Multi-task recognition intelligent sensor network optimized deployment method and device

Abstract

This application relates to the field of sensor network management and discloses a method and apparatus for optimizing the deployment of a smart sensor network for multi-task identification. The method includes: obtaining the node embedding matrix of sensor nodes in the smart sensor network; determining the contribution weight of each sensor node to a monitoring task set based on residual analysis of the node embedding matrix in the feature space; determining the connection weights between multiple sensor nodes using a fuzzy perception probability coverage model; and solving the membership degree of each sensor node to multiple monitoring tasks based on the interval type-II fuzzy set theory according to the multiple connection weights, wherein the smart sensor network includes multiple sensor nodes; calculating the generalized entropy of each sensor node to the monitoring task set based on the contribution weights and membership degrees of the multiple sensor nodes; identifying the top K sensor nodes with the largest generalized entropy among the multiple sensor nodes; and activating the top K sensor nodes to complete the optimized deployment of the smart sensor network.

Description

Technical Field

[0001] This application relates to the field of sensor network management, and in particular to a method and apparatus for optimizing the deployment of intelligent sensor networks with multi-task recognition. Background Technology

[0002] A smart sensor network is a wireless network composed of a large number of distributed sensor nodes. These nodes monitor the sensing area, collect data, and transmit it to a processing center. To ensure monitoring efficiency, the initial deployment strategy of sensor networks tends to deploy a certain amount of redundant sensor nodes. However, large-scale deployment of sensor nodes faces severe resource constraints, including limited energy, communication bandwidth, and computing power. Therefore, it is necessary to manage and reconfigure the nodes of the smart sensor network, i.e., optimize the network deployment. In the research on the optimized deployment of smart sensor networks, how to scientifically select the optimal subset from the large-scale deployment of sensor nodes to maximize the overall identification efficiency of the network for monitoring tasks under limited resource constraints is a research problem with significant theoretical and practical value.

[0003] Existing methods for optimizing sensor network deployment have significant limitations: First, most studies focus on optimizing a single task or simply treat multiple tasks as independent problems, failing to adequately consider the collaborative nature of these tasks. Second, constrained by factors such as measurement noise, environmental interference, and equipment performance limitations, node identification results often exhibit ambiguity. Most existing methods are based on deterministic models, lacking effective modeling and quantification of this perceptual ambiguity.

[0004] There is still no effective solution to the problem of how to select the optimal set of sensor nodes from a large number of sensor nodes under the constraint of limited resources in order to maximize the overall recognition performance of the network in multi-task scenarios. Summary of the Invention

[0005] To address this, this application provides a method and apparatus for optimizing the deployment of intelligent sensor networks for multi-task recognition, thereby overcoming the problem in the prior art of how to select the optimal set of sensor nodes from a large number of sensor nodes under limited resource constraints in order to maximize the overall recognition efficiency of the network in multi-task scenarios.

[0006] To achieve the above objectives, according to a first aspect of the embodiments of this application, a method for optimizing the deployment of a smart sensor network for multi-task identification is provided, comprising: obtaining a node embedding matrix of sensor nodes in the smart sensor network; determining the contribution weight of each sensor node to a monitoring task set based on residual analysis of the node embedding matrix in the feature space; wherein the node embedding matrix includes multiple node embeddings, the multiple node embeddings being used to indicate the digital vectors of the sensor nodes for multiple monitoring tasks, the monitoring task set including the multiple monitoring tasks, and the multiple node embeddings corresponding one-to-one with the multiple monitoring tasks; determining the connection weights between the multiple sensor nodes through a fuzzy perception probability coverage model, and solving the membership degree of each sensor node to the multiple monitoring tasks based on the interval type II fuzzy set theory according to the multiple connection weights; calculating the generalized entropy of each sensor node to the monitoring task set based on the contribution weights and membership degrees of the multiple sensor nodes, and identifying the top K sensor nodes with the largest generalized entropy among the multiple sensor nodes, activating the top K sensor nodes to complete the optimized deployment of the smart sensor network, wherein K is a positive integer.

[0007] In an optional embodiment, the node embedding matrix of sensor nodes in the intelligent sensor network is obtained. Based on the residual analysis of the node embedding matrix in the feature space, the contribution weight of each sensor node to the monitoring task set is determined, including: obtaining the node embedding of the sensor node for different monitoring tasks based on a graph attention autoencoder, obtaining the multiple node embeddings, and concatenating the multiple node embeddings to obtain the node embedding matrix; standardizing the node embedding matrix and calculating the covariance matrix of the standardized node embedding matrix; performing eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues, and determining the contribution weight of each sensor node to the monitoring task set from the multiple eigenvalues. L largest eigenvalues ​​are determined, and a principal feature subspace is constructed based on the eigenvectors corresponding to the L largest eigenvalues. The number of principal components L is determined based on the cumulative contribution rate threshold, and L is a positive integer. A residual subspace projection matrix is ​​constructed based on the principal feature subspace, and the standardized embedding vector of each sensor node is projected onto the residual subspace through the residual subspace projection matrix to obtain the corresponding residual vector. The principal feature subspace and the residual subspace are orthogonal, and the standardized embedding vector is the row vector of the standardized node embedding matrix. The contribution weight of each sensor node is calculated based on the squared L2 norm of the residual vector.

[0008] In an optional embodiment, based on a graph attention autoencoder, the node embeddings of the sensor nodes for different monitoring tasks are obtained to obtain the plurality of node embeddings. This includes: for any target monitoring task among the plurality of monitoring tasks, determining the node connection relationship between the first sensor node and the second sensor node under the target monitoring task based on the distance between any first sensor node and a second sensor node among the plurality of sensor nodes and the effective monitoring radius of the second sensor node under the target monitoring task, and generating an adjacency matrix of the second sensor node under the target monitoring task according to the plurality of node connection relationships; processing the plurality of adjacency matrices and the plurality of feature matrices composed of node perception data of the sensor nodes through the graph attention autoencoder to obtain the plurality of node embeddings, wherein the plurality of adjacency matrices and the plurality of feature matrices correspond one-to-one with the plurality of monitoring tasks.

