Iot perception device reliability evaluation method based on graph neural network and multi-task learning

Abstract

The application discloses a kind of based on graph neural network and multi-task learning's thing connection perception device reliability evaluation method, specifically including the following steps: data collection and pre-processing;Multi-level model design, evaluate multi-level model reliability, multi-level model design includes sensor level design, component level design, board card level design and whole machine level design;Multi-level network fusion;Through particle swarm optimization multi-level model parameter, improve reliability evaluation performance.The application improves the accuracy of reliability evaluation between each level by constructing multi-level full-quantity mapping relationship, and accurately predicts the overall failure trend of the equipment, thereby enhancing the stability of the overall system, especially in complex electromagnetic environment and environmental stress, the reliability of the equipment is more accurately evaluated;For the operation and maintenance of digital converter station is essential, can identify potential failure in advance, optimize maintenance plan, avoid the outage and economic loss caused by sudden failure.

Description

Technical Field

[0001] This application relates to the field of Internet of Things (IoT) technology in power systems, and in particular to a reliability assessment method for IoT sensing devices based on graph neural networks and multi-task learning. Background Technology

[0002] The application of IoT sensing device reliability assessment in digital converter stations is of great significance. As a key node in the new power system, the stability and reliability of the IoT sensing network in the digital converter station directly affects the efficiency of intelligent management and equipment fault detection in the power system. Due to the complex electromagnetic environment in which the converter station is located, and the influence of high and low temperatures, salt spray corrosion, alternating damp heat, and other operating conditions, IoT sensing devices often face interference from many uncertain factors. These factors can affect the communication efficiency and data accuracy of the equipment, thereby reducing the system's operating efficiency and service quality. Therefore, relying on advanced IoT, edge computing, and artificial intelligence technologies, establishing a set of equipment status assessment methods for complex environments can not only detect potential equipment faults in advance and improve the accuracy of fault prediction, but also ensure the stable operation of the digital converter station through real-time monitoring and intelligent analysis of equipment status, thus promoting the digital transformation and intelligent upgrading of equipment management.

[0003] Existing reliability assessment methods for IoT sensing devices mostly focus on single-level assessments or analyses based on expert experience and rules, making it difficult to comprehensively consider the interdependencies between different levels of the device and the impact of complex environmental factors. Single-level reliability assessment methods typically focus only on the reliability of a single level of the device (such as the entire device, components, or modules), ignoring the mutual influence and dependencies between different levels. For example, degradation of sensing materials may lead to performance degradation of components, ultimately affecting the reliability of the entire device. While empirical and rule-driven methods, such as Fault Tree Analysis (FTA) and Failure Mode and Effects Analysis (FMEA), can provide intuitive failure mode analysis for devices, they heavily rely on expert experience and assumptions and cannot handle novel or unknown failure modes. Furthermore, as device technology advances and the operating environment changes, existing empirical rules often become inadequate for new failure scenarios, thus reducing the accuracy and flexibility of the assessment. In addition, Fault Tree Analysis and Failure Mode and Effects Analysis typically lack in-depth consideration of the multi-level coupling relationships of the device, making it difficult to provide accurate predictions in complex operating environments. Summary of the Invention

[0004] The technical problem this invention aims to solve is to provide a reliability assessment method for IoT sensing devices based on graph neural networks and multi-task learning. This assessment method establishes multi-level full-data mapping relationships, combines graph neural networks and multi-task learning to comprehensively assess the reliability of the device, and accurately models the interdependencies and influences between different levels. Through the fusion and optimization of multi-source heterogeneous data, it can effectively cope with interference from complex environments, provide a more accurate and comprehensive reliability assessment, and possess adaptive optimization capabilities. It has significant advantages in equipment management and fault prediction, greatly improving the stability and intelligence level of the system.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following means:

[0006] The reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning specifically includes the following steps:

[0007] S1. Data collection and preprocessing;

[0008] S2. Multi-level model design and evaluation of the reliability of the multi-level model. The multi-level model design includes sensor-level design, component-level design, board-level design and overall system-level design.

[0009] S3, multi-level network convergence;

[0010] S4. Improve reliability assessment performance by optimizing multi-level model parameters through particle swarm optimization.

