Industrial Internet of Things cloud platform data analysis method

Abstract

The invention relates to the technical field of data processing, and discloses an industrial Internet of Things cloud platform data analysis method, which realizes distributed model training of a multi-node joint model and industrial Internet of Things data mining by adopting a distributed federated learning model training architecture so as to break through the problem of data resource islands of a traditional centralized training mode. Aggregating to obtain more effective industrial Internet of Things data; meanwhile, the distributed federated learning architecture can prevent leakage of local industrial Internet of Things data of each node, local use of the industrial Internet of Things data and local training of the model, thereby effectively ensuring the security of the industrial Internet of Things data.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a method for analyzing data on an industrial Internet of Things cloud platform. Background Art

[0002] With the rapid development of industrial Internet of Things technology, more and more industrial devices and sensors are connected to the network, generating a large amount of data. This data contains rich information and is of great value for optimizing production processes, improving equipment efficiency, and promptly responding to equipment failures. However, this data is often distributed in different geographical locations and organizations. Traditional centralized data processing methods not only face the risk of data privacy leakage but also have difficulty achieving efficient data sharing and utilization due to resource limitations and regulatory restrictions. In addition, existing data augmentation techniques often cannot fully reflect the complexity of the actual operating environment, resulting in problems with insufficient generalization ability in model training.

[0003] Chinese invention patent CN202011083175.5 proposes an online monitoring system for the status of power equipment based on the Internet of Things, including a data acquisition and aggregation platform, and a data comprehensive processing platform, a data analysis, deployment and application platform, and a device status warning platform that are sequentially connected to the data acquisition and aggregation platform. This system uses technologies and means such as IoT terminals, data acquisition, intelligent perception, standard protocol conversion, information fusion, intelligent diagnosis, and monitoring and warning to achieve the integration and analysis of IoT terminal data in a substation. It uses methods such as information interaction, data sharing, data analysis, and standardized interfaces to achieve information interaction with other external systems. It uses hierarchical, graded, and customized information acquisition means to achieve functions such as IoT terminal data acquisition, standard protocol conversion, and data analysis. Through this system, it can quickly achieve the acquisition and analysis of IoT terminal data in the station, as well as functions such as device status perception and monitoring, device status evaluation, and device status anomaly warning.

[0004] Chinese invention patent CN202410439099.9 proposes an industrial Internet gateway and a device health analysis model based on topological data analysis, which relates to the field of the Internet of Things; this Internet gateway is a high-performance edge computing gateway with communication protocols of 4G / 5G, Ethernet, and WiFi; a low-power data acquisition gateway with communication protocols of 4G / 5G, NB-loT, and WiFi; a general wireless data transmission gateway with a communication protocol of 4G Cat.1; a high-performance OPCUA acquisition gateway with communication protocols of 4G / 5G, WiFi, and Ethernet. The working logic of this model includes the following steps: data acquisition and preprocessing; data acquisition, where the gateway actually acquires the operating parameters of the device. The present invention has innovation, practicability, and operability, and is of great significance for improving equipment operation efficiency, reducing the occurrence of failures, and lowering maintenance costs.

[0005] Chinese invention patent CN202410550168.3 proposes a method and system for analyzing data of a palm-sized ultrasonic device based on the Internet of Things. The method includes initializing the device information of the ultrasonic device terminal; uploading the operation data to the cloud; continuously monitoring the connection status between the ultrasonic device terminal and the cloud and detecting the connection status between the ultrasonic device and the control terminal, recording connection abnormal information and reporting it to the cloud; analyzing the working status of the ultrasonic device, predicting the damage probability and remaining service life of the ultrasonic device; and generating a statistical report after the cloud receives the operation data and abnormal information. Through Internet of Things technology, the present invention realizes real-time data collection, transmission and processing of devices, comprehensively counts information such as the scanning time, device status, damage situation, scanning abnormalities, and link abnormalities of devices, helps device manufacturers to understand the usage of devices scattered around the world in real time, and optimizes the device management and maintenance processes.

[0006] Although the above technical solutions solve some technical problems, the following problems still need to be further solved: 1. In traditional centralized model training, all data needs to be centrally stored and processed, which easily leads to data privacy leakage problems. Moreover, due to the centralized storage of data, the demand for resources is high, which limits the scalability of the model.

[0007] 2. Traditional data augmentation techniques often suffer from insufficient sample diversity and cannot fully simulate the complex data distribution in the real world, resulting in weak generalization ability of the model.

[0008] 3. When training a neural network using a standard optimization algorithm, it is easy to encounter problems of gradient vanishing or explosion, which affects the training stability and final performance of the model.

