous level can be combined to calculate the posterior probability of the target intention at each level, thereby providing a comprehensive and detailed intention analysis. In addition, the use of embedding representation to map the discrete intention sequence to a continuous vector space enables the model to process discrete variables. The dimensions and features of the embedded representation are combined to form a new feature vector. This vector undergoes a linear transformation using a weight matrix and bias vector, and then passes through a nonlinear activation function to generate a new feature representation. These feature representations are then processed using variational inference to generate an approximate posterior probability distribution for each layer of intent. Through layer-by-layer calculation and updates, this approach captures the complex dependencies between intents at different levels, thereby improving the accuracy of intent inference.

[0099] Example 8: Minimize the KL divergence using the following formula:

[0100] ;

[0101] in, is the loss function; represents the KL divergence; importance sampling is used to approximate the expectation using the following formula:

[0102] +const;

[0103] in, is the number of sampling times; is the kth intention layer obtained by the sth sampling: const is a constant term; the parameters of variational inference are optimized using stochastic gradient descent using the following formula:

[0104] ;

[0105] in, For the Parameters of variational inference after iterative optimization; The parameters of the variational inference after optimization at the t+1th iteration; is the learning rate; is the gradient of the loss function with respect to the parameter θ.

[0106] Specifically, in the present invention, the loss function Used to represent KL divergence. By minimizing this function, the parameters of variational inference can be optimized. The form of the loss function integrates the log-likelihood function, the logarithm of the prior distribution, and the logarithm of the approximate posterior distribution. Specifically, this process involves calculating the logarithm of the likelihood function and the prior distribution and subtracting the logarithm of the approximate posterior distribution to obtain the loss function During the calculation process, the evidence item It is usually regarded as a constant term and can be ignored to simplify the calculation. In order to approximate the expected value, the present invention uses the importance sampling method. Importance sampling is a Monte Carlo method that estimates the expected value by sampling the distribution multiple times. In this method, by sampling the samples multiple times, an approximation of the expected value can be obtained, thereby effectively estimating the expected value of the loss function. Specifically, the approximate representation of the loss function can be estimated by taking a weighted sum of the intentions obtained by sampling. This approximate calculation method can greatly reduce the computational complexity while maintaining a high accuracy. During the optimization process, the present invention uses stochastic gradient descent (SGD) to adjust the parameters of variational inference SGD is an iterative optimization algorithm that calculates the loss function About parameters Gradient , gradually update the parameters to minimize the loss function. At each iteration, by adjusting the parameters By making small adjustments, we can gradually approach the optimal solution. The update formula is ,in For the The parameters after iterations, For the The parameters after iterations, is the learning rate. Through this iterative optimization approach, the optimal parameters can be gradually approached, making the approximate posterior distribution as close as possible to the true posterior distribution. This optimization process is of great significance in the analysis of militarized target behavior. In complex military scenarios, target behavior often exhibits a high degree of uncertainty and complex dependencies. By minimizing the KL divergence, the model can more accurately capture these complex relationships, thereby improving the ability to predict and analyze target behavior. For example, when analyzing a target's strategic intent, the probability distribution of the target's intent under different scenarios can be calculated using Bayesian inference, combining currently observed data with prior knowledge. These probability distributions can then be further optimized by minimizing the KL divergence to better reflect the actual situation. In practical applications, the combination of importance sampling and stochastic gradient descent enables the model to maintain efficiency and accuracy even when processing large amounts of data. Importance sampling effectively estimates the expected value by sampling the distribution multiple times, thus avoiding the problem of directly calculating complex integrals. Stochastic gradient descent, on the other hand, can find the optimal solution in a relatively short time by gradually optimizing the parameters. This method is particularly suitable for processing high-dimensional spaces and large amounts of data, making the present invention widely applicable in practical applications. To implement this optimization process, the prior probabilities of intent at each layer must first be modeled. These prior probabilities are represented using a recurrent neural network (RNN), with the prior probability of each layer's intent dependent on the hidden state of the previous layer. This recursive modeling approach captures the complex dependencies between intents at each layer, providing accurate prior information for subsequent variational inference. When calculating the likelihood function, the likelihood values ​​corresponding to each layer's intent are derived by combining the current features with the lower-level intent embeddings. These likelihood values ​​are then combined with the prior probabilities to form a complete posterior probability calculation framework. In practical applications, this optimization method enables the present invention to efficiently handle complex multi-layer intent inference problems. For example, in a complex military scenario, it may be necessary to analyze the target's intent at different levels, including strategic intent such as attack and defense, and tactical intent such as encirclement and assault. The present method combines observed feature data with intent information from previous layers to calculate the posterior probability of the target's intent at each layer, providing a comprehensive and detailed intent analysis. This analysis not only considers current observation data, but also combines historical data and the complex dependencies between intents at each layer, making intent inference more accurate.