[0011] Furthermore, the specific method for step S1 is as follows:

[0012] Relevant data is collected from sensors, components, boards, and the entire device of the Internet of Things sensing system. The data includes fault data, operating status data, and environmental data. The data is cleaned and preprocessed to remove outliers and fill in missing data to ensure data consistency and integrity.

[0013] Furthermore, the specific method for sensor hierarchy design in step S2 is as follows:

[0014] The reliability assessment of sensor materials employs a convolutional neural network (CNN) model. The model's objective is to output the reliability D1 of the sensor material. The input consists of multidimensional features of the sensor material, including its physical properties and environmental data. These features constitute the input data. N represents the number of sample data, M represents the feature dimension of each sample data, and the reliability of the sensor material is obtained through a convolutional neural network, with the specific expression as follows:

[0015] D1 = f1(X1)

[0016] Where f1(·) represents the convolutional neural network model, and D1 represents the reliability of the sensor material, which is a value between [0,1].

[0017] The computation process of the convolutional neural network model includes the following steps:

[0018] 1) Convolutional layer

[0019] Convolutional layers use convolutional weights K represents the size of the convolutional kernel, and the expression for the output of the convolutional layer is as follows:

[0020] Z1 = X1·W1 + b1

[0021] Where Z1 represents the output of the convolutional layer, and b1 represents the bias term;

[0022] 2) Activation function

[0023] The output of the convolutional layer is activated by the ReLU function, the expression of which is as follows:

[0024] A1 = ReLU(Z1) = max(0, Z1)

[0025] The ReLU activation function turns all negative values ​​to 0 to increase the nonlinearity of the network;

[0026] 3) Pooling layer

[0027] The pooling layer selects the maximum value in a local region. The pooling operation uses a 2×2 window with a stride of 2. The expression output by the pooling layer is as follows:

[0028] P1 = MaxPool(A1)

[0029] Where P1 represents the output of the pooling layer, which reduces the data dimension and improves computational efficiency;

[0030] 4) Fully connected layer

[0031] The pooled data is flattened and input into the fully connected layer. The weights of the fully connected layer are W2, and the expression for the output of the fully connected layer is as follows:

[0032] F1 = P1·W2 + b2

[0033] Where F1 represents the output of the fully connected layer, and b2 represents the bias term;

[0034] 5) Output layer

[0035] The output layer of the convolutional neural network model is processed by the Sigmoid activation function to obtain the reliability D1, the specific expression of which is as follows:

[0036]

[0037] Where σ represents the Sigmoid activation function.

[0038] Furthermore, the specific method for component-level design in step S2 is as follows:

[0039] The component level is modeled using a graph neural network (GNN). The model objective is to output the reliability D2 of the component level. The input is the input data X2. The reliability of the component level is obtained through the graph neural network, and the specific expression is as follows:

[0040] D² = g²(X²)

[0041] Where g2 represents the function obtained through the GNN model;

[0042] In the graph, nodes represent components, and edges represent the dependencies between components. The graph is represented as G = (V, E), where V = {v1, v2, ..., v...} n} represents a node set, E = {e1, e2, ..., e m} represents the edge set, where each node v i eigenvectors Let represent the feature vector of the l-th layer node; the graph convolutional layer updates the feature vector of each node by a weighted average of the neighbor node information, as shown in the following expression:

[0043]

[0044] in, Represents node v i The set of neighboring nodes, W( l ) represents the learning parameters of the l-th layer, b( l ) represents the bias term; σ represents the activation function;

[0045] The update process iterates, and the final node feature vector The specific expression for predicting the reliability D2 of components is as follows:

[0046]

[0047] Here, φ(·) represents a subsequent neural network layer used to map node features to reliability scores.

[0048] Furthermore, the specific method for the board hierarchy design in step S2 is as follows:

[0049] The board-level reliability is jointly trained using a multi-task learning (MTL) model. The model objective is to output the board-level reliability D3, with the input data X3. The board-level reliability is obtained through the MTL model, and the specific expression is as follows:

[0050]

[0051] Among them, f k (·) represents the k-th task model using the MTL model, θ k This represents the model parameters for the k-th task;

[0052] The multi-task set T = {T1, T2, ..., T} at the board level K}, each task T k For a given objective function, the model outputs multi-task prediction results, as shown in the following expression:

[0053]

[0054] In the MTL model, parameters are divided into shared parameters and task-specific parameters. The shared parameters are used to capture the commonalities of all tasks, while the task-specific parameters are used to capture the unique characteristics of each task. The output expression for each task is as follows:

[0055] f k =W shared X3+W k X3

[0056] Among them, W shared W represents shared parameters. k Indicates task-specific parameters;

[0057] The goal of the MTL model is to minimize the weighted sum of the multi-task loss functions, as shown in the following expression:

[0058]

[0059] Among them, L k Let λ represent the loss function for the k-th task. k Represents task T k The weights of Y k This indicates the actual label.