[0009] 4. Traditional graph neural networks usually perform information aggregation only in a single direction, failing to fully utilize the global and local data relationships, which reduces the ability of the model to process complex data structures. Summary of the Invention

[0010] To solve the deficiencies in the prior art, the present invention provides a method for analyzing data of an industrial Internet of Things cloud platform. By adopting a distributed federated learning model training architecture, it realizes distributed model training of a multi-node joint model and data mining of industrial Internet of Things, so as to break the data resource island problem of the traditional centralized training mode and aggregate and obtain more effective industrial Internet of Things data. At the same time, the distributed federated learning architecture can prevent the leakage of industrial Internet of Things data of each node. The industrial Internet of Things data is used locally and the model is trained locally, effectively ensuring the security of industrial Internet of Things data.

[0011] To achieve the above object, the present invention is realized through the following technical solutions: A method for analyzing data of an industrial Internet of Things cloud platform includes the following steps: Obtain the industrial Internet of Things data of each node; Input the industrial Internet of Things (IIoT) data of each node into a pre-constructed distributed federated learning model, and upload the trained model parameters to the central server; Use the generative adversarial network algorithm based on quantum state iteration to augment the obtained IIoT data; Input the augmented IIoT data into a neural network model for data mining; Input the mined data into a pre-constructed classifier model for classification prediction to obtain classified data; the classifier model is a graph neural network model based on a bidirectional graph matching strategy; Determine the data analysis result of the IIoT cloud platform based on the classified data.

[0012] Furthermore, during the model training process of distributed federated learning, the federated averaging algorithm is used as the model aggregation algorithm, and the federated averaging algorithm realizes the collaborative training of local models through multiple global iterations.

[0013] Furthermore, the specific method of using the generative adversarial network algorithm based on quantum state iteration is as follows: Randomly initialize the parameters of the generator and discriminator; Update the generator. The generator receives random noise as input and generates new IIoT data samples; Update the discriminator. The discriminator receives the generated samples from the generator and the samples in the real IIoT dataset, evaluates the two types of samples, and outputs the probability that they are real samples. When the probability that the generated samples are judged as real samples by the discriminator is within the interval [45%, 55%], it indicates that the generated samples and the real samples are similar, and the perturbation factor of the discriminator is increased to further optimize the quality of the generated samples; During the adversarial training process, the generator and discriminator are alternately updated in the way of training 100 rounds each; Use the gradient descent method based on momentum to update the weight parameters of the generator and discriminator, and use the gradient descent method to update the bias parameters of the generator and discriminator.

[0014] Furthermore, the neural network model uses a neural network algorithm based on self-adjusting learning rate.

[0015] Furthermore, the specific method of the graph neural network model based on the bidirectional graph matching strategy is as follows: Construct an initial graph according to the data features after feature extraction; For the feature vectors of each sample after feature extraction, convert the features into low-dimensional vectors through the embedding layer; Through the bidirectional graph matching strategy, perform bidirectional information aggregation from two directions: from global to local and from local to global; Fuse the aggregated features and use a conditional random field layer to process the fused node features.

[0016] Furthermore, the generator adopts a dynamic adjustment mechanism for the probability amplitude of the quantum state during the generation process.

[0017] Furthermore, the industrial Internet of Things data of each node is sourced from the industrial Internet of Things cloud platform, covering data from sensors, machinery, and industrial control systems on the production line.

[0018] Furthermore, the initial graph is constructed based on the similarity between devices, and the construction method is expressed as: ; where represents the edge weight between the -th node and the -th node in the graph; is the bandwidth parameter of the Gaussian kernel, used to control the sensitivity of the edge weight; is the -th sample in the dataset after feature extraction, i.e., the -th node in the graph; is the -th sample in the dataset after feature extraction, i.e., the -th node in the graph; represents the exponential function with the natural constant as the base; is the L2 norm.

[0019] Furthermore, the conditional random field layer considers the dependencies between nodes and optimizes the classification decision-making process by maximizing the posterior probability to achieve the final classification of Internet of Things devices.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In the data analysis task of the industrial Internet of Things cloud platform, the present invention adopts a distributed federated learning framework. By locally training the model on each node and aggregating and updating the model parameters to the central server, the problems of data privacy leakage and resource limitations in the traditional centralized training mode are solved.

[0021] 2. In the data analysis task of the industrial Internet of Things cloud platform, a generative adversarial network with dynamic feature enhancement of quantum states is adopted to optimize the feature distribution of data samples through quantum probability amplitude adjustment without revealing the original data, solving the problems of insufficient sample quantity and insufficient sample diversity in the dataset.