[0009] In an optional embodiment, the connection weights between multiple sensor nodes are determined using a fuzzy perception probability coverage model, and the membership degree of each sensor node to the multiple monitoring tasks is solved based on the multiple connection weights according to the interval type II fuzzy set theory. This includes: based on the fuzzy perception probability coverage model, for a third sensor node among the multiple sensor nodes, converting the node connection relationship between the third sensor node and multiple fourth sensor nodes under the multiple monitoring tasks into connection weights, wherein the node connection relationship is a binary relationship, the connection weight is a continuous probability value, and the connection weight is determined based on the distance between sensor nodes, path attenuation coefficient, path loss coefficient, and node noise immunity coefficient; based on the connection weights... The node embedding of the third sensor node is re-determined based on weights, resulting in multiple node embedding features of the third sensor node. An interval-type II fuzzy membership function set is constructed based on a Gaussian function, where the interval-type II fuzzy membership function set includes an upper bound membership function and a lower bound membership function, which is used to construct the uncertain trajectory region. Based on the multiple node embedding features, the upper and lower bound membership degrees of the third sensor node for the multiple monitoring tasks are calculated using the interval-type II fuzzy membership function set, where the upper and lower bound membership degrees include an upper bound membership degree and a lower bound membership degree. An interval-weighted average is then performed on the upper and lower bound membership degrees to obtain the membership degree of the third sensor node for the multiple monitoring tasks.

[0010] In an optional embodiment, calculating the generalized entropy of each sensor node with respect to the monitoring task set based on the contribution weights and membership degrees of the plurality of sensor nodes includes: for the fifth sensor node among the plurality of sensor nodes, calculating the fuzzy entropy of the fifth sensor node with respect to the plurality of monitoring tasks based on the membership degree of the fifth sensor node with respect to the plurality of monitoring tasks, wherein the fuzzy entropy is used to indicate the effective information content in the node sensing data of the fifth sensor node; summing the fuzzy entropies of the plurality of monitoring tasks to obtain the total fuzzy entropy of the fifth sensor node with respect to the monitoring task set; and multiplying the contribution weight of the fifth sensor node by the total fuzzy entropy to obtain the generalized entropy of the fifth sensor node.

[0011] In an optional embodiment, calculating the fuzzy entropy of the fifth sensor node for the plurality of monitoring tasks based on the membership degree of the fifth sensor node for the plurality of monitoring tasks includes: using the formula Calculate the fuzzy entropy, where, For sensor nodes For monitoring tasks fuzzy entropy, For sensor nodes For monitoring tasks The degree of membership.

[0012] In an optional embodiment, the method further includes: optimizing the function parameters of the interval type-2 fuzzy membership function set using an unsupervised learning framework, and using the weighted sum of spatial consistency loss and uncertainty quantization loss as the loss function of the interval type-2 fuzzy membership function set, wherein the spatial consistency loss is used to indicate the difference in membership degree of adjacent sensor nodes to the same monitoring task, the uncertainty quantization loss is used to indicate the fuzzy entropy of the sensor node, and the fuzzy entropy is used to indicate the effective information content of the node sensing data of the sensor node.

[0013] According to a second aspect of the embodiments of this application, a multi-task identification intelligent sensor network optimization deployment device is also provided, comprising: a weight quantization module, configured to acquire a node embedding matrix of sensor nodes in the intelligent sensor network, and determine the contribution weight of each sensor node to a monitoring task set based on residual analysis of the node embedding matrix in the feature space, wherein the node embedding matrix includes multiple node embeddings, the multiple node embeddings are used to indicate the digital vectors of the sensor nodes for multiple monitoring tasks, the monitoring task set includes the multiple monitoring tasks, and the multiple node embeddings correspond one-to-one with the multiple monitoring tasks; and a membership degree solving module, configured to... The connection weights among multiple sensor nodes are determined using a fuzzy perception probability coverage model, and the membership degree of each sensor node to the multiple monitoring tasks is solved based on the interval type II fuzzy set theory according to the multiple connection weights. The intelligent sensor network includes multiple sensor nodes. An optimization selection module is used to calculate the generalized entropy of each sensor node to the monitoring task set based on the contribution weights and membership degrees of the multiple sensor nodes, and to determine the top K sensor nodes with the largest generalized entropy among the multiple sensor nodes, and activate the top K sensor nodes to complete the optimized deployment of the intelligent sensor network, where K is a positive integer.

[0014] According to a third aspect of the embodiments of this application, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the intelligent sensor network optimization deployment method for multi-task identification as described in the first aspect through the computer program.

[0015] According to a fourth aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the intelligent sensor network optimization deployment method for multi-task identification as described in the first aspect.

[0016] This application transforms the complex node selection problem into a highly efficient global optimization problem through a three-level quantization and fusion mechanism. First, at the node layer, residual analysis of the node embedding matrix quantifies the deviation of each node from the network's mainstream pattern into its contribution weight, thereby identifying the uniqueness and discriminative value of its information. Second, at the task layer, a fuzzy perception probability coverage model and interval type-II fuzzy set theory are used to quantify real-world interference factors into the membership degree of each node to various tasks, characterizing the reliability and uncertainty of its perception. Finally, at the network layer, weights and membership degrees are fused into a unified generalized entropy index, which comprehensively represents the total information value provided by a subset of nodes. Based on this, the optimal deployment problem is simplified to a ranking problem of directly selecting the top K nodes with the largest generalized entropy, thus theoretically maximizing the overall recognition performance of the network in multi-task scenarios while strictly satisfying the resource constraint K. This solves the problem in related technologies of how to select the optimal set of sensor nodes from a large number of sensor nodes under limited resource constraints to maximize the overall recognition performance of the network in multi-task scenarios.

[0017] It should be understood that the description in the Summary Section is not intended to limit the key or essential features of the embodiments of this application, nor is it intended to restrict the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0018] The above and other features, advantages, and aspects of the embodiments of this application will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of the application. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0019] Figure 1 A flowchart illustrating an optional multi-task identification-based intelligent sensor network optimization deployment method according to an embodiment of this application;

[0020] Figure 2 This is an architecture diagram of an optional intelligent sensor network optimization deployment mechanism according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of the architecture of an optional node contribution weight inference model according to an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of the architecture of an optional node membership solution model according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of an optional node optimization selection mechanism according to an embodiment of this application.