[0060] Furthermore, the specific method for the overall machine-level design in step S2 is as follows:

[0061] At the system-wide level, a reinforcement learning (RL) model is used to consider the system's reliability D4 under different operating conditions and environments. The model is trained using deep Q-learning to explore the optimal reliability assessment under various operating conditions. The model objective is to output the system-wide reliability D4. The reinforcement learning model is a Markov decision process, denoted as a quintuple (S, A, P, R, γ).

[0062] The state space S represents the overall operating state of the machine, s t ∈S represents the running state at time t;

[0063] Action space A represents machine-level actions, including adjusting workload, modifying maintenance intervals, and changing operating modes, etc. t ∈A represents the action at time t;

[0064] The state transition probability P(s'|s,a) represents the probability distribution of the overall machine operating state changing from s to s' when action a is performed;

[0065] Reward function R(s) t ,a t This represents the reward function obtained from performing an action, which is related to the device's health status.

[0066] R(s t ,a t )=-C failure -α r ·C maintenance +βr·D4

[0067] Among them, C failure C represents the cost (high value) resulting from equipment failure. maintenance α represents the cost of maintenance actions. r β r These represent the weight parameters of the reward function, used to balance faults, maintenance costs, and reliability;

[0068] The discount factor γ represents the weight of the influence on controlling future rewards, and its range is 0 < γ < 1;

[0069] The goal of reinforcement learning models is to find an optimal policy π. * The expression for (a|s) maximizes the long-term cumulative reward, and is as follows:

[0070]

[0071] The state-action value function Q(s,a) is expressed as follows:

[0072]

[0073] in, Let Q(s,a) represent the expected reliability after taking action a in state s, a' represent the next action that can be taken, and s' represent the system state after taking the current action a.

[0074] The overall system-level reliability D4 is optimized through deep Q-learning, and the state-action value function Q(s,a) is updated as follows:

[0075]

[0076] Where α represents the learning rate;

[0077] After training, select the optimal action a given state s. * The expression is as follows:

[0078]

[0079] Execute the optimal action a * The corresponding system-wide reliability is obtained using the following expression:

[0080] D4=Q(s,a * ).

[0081] Furthermore, the specific method for step S3 is as follows:

[0082] The reliability models of the sensor level, component level, board level and whole machine level in step S2 are fused to construct a multi-level network. The fusion of the multi-level network establishes the reliability mapping relationship between the levels, so that the overall reliability of the device is not only based on the individual evaluation results of each level, but also takes into account the interaction between levels, thereby improving the overall evaluation effect.

[0083] A multilayer perceptron is used to fuse the reliability levels D1, D2, D3, and D4 from different levels. The process expression is as follows:

[0084] h′1=σ(W1[D1,D2,D3,D4]+b1)

[0085] h′2=σ(W2h′1+b2)

[0086] D′=σ(W3h′2+b3)

[0087] Where D' represents the overall reliability as the final evaluation result, W1, W2, and W3 represent the weight matrix, and b1, b2, and b3 represent the bias terms.

[0088] Furthermore, the specific method for step S4 is as follows:

[0089] Particle Swarm Optimization (PSO) is used to optimize the parameters of multi-layer networks. PSO helps search for optimal network weights and biases, minimizing the final evaluation error. The specific expression is as follows:

[0090] min L(D,D actual )

[0091] Where L represents the loss function using the mean squared error, and D actual Indicates actual reliability;

[0092] Particle swarm optimization (PSO) searches for the optimal solution by adjusting the position and velocity of particles. The PSO process is as follows:

[0093] Let the number of particles be P, and each particle represent the parameter set Θ = {W1, W2, W3, b1, b2, b3} of the model; calculate the loss value L for each particle under the current parameters, and update the particle velocity and position expressions as follows:

[0094] v i,t+1 =ωv i，t +c1r1(p best,i -Θ i )+c2r2(g best -Θ i )

[0095] Θ i，t+1 =Θ i，t +v it+1

[0096] Among them, v i,t Let represent the velocity of particle i in generation t, ω represent the inertia weight used to balance local and global search, c1 and c2 represent learning factors that control the influence of individual and global optimal positions, r1 and r2 represent random numbers to ensure the particle's search capability, and p best,i Let g represent the best solution to particle i so far. best Indicates the globally optimal solution;

[0097] If the loss function L converges or reaches the maximum number of iterations, then the optimization stops and the optimal model parameters are obtained.

[0098] Compared with the prior art, the present invention has the following beneficial effects:

[0099] This invention improves the accuracy of reliability assessment between different levels by constructing a multi-level full mapping relationship and accurately predicts the overall failure trend of the equipment, thereby enhancing the stability of the overall system. Especially under complex electromagnetic environments and environmental stress, the equipment reliability is more accurately assessed. This is crucial for the operation and maintenance management of digital converter stations, enabling the early identification of potential faults, optimization of maintenance plans, and avoidance of downtime and economic losses caused by sudden failures.

[0100] By employing techniques such as deep learning, graph neural networks, multi-task learning, and reinforcement learning for modeling, the system is able to extract effective features from large amounts of data, perform accurate predictions and analyses, and possess stronger learning and generalization capabilities. Through the combination of particle swarm optimization and reinforcement learning, not only are the parameters of each level of the model optimized, but the evaluation strategy is also dynamically adjusted according to changes in the device status, thereby enhancing the system's adaptability.

[0101] It can effectively process various heterogeneous data from different levels and sources, and through comprehensive analysis, it improves the efficiency of data utilization and reduces the complexity of data preprocessing. Attached Figure Description

[0102] Figure 1 This is a flowchart illustrating the reliability assessment method of the present invention. Detailed Implementation

[0103] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples.

[0104] This invention evaluates the reliability of IoT sensing devices under complex environmental conditions through a multi-level model. Its core is to divide the reliability of the device into four levels: sensing materials, components, circuit boards, and the whole machine. By using graph neural networks and multi-task learning techniques, a reliability mapping relationship between each level is established to achieve an accurate assessment of the overall reliability of the device. At the same time, particle swarm optimization is combined to optimize the model parameters to improve the accuracy and robustness of the evaluation model.

[0105] like Figure 1 As shown, the reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning specifically includes the following steps:

[0106] S1. Data collection and preprocessing, the specific methods are as follows:

[0107] Relevant data is collected from sensors, components, boards, and the entire device of the Internet of Things sensing system. The data includes fault data, operating status data, and environmental data. The data is cleaned and preprocessed to remove outliers and fill in missing data to ensure data consistency and integrity.

[0108] S2. Multi-level model design and reliability evaluation. The multi-level model design includes sensor-level design, component-level design, board-level design, and system-level design. The specific method for sensor-level design is as follows:

[0109] The reliability assessment of sensor materials employs a convolutional neural network (CNN) model. The model's objective is to output the reliability D1 of the sensor material. The input consists of multidimensional features of the sensor material, including physical properties (such as hardness, conductivity, temperature, and pressure) and environmental data (such as humidity, temperature, and radiation). These features constitute the input data. N represents the number of sample data, M represents the feature dimension of each sample data, and the reliability of the sensor material is obtained through a convolutional neural network, with the specific expression as follows:

[0110] D1 = f1(X1)

[0111] Where f1(·) represents the convolutional neural network model, and D1 represents the reliability of the sensor material, which is a value between [0,1], representing the reliability of the material under the current environment;

[0112] The computation process of the convolutional neural network model includes the following steps:

[0113] 1) Convolutional layer

[0114] Convolutional layers use convolutional weights K represents the size of the convolutional kernel, and the expression for the output of the convolutional layer is as follows:

[0115] Z1 = X1·W1 + b1

[0116] Where Z1 represents the output of the convolutional layer, and b1 represents the bias term;

[0117] 2) Activation function

[0118] The output of the convolutional layer is activated by the ReLU function, the expression of which is as follows:

[0119] A1 = ReLU(Z1) = max(0, Z1)

[0120] The ReLU activation function turns all negative values ​​to 0 to increase the nonlinearity of the network;

[0121] 3) Pooling layer

[0122] The pooling layer selects the maximum value in a local region. The pooling operation uses a 2×2 window with a stride of 2. The expression output by the pooling layer is as follows:

[0123] P1 = MaxPool(A1)

[0124] Where P1 represents the output of the pooling layer, which reduces the data dimension and improves computational efficiency;

[0125] 4) Fully connected layer

[0126] The pooled data is flattened and input into the fully connected layer. The weights of the fully connected layer are W2, and the expression for the output of the fully connected layer is as follows:

[0127] F1 = P1·W2 + b2

[0128] Where F1 represents the output of the fully connected layer, and b2 represents the bias term;

[0129] 5) Output layer

[0130] The output layer of the convolutional neural network model is processed by the Sigmoid activation function to obtain the reliability D1, the specific expression of which is as follows:

[0131]

[0132] Where σ represents the Sigmoid activation function.

[0133] The specific methods for component-level design are as follows:

[0134] Graph Neural Networks (GNNs) are deep learning models that can effectively process graph-structured data and are suitable for describing the interdependencies between component levels. In the component level of IoT sensing devices, components typically form a graph structure through electrical connections, signal dependencies, and other relationships. These relationships have a significant impact on the reliability of the components. The component level is modeled using a Graph Neural Network (GNN). The model objective is to output the reliability D2 of the component level, with the input data X2. The reliability of the component level is obtained through the Graph Neural Network, and the specific expression is as follows:

[0135] D² = g²(X²)

[0136] Where g2 represents the function obtained through the GNN model;

[0137] The core of a graph neural network is to update the representation of each node through information propagation between nodes and edges. The update of a node is a weighted summation process based on information from its neighboring nodes. In a graph structure, the nodes are components, and the edges represent the dependencies between these components. The graph is represented as G = (V, E), where V = {v1, v2, ..., v...}. n} represents a node set, E = {e1, e2, ..., e m} represents the edge set, where each node v i eigenvectors Let represent the feature vector of the l-th layer node; the graph convolutional layer updates the feature vector of each node by a weighted average of the neighbor node information, as shown in the following expression:

[0138]

[0139] in, Represents node v i The set of neighboring nodes, W (l) b represents the learning parameters of the l-th layer. (l) σ represents the bias term; σ represents the activation function.

[0140] The update process iterates, and the final node feature vector The specific expression for predicting the reliability D2 of components is as follows:

[0141]

[0142] Here, φ(·) represents a subsequent neural network layer used to map node features to reliability scores.

[0143] The specific methods for board hierarchy design are as follows:

[0144] The board-level reliability is jointly trained using a multi-task learning (MTL) model. The model objective is to output the board-level reliability D3, with the input data X3. The board-level reliability is obtained through the MTL model, and the specific expression is as follows:

[0145]

[0146] Among them, f k (·) represents the k-th task model using the MTL model, θ k This represents the model parameters for the k-th task;

[0147] The multi-task set T = {T1, T2, ..., T} at the board level K}, each task T k For a given objective function, the model outputs multi-task prediction results, as shown in the following expression:

[0148]

[0149] In the MTL model, parameters are divided into shared parameters and task-specific parameters. The shared parameters are used to capture the commonalities of all tasks, while the task-specific parameters are used to capture the unique characteristics of each task. The output expression for each task is as follows:

[0150] f k =W shared X3+W k X3

[0151] Among them, W shared W represents shared parameters. k Indicates task-specific parameters;

[0152] The goal of the MTL model is to minimize the weighted sum of the multi-task loss functions, as shown in the following expression:

[0153]

[0154] Among them, L k Let λ represent the loss function for the k-th task. k Represents task T k The weights (set to 1 / k), Y k Representing the true label; by jointly optimizing the loss function of multiple tasks, MTL effectively improves the accuracy of board-level reliability prediction.

[0155] The specific methods for overall machine hierarchical design are as follows:

[0156] At the system-wide level, a reinforcement learning (RL) model is used to consider the system's reliability D4 under different operating conditions and environments. The model is trained using deep Q-learning to explore the optimal reliability assessment under various operating conditions. The model objective is to output the system-wide reliability D4. The reinforcement learning model is a Markov decision process, denoted as a quintuple (S, A, P, R, γ).