[0022] 3. In the data analysis task of the industrial Internet of Things cloud platform, the self - adjusting learning rate optimization algorithm is adopted to optimize the parameter training of the neural network. By utilizing the sensitivity of the self - adaptive adjustment of the learning rate, the response ability of the model to the initial conditions of the parameters is enhanced, and the problems of gradient disappearance or gradient explosion that may occur during the model training process are solved.

[0023] 4. In the data analysis task of the industrial Internet of Things cloud platform, by adopting a two - way graph matching strategy in the graph neural network to aggregate information from both the global - to - local and local - to - global directions, the problem that one - way information aggregation cannot comprehensively capture the complex relationships between nodes is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] FIG. Figure 1 is the flowchart of the present invention; FIG. Figure 2 is the model training architecture of the distributed federated learning of the present invention; FIG. Figure 3 is the ablation experiment result graph of the two - way aggregation mechanism of the present invention, where (a) is the comparison experiment result graph of the classification accuracy of five models, and (b) is the comparison experiment result graph of the training time of five models; FIG. Figure 4 is the schematic diagram of the influence of industrial scenario complexity on performance; FIG. Figure 5 is the comparison experiment graph of noise robustness. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by this application.

[0026] The present invention proposes a data analysis method for an industrial Internet of Things cloud platform. The industrial Internet of Things data analysis task requires a large amount of effective training data for feature extraction, analysis, and learning. The present invention realizes the distributed model training of a multi - node joint model and the data mining of the industrial Internet of Things by adopting the model training architecture of distributed federated learning, so as to break the problem of data resource islands in the traditional centralized training mode and aggregate and obtain more effective industrial Internet of Things data. At the same time, the distributed federated learning architecture can prevent the leakage of industrial Internet of Things data of each node. The industrial Internet of Things data is used locally, and the model is trained locally, effectively ensuring the security of industrial Internet of Things data.

[0027] For the convenience of understanding this embodiment, first, a method for constructing a device fault diagnosis model based on small samples disclosed in the embodiments of the present invention will be introduced in detail. Figure 1The flowchart of a data analysis method for an industrial Internet of Things cloud platform disclosed in an embodiment of the present invention is shown as Figure 1 shown. The data analysis method for the industrial Internet of Things cloud platform includes the following steps: S1. Obtain the industrial Internet of Things data of each node; The industrial Internet of Things data collection of each node in the present invention is sourced from the industrial Internet of Things cloud platform, covering data from sensors, machine equipment, and industrial control systems on the production line. The collected industrial Internet of Things data reflects the operating status of the machine, environmental parameters, and key indicators in the production process. The industrial Internet of Things data is stored in a structured JSON format. In one embodiment, the attributes of the industrial Internet of Things data include: temperature, reflecting the temperature value of the equipment or environment; humidity, the humidity level of the environment; pressure, the pressure index during equipment operation; vibration level, the degree of vibration generated during machine operation; energy consumption, the energy consumption of the equipment during operation; running time, the cumulative running time of the equipment; production speed, the working rate of the production line; failure rate, the frequency of equipment failures; maintenance record, the historical record of equipment maintenance; and safety status, the safe operating status of the machine.

[0028] It should be noted that this embodiment is only for illustrating an industrial Internet of Things data format and type of the present invention. In actual applications, the attributes of industrial Internet of Things data are usually more than 10, and the number of attributes of industrial Internet of Things data may reach dozens or even hundreds.

[0029] S2. Input the industrial Internet of Things data of each node into a pre-constructed distributed federated learning model, and upload the trained model parameters to the central server; In the federated learning framework, each node does not need to share the industrial Internet of Things data of its local training model. Instead, it trains the local model and sends the updated model to the centralized learning unit for aggregation. The model training architecture of distributed federated learning is as Figure 2 shown: In the model training architecture of distributed federated learning, in each round of iteration, each node separately conducts local model training and uploads the trained model parameters to the central server. The central server completes parameter aggregation and update, and issues the updated parameters to each node to start a new round of iteration until the training converges.

[0030] In the model training architecture of distributed federated learning, the federated average algorithm is used as the model aggregation algorithm. The federated average algorithm realizes the collaborative training of local models through multiple global iterations. Specifically, for each global iteration, let the number of nodes be , the total number of samples it owns be , and the number of samples of the th node be , the objective function of federated learning is defined as: ; ; In the formula, is the objective function of federated learning, is the model parameter, is the model parameter for the loss prediction of the th industrial Internet of Things data sample, is the training loss, is the th industrial Internet of Things data sample's sample feature, The th industrial Internet of Things data sample's label. Preferably, the training loss is calculated using the cross-entropy loss function.

[0031] Furthermore, for the th node, the objective function of this node is defined as: ; In the formula, is the th node's sample quantity, is the th node's objective function, is the th node's industrial Internet of Things data distribution.