[0024] Figure 6This is a structural block diagram of an optional multi-task identification intelligent sensor network optimization deployment device according to an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0027] To address the technical problems existing in related technologies, this embodiment provides a method for optimizing the deployment of intelligent sensor networks with multi-task identification. Figure 1 This is a flowchart of a multi-task identification-based intelligent sensor network optimization deployment method according to an embodiment of this application. The process includes the following steps:

[0028] Step S102: Obtain the node embedding matrix of sensor nodes in the intelligent sensor network. Based on the residual analysis of the node embedding matrix in the feature space, determine the contribution weight of each sensor node to the monitoring task set. The node embedding matrix includes multiple node embeddings, which are used to indicate the digital vectors of the sensor nodes for multiple monitoring tasks. The monitoring task set includes the multiple monitoring tasks, and the multiple node embeddings correspond one-to-one with the multiple monitoring tasks.

[0029] Step S104: Determine the connection weights between multiple sensor nodes using a fuzzy perception probability coverage model, and solve the membership degree of each sensor node to the multiple monitoring tasks based on the multiple connection weights according to the interval type II fuzzy set theory. The intelligent sensor network includes multiple sensor nodes.

[0030] Step S106: Calculate the generalized entropy of each sensor node for the monitoring task set based on the contribution weights and membership degrees of the multiple sensor nodes, and determine the top K sensor nodes with the largest generalized entropy among the multiple sensor nodes, and activate the top K sensor nodes to complete the optimized deployment of the intelligent sensor network, where K is a positive integer.

[0031] like Figure 2 As shown in the embodiments of this application, an unsupervised optimization deployment mechanism for intelligent sensor networks driven by generalized entropy and multi-task is proposed. The core of this mechanism lies in three-level quantization and fusion. First, sensor network features are processed in parallel at the node layer: the contribution weight inference model uses graph neural networks and residual subspace analysis to infer the contribution weight of each node to the task; simultaneously, the edge weights of the graph topology are redefined through a fuzzy perception probability coverage model to simulate interference from various factors in reality. The newly obtained node features and graph topology are fed into the membership degree solution model to calculate the membership degree of nodes to multiple tasks. Then, the membership degree and contribution weight are input into an optimization selection mechanism based on maximizing generalized entropy, where the membership degree is transformed into fuzzy entropy at the task layer and fused with the weights to form a unified generalized entropy index. Finally, the generalized entropy is optimized at the network layer according to resource constraints, directly outputting an optimal subset of sensor nodes, i.e., the set of activated nodes (nodes that are not activated are placed in dormancy), thereby maximizing the overall performance of the network in multi-task scenarios.

[0032] Through the above steps, a three-level quantization and fusion mechanism transforms the complex node selection problem into a highly efficient global optimization problem. First, at the node layer, residual analysis of the node embedding matrix quantifies the deviation of each node from the network's mainstream pattern into its contribution weight, identifying the uniqueness and discriminative value of its information. Second, at the task layer, fuzzy perception probability coverage models and interval type-II fuzzy set theory quantify real-world interference factors into the membership degrees of nodes to various tasks, characterizing the reliability and uncertainty of their perception. Finally, at the network layer, weights and membership degrees are fused into a unified generalized entropy index, which comprehensively represents the total information value provided by a subset of nodes. Based on this, the optimal deployment problem is simplified to a ranking problem of directly selecting the top K nodes with the largest generalized entropy, thus theoretically maximizing the overall recognition performance of the network in multi-task scenarios while strictly satisfying the resource constraint K. This solves the problem in related technologies of how to select the optimal set of sensor nodes from a large number of sensor nodes under limited resource constraints to maximize the overall recognition performance of the network in multi-task scenarios.

[0033] Optionally, based on a graph attention autoencoder, the node embeddings of the sensor nodes for different monitoring tasks are obtained to obtain the plurality of node embeddings, including: for any target monitoring task among the plurality of monitoring tasks, determining the node connection relationship between the first sensor node and the second sensor node under the target monitoring task based on the distance between any first sensor node and the second sensor node among the plurality of sensor nodes and the specific monitoring radius of the second sensor node under the target monitoring task, and generating an adjacency matrix of the second sensor node under the target monitoring task according to the plurality of node connection relationships; processing the plurality of adjacency matrices and the plurality of feature matrices composed of node perception data of the sensor nodes through the graph attention autoencoder to obtain the plurality of node embeddings, wherein the plurality of adjacency matrices and the plurality of feature matrices correspond one-to-one with the plurality of monitoring tasks.

[0034] Optionally, obtain the node embedding matrix of sensor nodes in the intelligent sensor network, and determine the contribution weight of each sensor node to the monitoring task set based on the residual analysis of the node embedding matrix in the feature space. This includes: obtaining the node embedding of the sensor node for different monitoring tasks based on a graph attention autoencoder, obtaining the multiple node embeddings, and concatenating the multiple node embeddings to obtain the node embedding matrix; standardizing the node embedding matrix and calculating the covariance matrix of the standardized node embedding matrix; performing eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues, and determining L from the multiple eigenvalues. L largest eigenvalues ​​are identified, and a principal feature subspace is constructed based on the eigenvectors corresponding to these L largest eigenvalues. The number of principal components L is determined based on a cumulative contribution rate threshold, and L is a positive integer. A residual subspace projection matrix is ​​constructed based on the principal feature subspace, and the standardized embedding vector of each sensor node is projected onto the residual subspace through the residual subspace projection matrix to obtain the corresponding residual vector. The principal feature subspace and the residual subspace are orthogonal, and the standardized embedding vector is the row vector of the standardized node embedding matrix. The contribution weight of each sensor node is calculated based on the squared L2 norm of the residual vector.

[0035] In this application embodiment, to solve the problem of node weight quantization, this application proposes a method for solving node contribution weights based on embedded residual features, the process of which is as follows: Figure 3 As shown.