[0107] Example 9: In step 4, the approximate posterior distribution is used to perform intent inference using the following formula:

[0108] ;

[0109] in, is the result of intention reasoning; then use the following formula to calculate the uncertainty of intention reasoning:

[0110] ;

[0111] in, Reasoning about uncertainty in intentions.

[0112] Specifically, the core of intention reasoning is to use the approximate posterior distribution To determine the most likely intent. Specifically, by calculating the probability of each possible intent and selecting the one with the highest probability, the intention inference result is obtained. This process ensures that the inferred intent is the best explanation based on the current observation data and the previous intention sequence, providing the most likely prediction of the target's behavior. This method leverages the efficient computational properties of the approximate posterior distribution, enabling the model to quickly and accurately infer intent in complex and changing environments. In addition to inferring the most likely intent, the present invention also places special emphasis on assessing the uncertainty of the inference results. By calculating the entropy of the approximate posterior distribution to measure uncertainty, the reliability of the prediction results can be quantified. Entropy is a measure of the dispersion of a probability distribution. Higher entropy values ​​indicate a more dispersed distribution and greater uncertainty in the inferred results; lower entropy values ​​indicate a more concentrated distribution and higher confidence in the inferred results. This method effectively quantifies the uncertainty of the prediction results, allowing decision makers to more clearly understand the reliability of the model's inferences. This uncertainty assessment is particularly important in militarized target behavior analysis. Decision makers not only need to know the most likely target intent but also understand the credibility of the inferred results. For example, in a complex tactical scenario, if the uncertainty of the inference result is high, the decision maker may need more information or further analysis to confirm the target's intention; if the uncertainty is low, the decision maker can act more confidently based on the inference result. The specific implementation of this method includes the following steps. First, the probability of each possible intention is calculated using the approximate posterior distribution, and the intention with the highest probability is selected as the inference result. This process is achieved by maximizing the approximate posterior distribution, that is, finding the intention that makes The greatest intention . Then, the uncertainty of the inference result is evaluated by calculating the entropy of the approximate posterior distribution. The calculation of entropy involves summing and taking the logarithm of the probability distribution of all possible intentions, thereby obtaining a quantitative uncertainty measure. Through this method, the present invention can provide not only accurate intention inference results, but also provide a clear confidence assessment for these results. In practical applications, this method can help military commanders rely more confidently on the inference results of the model when formulating strategies. For example, when faced with complex enemy actions, commanders can use the enemy intentions inferred by the model and the corresponding uncertainty assessment to respond quickly and formulate corresponding tactics.

[0113] Example 10: A militarized target behavior analysis intelligent agent testing device, characterized in that it includes: a data acquisition part, used to acquire multi-source data of militarized targets, perform data preprocessing on each category of data in the multi-source data, perform timestamp alignment and spatial registration, and then fuse them into an input tensor; an intention model construction part, used to extract tensor features of the input tensor; based on the tensor features, define a multi-layer intention hierarchy, each intention layer contains multiple possible intentions, and use conditional probability to represent the dependency relationship between the intention layers; an intention reasoning model construction part, used to define the likelihood function of each intention layer; use a recursive neural network to model the prior probability; use the Bayesian rule to calculate the posterior probability; iteratively execute the following process to complete training: use variational inference to define an approximate posterior distribution, and then minimize the KL divergence; use importance sampling to approximate expectations, and use stochastic gradient descent to optimize the parameters of variational inference; when the number of iterative executions reaches the set number, execute step 4: the intention reasoning part, used to perform intention reasoning using the approximate posterior distribution, and calculate the uncertainty of intention reasoning.