[0157] The state space S represents the overall operating state of the machine, s t ∈S represents the running state at time t;

[0158] Action space A represents machine-level actions, including adjusting workload, modifying maintenance intervals, and changing operating modes, etc. t ∈A represents the action at time t;

[0159] The state transition probability P(s'|s,a) represents the probability distribution of the overall machine operating state changing from s to s' when action a is performed;

[0160] Reward function R(s) t ,a t This represents the reward function obtained from performing an action, which is related to the device's health status.

[0161] R(s t ,a t )=-C failure -α r ·C maintenance +β r ·D4

[0162] Among them, C failure C represents the cost (high value) resulting from equipment failure. maintenance α represents the cost of maintenance actions. r β r These represent the weight parameters of the reward function, used to balance faults, maintenance costs, and reliability;

[0163] The discount factor γ represents the weight of the influence on controlling future rewards, and its range is 0 < γ < 1;

[0164] The goal of reinforcement learning models is to find an optimal policy π. * The expression for (a|s) maximizes the long-term cumulative reward, and is as follows:

[0165]

[0166] The state-action value function Q(s,a) is expressed as follows:

[0167]

[0168] in, Let Q(s,a) represent the expected reliability after taking action a in state s, a' represent the next action that can be taken, and s' represent the system state after taking the current action a.

[0169] The overall system-level reliability D4 is optimized through deep Q-learning, and the state-action value function Q(s,a) is updated as follows:

[0170]

[0171] Where α represents the learning rate;

[0172] After training, select the optimal action a given state s. * The expression is as follows:

[0173]

[0174] Execute the optimal action a * The corresponding system-wide reliability is obtained using the following expression:

[0175] D4=Q(s,a * ).

[0176] S3, multi-level network fusion, the specific methods are as follows:

[0177] The reliability models of the sensor level, component level, board level and whole machine level in step S2 are fused to construct a multi-level network. The fusion of the multi-level network establishes the reliability mapping relationship between the levels, so that the overall reliability of the device is not only based on the individual evaluation results of each level, but also takes into account the interaction between levels, thereby improving the overall evaluation effect.

[0178] A multilayer perceptron is used to fuse the reliability levels D1, D2, D3, and D4 from different levels. The process expression is as follows:

[0179] h′1=σ(W1[D1,D2,D3,D4]+b1)

[0180] h′2=σ(W2h′1+b2)

[0181] D′=σ(W3h′2+b3)

[0182] Where D' represents the overall reliability as the final evaluation result, W1, W2, and W3 represent the weight matrix, and b1, b2, and b3 represent the bias terms.

[0183] S4. Improve reliability assessment performance by optimizing multi-level model parameters through particle swarm optimization. The specific method is as follows:

[0184] Particle Swarm Optimization (PSO) is used to optimize the parameters of multi-layer networks. PSO helps search for optimal network weights and biases, minimizing the final evaluation error. The specific expression is as follows:

[0185] min L(D′,D actual )

[0186] Where L represents the loss function using the mean squared error, and D actual Indicates actual reliability;

[0187] Particle swarm optimization (PSO) searches for the optimal solution by adjusting the position and velocity of particles. The PSO process is as follows:

[0188] Let the number of particles be P, and each particle represent the parameter set Θ = {W1, W2, W3, b1, b2, b3} of the model; calculate the loss value L for each particle under the current parameters, and update the particle velocity and position expressions as follows:

[0189] v i,t+1 =ωv i,t +c1r1(p best,i -Θ i )+c2r2(g best -Θ i )

[0190] Θ i,t+1 =Θ i,t +v i,t+1

[0191] Among them, v i,t Let represent the velocity of particle i in generation t, ω represent the inertia weight used to balance local and global search, c1 and c2 represent learning factors that control the influence of individual and global optimal positions, r1 and r2 represent random numbers to ensure the particle's search capability, and p best,i Let g represent the best solution to particle i so far. best Indicates the globally optimal solution;

[0192] If the loss function L converges or reaches the maximum number of iterations, then the optimization stops and the optimal model parameters are obtained.