[0036] When using graph neural networks to obtain node embedding matrices and infer subsequent node contribution weights, the primary task is to effectively construct the topological graph structure. As a crucial element of the graph model, the definition of edge sets fundamentally determines the paths of information propagation and aggregation within the network. In intelligent sensor networks, the collaborative sensing capability between nodes is primarily constrained by their relative spatial positions. Only when two sensor nodes are within each other's effective monitoring range can they collaboratively observe the same target, and only then will their measurement data possess spatial correlation and information fusion value. In particular, sensor nodes exhibit differences in their ability to identify different tasks, and this difference is directly reflected in the target monitoring radius. Based on this principle, a graph construction method constrained by sensor node distance thresholds and monitoring radius directly reflects the physical connectivity between nodes. The connection relationships between nodes are defined using a binary function as shown in Equation 1.

[0037] Formula 1;

[0038] in, Represents sensor nodes and The connection between them This indicates the distance between two sensor nodes. For sensors in specific tasks The target monitoring radius is below. Indicates in the task Next, node and If the distance between the nodes is within the effective monitoring radius, i.e., within the collaborative sensing range, it means that there is spatial correlation between the monitoring data of the two nodes, and vice versa. Ultimately, this yields task-oriented results. adjacency matrix .

[0039] Accurately quantifying the contribution weights of each node in a homogeneous sensor network to task recognition hinges on generating high-quality node embeddings that integrate node attributes with the network topology. To this end, this section utilizes a graph attention auto-encoder (GATE) composed of an encoder and decoder to obtain node embeddings.

[0040] Specifically, the encoder is constructed using a multi-layered graph attention network (GAT) stack. Each sensor node... The Layer Embedded Representation for:

[0041] Formula 2;

[0042] in, Represents a node The neighborhood group, It is a non-linear activation function. is a trainable weight matrix. The normalized attention coefficient is calculated as follows:

[0043] Formula 3;

[0044] in, Let be a learnable attention parameter vector, and || denote the vector concatenation operation. For a trainable parameter matrix, For nodes Embedded representation, For nodes A neighbor node Embedded representation, Nodes The embedding representation of the neighboring nodes, where the subscripts are... It is a general term representing a node. any of its neighboring nodes.

[0045] The decoder uses inner product decoding to reconstruct the original graph structure based on the learned node embeddings. Nodes and The probability of a connection existing between them It is predicted to be:

[0046] Formula 4;

[0047] in, and These are the nodes output by the encoder. and nodes The final node embedding. This is a transpose operation. The model is trained by minimizing the reconstruction error, and the loss function is... Defined as an adjacency matrix With reconstruction matrix Binary Cross Entropy (BCE):

[0048] Formula 5;

[0049] This represents the total number of nodes in the network.

[0050] After training, for each sensor node The specific embedding representations learned under different tasks are concatenated to form an enhanced node embedding vector that integrates multi-task semantics. . Represents a node The label, This represents the total embedding dimension after concatenation. Subsequently, the enhanced embedding vectors of all sensor nodes are aggregated to construct a node embedding matrix for multi-task fusion. This matrix not only integrates node attributes and local and global topological information, but also introduces task-related discriminative features, forming a comprehensive node representation with multi-task semantics.

[0051] The contribution weight inference method based on residual subspace analyzes the distribution characteristics of nodes embedded in the feature space and establishes an effective measurement mechanism to quantify the contribution of each node to task recognition.

[0052] First, Z-score normalization is required for the multi-dimensional node embedding vectors to eliminate differences in the dimensions of embedding features from different tasks, ensuring fairness and accuracy in subsequent analyses. Calculating the covariance matrix is ​​a crucial step in understanding the correlation structure between node embeddings. The normalized embedding matrix... The covariance matrix is ​​defined as:

[0053] Formula 6;

[0054] in, These are the normalization coefficients. The eigenvalue decomposition of this symmetric matrix. It can reveal the main directions of variation in the embedding space. Covariance matrix eigenvector matrix, It is a diagonal matrix whose diagonal elements are... The eigenvalues ​​correspond one-to-one with the eigenvectors. The number of principal components is determined using the cumulative contribution rate criterion, with a threshold set to determine the number of principal components to retain. This choice ensures that Zhang Cheng's main feature subspace can capture the most significant and consistent variation patterns in the embeddings of all nodes, which can be understood as "consensus information" of network-aware data, while also effectively reducing dimensionality.

[0055] Residual subspace projection matrix Based on The principal feature subspace mapping formed by the eigenvectors corresponding to the largest eigenvalues. Construct, defined as:

[0056] Formula 7;

[0057] It is an identity matrix. Projecting vectors onto the complement space, orthogonal to the principal feature subspace, i.e., the residual subspace, effectively separates the main variation patterns and residual variation components in the data. (Residual mapping of node embeddings) This is achieved through the following transformation:

[0058] Formula 8;

[0059] Mapped vector It includes components in the node embedding that are orthogonal to the main mutation pattern. The square of the L2 norm The differences and uniqueness of the node's behavioral pattern compared to the network's mainstream "consensus information" were quantified. This difference is not a manifestation of noise or unreliability, but rather characterizes the unique perceptual perspective and information content provided by the node. A higher value indicates a more unique behavioral pattern and higher discriminative value of the information provided; conversely, a lower value indicates a high degree of consistency between the node's behavioral pattern and the network's mainstream pattern, with significant redundancy in the information provided compared to the crucial components constituting the "consensus foundation" of the perceptual task. Based on spectral theory and the principle of residual subspace mapping, the node... Contribution weight It is proportional to the square of the norm, specifically defined as follows:

[0060] Formula 9;

[0061] This represents the total number of nodes in the network. Clearly, Satisfy normalization conditions .

[0062] Optionally, the connection weights among multiple sensor nodes are determined using a fuzzy perception probability coverage model, and the membership degree of each sensor node to the multiple monitoring tasks is solved based on the multiple connection weights according to the interval type-2 fuzzy set theory. This includes: based on the fuzzy perception probability coverage model, for a third sensor node among the multiple sensor nodes, converting the node connection relationship between the third sensor node and multiple fourth sensor nodes under the multiple monitoring tasks into connection weights, wherein the node connection relationship is a binary relationship, the connection weight is a continuous probability value, and the connection weight is determined based on the distance between sensor nodes, path attenuation coefficient, path loss coefficient, and node noise immunity coefficient; based on the connection weights... The node embedding of the third sensor node is redefined to obtain multiple node embedding features of the third sensor node. An interval-type II fuzzy membership function set is constructed based on a Gaussian function, wherein the interval-type II fuzzy membership function set includes an upper bound membership function and a lower bound membership function, and the interval-type II fuzzy membership function set is used to constitute an uncertain trajectory region. Based on the multiple node embedding features, the upper and lower bound membership degrees of the third sensor node for the multiple monitoring tasks are calculated using the interval-type II fuzzy membership function set, wherein the upper and lower bound membership degrees include an upper bound membership degree and a lower bound membership degree. An interval-weighted average is performed on the upper and lower bound membership degrees to obtain the membership degree of the third sensor node for the multiple monitoring tasks.