[0114] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these specific embodiments are merely illustrative, and that those skilled in the art may omit, substitute, and modify the details of the methods and systems described above without departing from the principles and spirit of the present invention. For example, combining the above method steps to perform substantially the same functions in substantially the same manner to achieve substantially the same results falls within the scope of the present invention. Accordingly, the scope of the present invention is limited solely by the appended claims.

Claims

1. A method for testing a militarized target behavior analysis agent, characterized in that: The method comprises: Step 1: Obtain multi-source data of militarized targets, pre-process the data of each category in the multi-source data, align the timestamps and spatially register them, and then fuse them into an input tensor; Step 2: Extract tensor features from the input tensor; based on the tensor features, define a multi-layer intent hierarchy, where each intent layer contains multiple possible intents, and use conditional probabilities to represent the dependencies between intent layers. Step 3: Define the likelihood function of each intent layer; use a recursive neural network to model the prior probability; use Bayes' rule to calculate the posterior probability; iteratively perform the following process to complete the training: use variational inference to define an approximate posterior distribution, and then minimize the KL divergence; use importance sampling to approximate the expectation, and use stochastic gradient descent to optimize the parameters of variational inference; when the number of iterations reaches the set number, execute step 4: Step 4: Use the approximate posterior distribution to perform intention inference and calculate the uncertainty of intention inference; In step 2, define Layer intention structure, Layer Intent Layer possible intentions; initialize the intent vector for each layer: ； in, , is the index number; Indicates the The first layer of the intention layer intention; each is a discrete variable and can be possible values; Tensor features The nonlinear transformation is performed by the following formula to obtain the feature , to accommodate different levels of intent reasoning: ； in, represents element-wise multiplication, and They are the preset feature weight matrix and feature bias matrix respectively; Use conditional probability to represent the dependencies between intent layers: ； in, Expressing the highest level of intention, Indicates the lowest level of intention; For a given feature The joint probability distribution of the entire intent hierarchy when ; For a given When considering layer intentions and features, the conditional probability of the lowest layer intention; For a given When layering intent and features, Conditional probability of layer intention; The conditional probability of the top-level intent given the features.

2. The militarized target behavior analysis agent testing method according to claim 1, characterized in that: The multi-layered intention hierarchy in step 2 includes at least three intention layers, namely: strategic intention layer, tactical intention layer and specific action intention layer; the strategic intention layer includes at least the following intentions: attack, defense, retreat, reorganization and deception; the tactical intention layer includes at least the following intentions: encirclement, breakthrough, reconnaissance, blocking, detour and feint; the specific action intention layer includes at least the following intentions: deploying weapon systems, establishing fortifications, carrying out air strikes, conducting electronic interference, launching ground attacks, conducting special operations, and establishing supply lines.

3. The militarized target behavior analysis agent testing method according to claim 2, characterized in that: The multi-source data includes at least the following categories of data: position data, speed data, acceleration data, direction data, environmental data, altitude data, longitude data and latitude data; Step 1: The process of preprocessing the data of each category in the multi-source data includes: performing data noise reduction and data normalization on the data of each category; In step 1, the pre-processed data of each category is timestamp aligned and spatially registered, and then the data is resampled and interpolated to form an input tensor; in step 2, the tensor features of the input tensor are extracted using a deep neural network. .

4. The militarized target behavior analysis agent testing method according to claim 3, characterized in that: In step 3, the following formula is used to define Likelihood function of the layer intention layer , which contains a multi-layer perceptron with a main branch and a multi-layer sensor with a parallel branch. The main branch uses the tanh activation function, and the parallel branch uses the LeakyReLU activation function, which is expressed by the following formula: ； in, From the 1st to the The intention sequence of layers; is the embedding representation of the low-level intent, which is a function that maps a discrete sequence of intents to a continuous vector space. Let the dimension after embedding be ;set up The latitude is , is the concatenation of feature and low-level intent embedding, with dimension ; For the The weight matrix of the layer intention layer has the dimension ,in It is Dimensions of the layer intention layer; For the The bias vector of the layer intent layer has a dimension of ; For the The weight matrix of the layer intention layer has the dimension ,in is the dimension of the output; For the The bias vector of the layer intent layer has a dimension of ; For the parallel branch The weight matrix of the layer intention layer has the dimension ; For the parallel branch The bias vector of the layer intent layer has a dimension of ; For the parallel branch The weight matrix of the layer intention layer has the dimension ; For the parallel branch The bias vector of the layer intent layer has a dimension of ; is the activation function of the rectified linear unit with leakage, is a function variable; is the Softmax function, which is used to convert the output into a probability distribution; is the Sigmoid activation function, which is used to compress the output to the (0,1) interval.