[0193] The above description is merely an embodiment of the present invention. It should be reiterated that those skilled in the art can make several improvements to the present invention without departing from the principle of the present invention, and these improvements are also included within the scope of protection of the claims of the present invention.

Claims

1. A reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning, characterized in that, Specifically, the following steps are included: S1. Data collection and preprocessing; S2. Multi-level model design and evaluation of the reliability of the multi-level model. The multi-level model design includes sensor-level design, component-level design, board-level design and overall system-level design. S3, multi-level network convergence; S4. Improve reliability assessment performance by optimizing multi-level model parameters through particle swarm optimization; The specific method for the overall machine-level design in step S2 is as follows: At the system-wide level, a reinforcement learning model is used. The model objective is to output the system-wide reliability D4. The reinforcement learning model is a Markov decision process, denoted as a quintuple (S, A, P, R, γ). The state space S represents the overall operating state of the machine, s t ∈S represents the running state at time t; Action space A represents machine-level actions, including adjusting workload, modifying maintenance intervals, and changing operating modes. t ∈A represents the action at time t; The state transition probability P(s′|s,a) represents the probability distribution of the overall machine operating state changing from s to s′ when action a is performed; Reward function R(s) t ,a t () represents the reward function obtained by performing the action: R(s t ,a t )=-C failure -a r ·C maintenance +b r ·D4 Among them, C failure C represents the cost incurred due to equipment failure. maintenance α represents the cost of maintenance actions. r β r These represent the weight parameters of the reward function, used to balance faults, maintenance costs, and reliability; The discount factor γ represents the weight of the influence on controlling future rewards, and its range is 0 < γ < 1; The goal of reinforcement learning models is to find an optimal policy π. * The expression for (a|s) maximizes the long-term cumulative reward, and is as follows: The state-action value function Q(s,a) is expressed as follows: in, Let Q(s,a) represent the expected reliability after taking action a in state s, a' represent the next action that can be taken, and s' represent the system state after taking the current action a. The overall system-level reliability D4 is optimized through deep Q-learning, and the state-action value function Q(s,a) is updated as follows: Where α represents the learning rate; After training, select the optimal action a given state s. * The expression is as follows: Execute the optimal action a * The corresponding system-wide reliability is obtained using the following expression: D4=Q(s,a * ); The specific method for step S3 is as follows: The reliability models of the sensor level, component level, board level and whole machine level in step S2 are fused to construct a multi-level network, and the multi-level network fusion establishes the reliability mapping relationship between the levels. A multilayer perceptron is used to fuse the reliability levels D1, D2, D3, and D4 from different levels. The process expression is as follows: h′1=σ(W1[D1,D2,D3,D4]+b1) h′2=σ(W2h′1+b2 D′=σ(W3h′2+n3) Where D' represents the overall reliability as the final evaluation result, W1, W2, and W3 represent the weight matrix, and b1, b2, and b3 represent the bias terms.

2. The reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning according to claim 1, characterized in that, The specific method for step S1 is as follows: Relevant data is collected from sensors, components, boards, and the entire device of the Internet of Things sensing system. The data includes fault data, operating status data, and environmental data. The data is cleaned and preprocessed to remove outliers and fill in missing data to ensure data consistency and integrity.

3. The reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning according to claim 1, characterized in that, The specific method for sensor hierarchy design in step S2 is as follows: The reliability assessment of sensor materials employs a convolutional neural network model. The model objective is to output the reliability D1 of the sensor material. The input consists of multidimensional features of the sensor material, including physical properties and environmental data. These features constitute the input data X1∈ N represents the number of sample data, M represents the feature dimension of each sample data, and the reliability of the sensor material is obtained through a convolutional neural network, with the specific expression as follows: D1 = f1(X1) Where f1(·) represents the convolutional neural network model, and D1 represents the reliability of the sensor material; The computation process of the convolutional neural network model includes the following steps: 1) Convolutional layer Convolutional layers use convolutional weights K represents the size of the convolutional kernel, and the expression for the output of the convolutional layer is as follows: Z1 = X1·W1 + b1 Where Z1 represents the output of the convolutional layer, and b1 represents the bias term; 2) Activation function The output of the convolutional layer is activated by the ReLU function, the expression of which is as follows: A1 = ReLU(Z1) = max(0, Z1) The ReLU activation function turns all negative values ​​to 0 to increase the nonlinearity of the network; 3) Pooling layer The pooling layer selects the maximum value in a local region. The pooling operation uses a 2×2 window with a stride of 2. The expression output by the pooling layer is as follows: P1 = MaxPool(A1) Where P1 represents the output of the pooling layer, which reduces the data dimension; 4) Fully connected layer The pooled data is flattened and input into the fully connected layer. The weights of the fully connected layer are W2, and the expression for the output of the fully connected layer is as follows: F1 = P1·W2 + b2 Where F1 represents the output of the fully connected layer, and b2 represents the bias term; 5) Output layer The output layer of the convolutional neural network model is processed by the Sigmoid activation function to obtain the reliability D1, the specific expression of which is as follows: Where σ represents the Sigmoid activation function.