[0063] To accurately characterize the fuzziness of node perception, this application constructs a complete node membership degree solution model. This model aims to quantify the inherent uncertainties of nodes during perception (such as measurement noise and channel interference) into a computable membership degree value using interval type II fuzzy set theory. The specific implementation path is as follows: First, a graph model that better reflects the actual physical connection quality needs to be established as the basis for analysis; then, a fuzzy inference mechanism is designed on this graph model by constructing an interval type II fuzzy membership function set. The overall architecture of the node membership degree solution model proposed in this embodiment is as follows: Figure 4 As shown.

[0064] The above embodiments define node connection relationships. The value can be either 0 or 1, essentially an idealized model of the effective connection relationship between nodes. This definition is based on a fundamental assumption: as long as a connection is determined between two sensor nodes, it is assumed that there is a lossless and interference-free ideal channel between them, and the connection strength is simplified to a maximum value of 1. This modeling method does not consider the complex impact of sensor measurement performance and factors such as path loss, multipath fading, and environmental noise interference commonly found in actual wireless communication environments on connection quality and signal strength. To overcome the limitations of the above idealized model and more realistically reflect actual wireless communication scenarios, this embodiment introduces a fuzzy perception probabilistic coverage model to refine the modeling of edge weights and node perception uncertainties in the graph model, constructing a more physically meaningful and robust multi-dimensional spatial graph. This model measures the sensor's sensing capability probabilistically, and the edge weights... Defined as a continuous probability value reflecting the signal transmission quality, as shown in Formula 10:

[0065] Formula 10;

[0066] Where, constant This represents the node's noise immunity coefficient, used to characterize the noise immunity capability of a sensor node due to its hardware characteristics. (Exponential decay term) Simulated wireless signal strength as a function of propagation distance A physical phenomenon of increase and decrease. Among them, This is the path attenuation coefficient. These are environment-related path loss exponents, collectively characterizing the channel fading characteristics under specific environments. Weights Reflects the nodes and The probability of successful and effective data exchange between nodes. The closer the distance, the better the channel conditions, and the better the node performance, the higher the connection weight and the greater the reliability of collaborative sensing.

[0067] The ability of sensor nodes to identify tasks is ambiguous, which can be reflected by membership degrees. Therefore, A fuzzy set is defined, using Zadeh notation, written as ,

[0068] in, express right The membership degree, and satisfying Furthermore, to accurately characterize this membership degree, this section introduces the theory of Interval Type-2 Fuzzy Sets (IT2FS) and establishes a node-aware fuzzy modeling method based on learnable parameters.

[0069] Traditional fuzzy set methods have limitations in handling uncertainty, while interval type II fuzzy sets, by introducing the "footprint of uncertainty" (FOU) region, can more effectively express the fuzziness in the perception process. Given the excellent mathematical properties of the Gaussian function—continuous differentiability, smoothness, and symmetry—it is well-suited to describing the normal distribution of sensor measurements around the true value, and meets the applicability conditions of the central limit theorem in practical measurement scenarios. Therefore, this embodiment selects the Gaussian function as the basic function for constructing the upper and lower bounds of the interval type II fuzzy membership function set. By adjusting the parameters of the Gaussian function, different perceptual fuzzinesses can be flexibly adapted. The core structure of the constructed interval type II fuzzy Gaussian membership function is as follows:

[0070] The node embedding obtained after redefining the edge weights is (i.e., the node embedding features mentioned above), the global statistical features obtained through the graph neural network are: and Then, for the task The upper and lower bound membership functions are defined as follows:

[0071] Formula 11;

[0072] Formula 12;

[0073] in, and These are learnable parameters, satisfying This ensures the formation of an effective region of uncertainty. Through this design, each node can obtain a fuzzy metric adapted to its embedded features.

[0074] After obtaining the FOU, it needs to be "reduced" to a type-1, computable membership value. Here, the common interval weighted average method is used. Final node. For recognition tasks membership degree It can be represented as follows:

[0075] Formula 13;

[0076] In the formula, For the task The number of samples.

[0077] Optionally, the method further includes: optimizing the function parameters of the interval type II fuzzy membership function set using an unsupervised learning framework, and using the weighted sum of spatial consistency loss and uncertainty quantization loss as the loss function of the interval type II fuzzy membership function set, wherein the spatial consistency loss is used to indicate the difference in membership degree of adjacent sensor nodes to the same monitoring task, the uncertainty quantization loss is used to indicate the fuzzy entropy of the sensor node, and the fuzzy entropy is used to indicate the effective information content of the node sensing data of the sensor node.

[0078] To achieve automatic optimization of fuzzy parameters, this embodiment also designs an unsupervised learning framework based on a dual loss function. Since topologically adjacent nodes should have similar fuzzy characteristics, a spatial consistency loss is constructed. Quantifying losses due to uncertainty It can promote the clear differentiation of ambiguity. The two are defined as follows:

[0079] Formula 14;

[0080] Formula 15;

[0081] in, Let be the set of edges in a graph neural network. For spatial consistency loss, Quantify the loss due to uncertainty.

[0082] The total loss is the weighted combination of the two:

[0083] Formula 16;

[0084] hyperparameters and These are two regularization coefficients.