5. The militarized target behavior analysis agent testing method according to claim 4, characterized in that: Using the following formula, we use a recurrent neural network to model the Prior probability of the intention layer , each intent layer corresponds to a hidden layer of the recurrent neural network: ； ； in, represents a gated recurrent unit; The recurrent neural network corresponds to Hidden layer of the intent layer; The recurrent neural network corresponds to Hidden layer of the intent layer; is the weight matrix of the recurrent neural network; is the bias matrix of the recurrent neural network.

6. The militarized target behavior analysis agent testing method according to claim 5, characterized in that: The Bayesian rule is used to calculate the Posterior probability of the intention layer : ； in, is the subscript index; using variational inference, define an approximate posterior distribution using the following formula: ； in, is the parameterized approximate posterior distribution; are the parameters of variational inference; is the weight matrix for variational inference; is the bias vector for variational inference.

7. The militarized target behavior analysis agent testing method according to claim 6, characterized in that: Minimize the KL divergence by the following formula: ； in, is the loss function; represents the KL divergence; importance sampling is used to approximate the expectation using the following formula: +const; in, is the number of sampling times; is the kth intention layer obtained by the sth sampling: const is a constant term; the parameters of variational inference are optimized using stochastic gradient descent using the following formula: ； in, For the The parameters of variational inference after iterative optimization; The parameters of the variational inference after optimization at the t+1th iteration; is the learning rate; is the gradient of the loss function with respect to the parameter θ.

8. The militarized target behavior analysis agent testing method according to claim 7, characterized in that: In step 4, the approximate posterior distribution is used to infer intent using the following formula: ； in, is the result of intention reasoning; then use the following formula to calculate the uncertainty of intention reasoning: ； in, Reasoning about uncertainty in intentions.

9. A military target behavior analysis agent testing device, characterized in that: include: The data acquisition part is used to obtain multi-source data of militarized targets, pre-process the data of each category in the multi-source data, align the timestamps and register the space, and then fuse them into an input tensor; Intent The model building part is used to extract tensor features of the input tensor. Based on the tensor features, a multi-layer intent hierarchy is defined. Each intent layer contains multiple possible intents, and conditional probabilities are used to represent the dependencies between the intent layers. The intention inference model construction part is used to define the likelihood function of each intent layer; a recurrent neural network is used to model the prior probability; Calculate posterior probabilities using Bayes' rule; The following process is iteratively executed to complete the training: using variational inference, an approximate posterior distribution is defined, and then the KL divergence is minimized; importance sampling is used to approximate the expectation, and stochastic gradient descent is used to optimize the variational inference parameters; when the number of iterations reaches the set number, step 4: the intention reasoning part is executed, which is used to perform intention reasoning using the approximate posterior distribution and calculate the uncertainty of intention reasoning; intention Model building part definition Layer intention structure, Layer Intent Layer possible intentions; initialize the intent vector for each layer: ； in, , is the index number; Indicates the The first layer of the intention layer intention; each is a discrete variable and can be possible values; Tensor features The nonlinear transformation is performed by the following formula to obtain the feature , to accommodate different levels of intent reasoning: ； in, represents element-wise multiplication, and They are the preset feature weight matrix and feature bias matrix respectively; Use conditional probability to represent the dependencies between intent layers: ； in, Expressing the highest level of intention, Indicates the lowest level of intention; For a given feature The joint probability distribution of the entire intent hierarchy when ; For a given When considering layer intentions and features, the conditional probability of the lowest layer intention; For a given When layering intent and features, Conditional probability of layer intention; The conditional probability of the top-level intent given the features.

Images