4. The reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning according to claim 1, characterized in that, The specific method for component-level design in step S2 is as follows: The component level is modeled using a graph neural network. The model objective is to output the reliability D2 of the component level. The input is the input data X2. The reliability of the component level is obtained through the graph neural network, and the specific expression is as follows: D² = g²(X²) Where g2 represents the function obtained through the GNN model; In the graph, nodes represent components, and edges represent the dependencies between components. The graph is represented as G = (V, E), where V = {v1, v2, ..., v...} n } represents a node set, E = {e1, e2, ..., e m } represents the edge set, where each node v i eigenvectors Let represent the feature vector of the l-th layer node; the graph convolutional layer updates the feature vector of each node by a weighted average of the neighbor node information, as shown in the following expression: in, Represents node v i The set of neighboring nodes, W (l) b represents the learning parameters of the l-th layer. (l) σ represents the bias term; σ represents the activation function. The update process is iterative, and the final node feature vector is used to predict the component reliability D2, the specific expression of which is as follows: Here, φ(·) represents a subsequent neural network layer used to map node features to reliability scores.

5. The reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning according to claim 1, characterized in that, The specific method for board hierarchy design in step S2 is as follows: The board-level reliability is jointly trained using a multi-task learning model. The model objective is to output the board-level reliability D3, with the input data X3. The board-level reliability is obtained through the multi-task learning model, and the specific expression is as follows: Among them, f k (·) represents the k-th task model using the MTL model, θ k This represents the model parameters for the k-th task; The multi-task set T = {T1, T2, ..., T} at the board level K }, each task T k For a given objective function, the model outputs multi-task prediction results, as shown in the following expression: In the MTL model, parameters are divided into shared parameters and task-specific parameters. The shared parameters are used to capture the commonalities of all tasks, while the task-specific parameters are used to capture the unique characteristics of each task. The output expression for each task is as follows: f k =W shared X3+W k X3 Among them, W shared W represents a shared parameter. k Indicates task-specific parameters; The goal of the MTL model is to minimize the weighted sum of the multi-task loss functions, as shown in the following expression: Among them, L k Let λ represent the loss function for the k-th task. k Indicates task T k The weights of Y k This indicates the actual label.

6. The reliability evaluation method for IoT sensing devices based on graph neural networks and multi-task learning according to claim 1, characterized in that, The specific method for step S4 is as follows: The parameters of a multi-level network are optimized using particle swarm optimization to minimize the final evaluation error. The specific expression is as follows: minL(D,D actual ) Where L represents the loss function using the mean squared error, and D actual Indicates actual reliability; Particle swarm optimization (PSO) searches for the optimal solution by adjusting the position and velocity of particles. The PSO process is as follows: Let the number of particles be P, and each particle represent the parameter set Θ = {W1, W2, W3, b1, b2, b3} of the model; calculate the loss value L for each particle under the current parameters, and update the particle velocity and position expressions as follows: v i,t+1 =ωv i,t +c1r1(p best,i -I i )+c2r2(g best -I i ) I i,t+1 =Θ i,t +v i,t+1 Among them, v i,t Let represent the velocity of particle i in generation t, ω represent the inertia weight used to balance local and global search, c1 and c2 represent learning factors that control the influence of individual and global optimal positions, r1 and r2 represent random numbers to ensure the particle's search capability, and p best,i Let g represent the best solution to particle i to date. best Indicates the globally optimal solution; If the loss function L converges or reaches the maximum number of iterations, then the optimization stops and the optimal model parameters are obtained.

Images