[0085] Optionally, the generalized entropy of each sensor node with respect to the monitoring task set is calculated based on the contribution weights and membership degrees of the multiple sensor nodes, including: for the fifth sensor node among the multiple sensor nodes, calculating the fuzzy entropy of the fifth sensor node with respect to the multiple monitoring tasks based on the membership degree of the fifth sensor node with respect to the multiple monitoring tasks, wherein the fuzzy entropy is used to indicate the effective information content in the node sensing data of the fifth sensor node; summing the fuzzy entropies of the multiple monitoring tasks to obtain the total fuzzy entropy of the fifth sensor node with respect to the monitoring task set; and multiplying the contribution weight of the fifth sensor node by the total fuzzy entropy to obtain the generalized entropy of the fifth sensor node.

[0086] In an exemplary embodiment, calculating the fuzzy entropy of the fifth sensor node for the plurality of monitoring tasks based on the membership degree of the fifth sensor node for the plurality of monitoring tasks includes: using the formula Calculate the fuzzy entropy, where, For sensor nodes For monitoring tasks fuzzy entropy, For sensor nodes For monitoring tasks The degree of membership.

[0087] In this embodiment, a comprehensive metric model is established to evaluate the overall performance of the network in task recognition. To this end, this embodiment constructs a task-oriented generalized entropy metric model for isomorphic sensor networks. This mechanism aims to establish a quantitative index system for the task recognition performance of sensor networks by integrating information theory and fuzzy logic, and thereby optimize the selection of nodes in the network. The specific process is as follows: Figure 5 As shown.

[0088] First, based on fuzzy logic theory and information science theory, nodes are defined. For the task The fuzzy entropy is calculated using the following formula:

[0089] Formula 17;

[0090] This refers to the effective information content in the node's perceived data. The higher the entropy value, the greater the information gain that the node's observation data pattern brings to the collaborative perception system, and the greater its potential contribution to building a global recognition model.

[0091] Secondly, nodes For containing A set of subtasks The fuzzy entropy is obtained by summing the fuzzy entropies of all tasks, that is:

[0092] Formula 18;

[0093] in, For task set The number of neutron missions.

[0094] Furthermore, define task-oriented The generalized entropy of a smart sensor network is:

[0095] Formula 19;

[0096] in, Enable an identifier for the node. This indicates that the node is enabled; otherwise, it will hibernate.

[0097] Generalized entropy Essentially, it is the weighted fuzzy entropy sum of a subset of enabled nodes. The larger the value, the richer the task-related information contained in the network, and the higher the recognition efficiency. Therefore, the problem of optimizing the deployment of sensor networks can be expressed as:

[0098] ,

[0099] Formula 20;

[0100] Due to the objective function Decision variables A linear function, and the constraints are from the initial state of the sensor network. Select from sensor nodes Given the cardinality constraint, this optimization problem has a separable structure, and the optimal solution can be obtained through a simple sorting mechanism. Since the terms... and It is a constant, and For binary variables, the objective function is under constraints. The maximum value is selected using the TOP-K method. The largest value Implemented by individual nodes. The selected node index is the overall score for each node, sorted from highest to lowest.

[0101] Formula 21;

[0102] Nodes form the optimal subset The corresponding node enable identifier is set to:

[0103] Formula 22;

[0104] The generalized entropy corresponding to the optimal subset is:

[0105] Formula 23.

[0106] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, 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 is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0107] Embodiments of this application also provide a multi-task identification-based intelligent sensor network optimization deployment device, such as... Figure 6 As shown, the device includes:

[0108] The weight quantization module 62 is used to obtain the node embedding matrix of sensor nodes in the intelligent sensor network, and to determine the contribution weight of each sensor node to the monitoring task set based on the residual analysis of the node embedding matrix in the feature space. The node embedding matrix includes multiple node embeddings, which are used to indicate the digital vector of the sensor node for multiple monitoring tasks. The monitoring task set includes the multiple monitoring tasks, and the multiple node embeddings correspond one-to-one with the multiple monitoring tasks.

[0109] Membership degree solving module 64 is used to determine the connection weights between multiple sensor nodes through a fuzzy perception probability coverage model, and to solve the membership degree of each sensor node to the multiple monitoring tasks based on the multiple connection weights according to the interval type II fuzzy set theory, wherein the intelligent sensor network includes multiple sensor nodes.

[0110] The optimization selection module 66 is used to calculate the generalized entropy of each sensor node for the monitoring task set based on the contribution weights and membership degrees of the multiple sensor nodes, and to determine the top K sensor nodes with the largest generalized entropy among the multiple sensor nodes, and activate the top K sensor nodes to complete the optimized deployment of the smart sensor network, where K is a positive integer.

[0111] The aforementioned device transforms the complex node selection problem into a highly efficient global optimization problem through a three-level quantization and fusion mechanism. First, at the node layer, residual analysis of the node embedding matrix quantifies the deviation of each node from the mainstream network pattern into its contribution weight, thereby identifying the uniqueness and discriminative value of its information. Second, at the task layer, a fuzzy perception probability coverage model and interval type-II fuzzy set theory are used to quantify real-world interference factors into the membership degree of each node to various tasks, characterizing the reliability and uncertainty of its perception. Finally, at the network layer, weights and membership degrees are fused into a unified generalized entropy index, which comprehensively represents the total information value provided by the node subset. Based on this, the optimal deployment problem is simplified to a ranking problem of directly selecting the top K nodes with the largest generalized entropy, thus theoretically maximizing the overall recognition performance of the network in multi-task scenarios while strictly satisfying the resource constraint K. This solves the problem in related technologies of how to select the optimal set of sensor nodes from a large number of sensor nodes under limited resource constraints to maximize the overall recognition performance of the network in multi-task scenarios.

[0112] Optionally, the aforementioned weight quantization module 62 is further configured to: obtain the node embeddings of the sensor nodes for different monitoring tasks based on a graph attention autoencoder, obtain the multiple node embeddings, and concatenate the multiple node embeddings to obtain the node embedding matrix; perform standardization processing on the node embedding matrix and calculate the covariance matrix of the standardized node embedding matrix; perform eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues, determine L largest eigenvalues ​​from the multiple eigenvalues, and construct a principal feature subspace based on the eigenvectors corresponding to the L largest eigenvalues, wherein the number of principal components L is determined based on the cumulative contribution rate threshold, and L is a positive integer; construct a residual subspace projection matrix based on the principal feature subspace, and project the standardized embedding vector of each sensor node onto the residual subspace through the residual subspace projection matrix to obtain the corresponding residual vector, wherein the principal feature subspace and the residual subspace are orthogonal, and the standardized embedding vector is the row vector of the standardized node embedding matrix; and calculate the contribution weight of each sensor node based on the squared L2 norm of the residual vector.

[0113] Optionally, the aforementioned weight quantization module 62 is further configured to, for any target monitoring task among the plurality of monitoring tasks, determine the node connection relationship between the first sensor node and the second sensor node under the target monitoring task based on the distance between any first sensor node and the second sensor node among the plurality of sensor nodes and the specific monitoring radius of the second sensor node under the target monitoring task, and generate an adjacency matrix of the second sensor node under the target monitoring task according to the plurality of node connection relationships; process the plurality of adjacency matrices and the plurality of feature matrices composed of node perception data of sensor nodes through the graph attention autoencoder to obtain the plurality of node embeddings, wherein the plurality of adjacency matrices and the plurality of feature matrices correspond one-to-one with the plurality of monitoring tasks.

[0114] Optionally, the membership calculation module 64 is further configured to, based on the fuzzy perception probability coverage model, convert the node connection relationship between the third sensor node and multiple fourth sensor nodes under the multiple monitoring tasks into connection weights for the third sensor node among the multiple sensor nodes. The node connection relationship is a binary relationship, and the connection weights are continuous probability values. The connection weights are determined based on the distance between sensor nodes, path attenuation coefficient, path loss coefficient, and node noise immunity coefficient. Based on the connection weights, the node embedding of the third sensor node is re-determined to obtain the multiple membership values ​​of the third sensor node. The system embeds features of each node; constructs a set of interval type II fuzzy membership functions based on Gaussian functions, wherein the set of interval type II fuzzy membership functions includes an upper bound membership function and a lower bound membership function, and is used to constitute an uncertain trajectory region; based on the multiple node embedding features, the system calculates the upper and lower bound membership degrees of the third sensor node for the multiple monitoring tasks through the set of interval type II fuzzy membership functions, wherein the upper and lower bound membership degrees include an upper bound membership degree and a lower bound membership degree; and performs an interval weighted average on the upper and lower bound membership degrees to obtain the membership degree of the third sensor node for the multiple monitoring tasks.

[0115] Optionally, the optimization selection module 66 is further configured to, for the fifth sensor node among the plurality of sensor nodes, calculate the fuzzy entropy of the fifth sensor node for the plurality of monitoring tasks based on the membership degree of the fifth sensor node for the plurality of monitoring tasks, wherein the fuzzy entropy is used to indicate the effective information content in the node sensing data of the fifth sensor node; sum the fuzzy entropies of the plurality of monitoring tasks to obtain the total fuzzy entropy of the fifth sensor node for the monitoring task set; and multiply the contribution weight of the fifth sensor node by the total fuzzy entropy to obtain the generalized entropy of the fifth sensor node.

[0116] Optionally, the aforementioned optimization selection module 66 is also used to select the option via formula. Calculate the fuzzy entropy, where, For sensor nodes For monitoring tasks fuzzy entropy, For sensor nodes For monitoring tasks The degree of membership.

[0117] Optionally, the membership degree solving module 64 is further configured to optimize the function parameters of the interval type II fuzzy membership function set using an unsupervised learning framework, and to use the weighted sum of spatial consistency loss and uncertainty quantization loss as the loss function of the interval type II fuzzy membership function set, wherein the spatial consistency loss is used to indicate the difference in membership degree of adjacent sensor nodes to the same monitoring task, the uncertainty quantization loss is used to indicate the fuzzy entropy of the sensor node, and the fuzzy entropy is used to indicate the effective information content of the node sensing data of the sensor node.

[0118] Embodiments of this application also provide a storage medium including a stored program, wherein the program, when executed, performs any of the methods described above. Optionally, in this embodiment, the storage medium may be configured to store program code for performing the steps of the methods described above.

[0119] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0120] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0121] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0122] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium storing the computer program product, wherein the computer program, when executed by a processor, implements the steps of the methods described in various embodiments of this application.

[0123] Optionally, in this embodiment, the computer program described above can be configured to implement the steps in any of the above method embodiments when executed by a processor.

[0124] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0125] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device in software or hardware. Terms such as "first," "second," etc., are used only for distinguishing descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0126] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. A method for optimizing the deployment of a multi-task identification intelligent sensor network, characterized in that, include: Obtain the node embedding matrix of sensor nodes in the intelligent sensor network. Based on the residual analysis of the node embedding matrix in the feature space, determine the contribution weight of each sensor node to the monitoring task set. The node embedding matrix includes multiple node embeddings, which are used to indicate the digital vectors of the sensor nodes for multiple monitoring tasks. The monitoring task set includes the multiple monitoring tasks, and the multiple node embeddings correspond one-to-one with the multiple monitoring tasks. The connection weights among multiple sensor nodes are determined using a fuzzy perception probabilistic coverage model. Based on interval type-II fuzzy set theory, the membership degree of each sensor node to the multiple monitoring tasks is solved according to these connection weights. The intelligent sensor network includes multiple sensor nodes. Based on the fuzzy perception probabilistic coverage model, for a third sensor node among the multiple sensor nodes, the node connection relationship between the third sensor node and multiple fourth sensor nodes under the multiple monitoring tasks is converted into connection weights. These node connection relationships are binary relationships, and the connection weights are continuous probability values. The connection weights are determined based on the distance between sensor nodes, path attenuation coefficient, path loss coefficient, and node noise immunity coefficient. The generalized entropy of each sensor node with respect to the monitoring task set is calculated based on the contribution weights and membership degrees of the multiple sensor nodes. The top K sensor nodes with the largest generalized entropy are then identified among the multiple sensor nodes, and the top K sensor nodes are activated to complete the optimized deployment of the intelligent sensor network, where K is a positive integer.

2. The method according to claim 1, characterized in that, Obtain the node embedding matrix of sensor nodes in the intelligent sensor network. Based on the residual analysis of the node embedding matrix in the feature space, determine the contribution weight of each sensor node to the monitoring task set, including: Based on a graph attention autoencoder, the node embeddings of the sensor nodes for different monitoring tasks are obtained, resulting in multiple node embeddings. These multiple node embeddings are then concatenated to obtain the node embedding matrix. The node embedding matrix is ​​standardized, and the covariance matrix of the standardized node embedding matrix is ​​calculated. The covariance matrix is ​​decomposed into eigenvalues ​​to obtain multiple eigenvalues. L largest eigenvalues ​​are determined from the multiple eigenvalues, and a principal feature subspace is constructed based on the eigenvectors corresponding to the L largest eigenvalues. The number of principal components L is determined based on the cumulative contribution rate threshold, and L is a positive integer. Based on the principal feature subspace, a residual subspace projection matrix is ​​constructed, and the standardized embedding vector of each sensor node is projected onto the residual subspace through the residual subspace projection matrix to obtain the corresponding residual vector. The principal feature subspace is orthogonal to the residual subspace, and the standardized embedding vector is the row vector of the standardized node embedding matrix. The contribution weight of each sensor node is calculated based on the squared L2 norm of the residual vector.

3. The method according to claim 2, characterized in that, Based on a graph attention autoencoder, the node embeddings of the sensor nodes for different monitoring tasks are obtained, resulting in the multiple node embeddings, including: For any target monitoring task among the plurality of monitoring tasks, the node connection relationship between the first sensor node and the second sensor node under the target monitoring task is determined based on the distance between any first sensor node and the second sensor node among the plurality of sensor nodes and the effective monitoring radius of the second sensor node under the target monitoring task, and an adjacency matrix of the second sensor node under the target monitoring task is generated according to the plurality of node connection relationships. The graph attention autoencoder processes multiple adjacency matrices and multiple feature matrices composed of node perception data from sensor nodes to obtain multiple node embeddings, wherein the multiple adjacency matrices and multiple feature matrices correspond one-to-one with the multiple monitoring tasks.

4. The method according to claim 1, characterized in that, The connection weights among multiple sensor nodes are determined using a fuzzy perception probability coverage model, and the membership degree of each sensor node to the multiple monitoring tasks is solved based on the interval type II fuzzy set theory according to the multiple connection weights, including: Based on the connection weights, the node embedding of the third sensor node is re-determined, resulting in multiple node embedding features of the third sensor node. An interval type II fuzzy membership function set is constructed based on the Gaussian function, wherein the interval type II fuzzy membership function set includes an upper bound membership function and a lower bound membership function, and the interval type II fuzzy membership function set is used to construct an uncertain trajectory region; Based on the embedding features of the multiple nodes, the upper and lower bound membership degrees of the third sensor node for the multiple monitoring tasks are calculated through the interval type II fuzzy membership function set, wherein the upper and lower bound membership degrees include upper bound membership degree and lower bound membership degree. The membership degree of the third sensor node to the multiple monitoring tasks is obtained by performing an interval weighted average of the upper and lower bound membership degrees.

5. The method according to claim 1, characterized in that, The generalized entropy of each sensor node with respect to the monitoring task set is calculated based on the contribution weights and membership degrees of multiple sensor nodes, including: For the fifth sensor node among the plurality of sensor nodes, the fuzzy entropy of the fifth sensor node to the plurality of monitoring tasks is calculated based on the membership degree of the fifth sensor node to the plurality of monitoring tasks, wherein the fuzzy entropy is used to indicate the effective information content in the node sensing data of the fifth sensor node; The fuzzy entropy of the multiple monitoring tasks is summed to obtain the total fuzzy entropy of the fifth sensor node for the monitoring task set; The contribution weight of the fifth sensor node is multiplied by the total fuzzy entropy to obtain the generalized entropy of the fifth sensor node.

6. The method according to claim 5, characterized in that, Calculating the fuzzy entropy of the fifth sensor node for the multiple monitoring tasks based on the membership degree of the fifth sensor node to the multiple monitoring tasks includes: Through formula Calculate the fuzzy entropy, where, For sensor nodes For monitoring tasks fuzzy entropy, For sensor nodes For monitoring tasks The degree of membership.

7. The method according to claim 4, characterized in that, The method further includes: An unsupervised learning framework is used to optimize the function parameters of the interval type II fuzzy membership function set, and the weighted sum of spatial consistency loss and uncertainty quantization loss is used as the loss function of the interval type II fuzzy membership function set. The spatial consistency loss is used to indicate the difference in membership degree of adjacent sensor nodes to the same monitoring task, and the uncertainty quantization loss is used to indicate the fuzzy entropy of the sensor node. The fuzzy entropy is used to indicate the effective information content of the node sensing data of the sensor node.

8. A multi-task recognition intelligent sensor network optimization deployment device, characterized in that, include: The weight quantization module is used to obtain the node embedding matrix of sensor nodes in the intelligent sensor network. Based on the residual analysis of the node embedding matrix in the feature space, the contribution weight of each sensor node to the monitoring task set is determined. The node embedding matrix includes multiple node embeddings, which are used to indicate the digital vectors of the sensor nodes for multiple monitoring tasks. The monitoring task set includes the multiple monitoring tasks, and the multiple node embeddings correspond one-to-one with the multiple monitoring tasks. The membership calculation module is used to determine the connection weights between multiple sensor nodes through a fuzzy perception probability coverage model, and to solve the membership degree of each sensor node to the multiple monitoring tasks based on the interval type II fuzzy set theory according to the multiple connection weights. The intelligent sensor network includes multiple sensor nodes. Based on the fuzzy perception probability coverage model, for the third sensor node among the multiple sensor nodes, the node connection relationship between the third sensor node and multiple fourth sensor nodes under the multiple monitoring tasks is converted into connection weights. The node connection relationship is a binary relationship, and the connection weight is a continuous probability value. The connection weight is determined based on the distance between sensor nodes, path attenuation coefficient, path loss coefficient, and node noise resistance coefficient. The optimization selection module is used to calculate the generalized entropy of each sensor node with respect to the monitoring task set based on the contribution weights and membership degrees of multiple sensor nodes, and to determine the top K sensor nodes with the largest generalized entropy among the multiple sensor nodes, and activate the top K sensor nodes to complete the optimized deployment of the smart sensor network, where K is a positive integer.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images