[0032] In one embodiment, taking the parameter update method of the th iteration of the th node as an example, let the parameter gradient of the th node be , then the way to update the model parameters at the th iteration is expressed as: ; In the formula, is the model parameter of the th iteration, is the model parameter of the th iteration, is the learning rate of the current parameter update, is the th node's sample quantity.

[0033] Furthermore, the parameter update method of the central server's global model is expressed as: ; In the formula, is the The parameters of the global model of the central server in the th iteration for the model parameters of the

[0034] Further, repeating the iteration operation means that the global model of the central server and the model training of each node are completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 5000 times.

[0035] S3. Use the generative adversarial network algorithm based on quantum state iteration to augment the obtained industrial Internet of Things data; The acquisition, annotation, and preprocessing of industrial Internet of Things training data are time-consuming and laborious, and insufficient training samples are likely to lead to poor model generalization ability and affect the accuracy of the model. The present invention uses the generative adversarial network algorithm based on quantum state iteration for sample generation, thereby realizing the augmentation of industrial Internet of Things data. The generative adversarial network algorithm consists of two parts: a generator and a discriminator. Based on the traditional generative adversarial network, when the generator generates industrial Internet of Things data, it adopts a quantum transcendence dynamic feature enhancement mechanism, dynamically adjusts based on the probability amplitude of the quantum state, optimizes the feature distribution of industrial Internet of Things data samples, and increases the diversity and complexity of the samples generated by the model; Specifically, the method of the generative adversarial network algorithm based on quantum state iteration is as follows: S301. Randomly initialize the parameters of the generator and the discriminator. Let the weight of the generator be and the bias of the generator be , the weight of the discriminator be , and the bias of the discriminator be . The initialization method is expressed as: ; ; In the formula, represents a normal distribution with a mean of 0 and as the variance; is a normal distribution; represents being subject to a specific distribution. Preferably, is set to 0.1.

[0036] S302. In each iteration, first update the generator. The generator receives random noise as input and generates new industrial Internet of Things data samples. The distribution of the random noise is: ; In the formula, represents the Gaussian noise vector input to the generator;​​ is the identity matrix; represents a normal distribution with a mean of 0 and an identity matrix as the variance.

[0037] Furthermore, the generated new industrial Internet of Things data samples have the same attributes as the real industrial Internet of Things data samples. For example, the attributes include: temperature, which reflects the temperature value of the device or environment; humidity, the humidity level of the environment; pressure, the pressure index during the operation of the device; vibration level, the degree of vibration generated when the machine is running; energy consumption, the energy consumption of the device during the operation process; running time, the cumulative running time of the device; production speed, the working rate of the production line; failure rate, the frequency of device failures; maintenance records, the historical records of device maintenance; safety status, the safe operation status of the machine. The generation method is expressed as: ; In the formula, is the hyperbolic tangent function; is the generated industrial Internet of Things data sample; is the coherence control parameter; is the quantum coherence function; represents the input as of the quantum coherence function. Preferably, is set to 0.1.

[0038] Industrial Internet of Things data usually has complexity and diversity. In order to generate data samples with similar attributes, the generator adopts a mechanism for dynamically adjusting the probability amplitude of quantum states during the generation process, which can optimize the feature distribution of the data samples, make the generated data closer to the statistical characteristics of real industrial data, and improve the diversity and complexity of the samples. For example, by adaptively calculating the quantum coherence function, industrial data with diverse attributes such as temperature, humidity, and pressure can be generated, thereby improving the generalization ability of the model. In one embodiment, the quantum coherence function is used to dynamically adjust the feature distribution of the generated samples, and the calculation method is expressed as: ; In the formula, is the phase of the th qubit, obtained through the interaction of qubits during the quantum calculation process; is the th component of the noise vector, is the total number of qubits.

[0039] S303. Update the discriminator. The discriminator receives the generated samples from the generator and the samples in the real industrial Internet of Things dataset, evaluates the two types of samples, and outputs the probability that they are real samples. When the probability that the generated samples are judged as real samples by the discriminator is within the interval of [45%, 55%], it indicates that the generated samples and the real samples are similar. Increase the perturbation factor of the discriminator to further optimize the quality of the generated samples, so as to more accurately simulate the real characteristics of the industrial Internet of Things data, which is expressed as: ; ; In the formula, is the Sigmoid activation function, is the real sample, is the generated sample; represents the discriminator function; is the discriminator perturbation factor, used to enhance or weaken the perturbation effect on the generated samples; is the perturbation adjustment function; represents that the input is of the discriminator function; represents that the input is of the discriminator function. Preferably, is set to 0.2.

[0040] In one embodiment, the perturbation adjustment function adopts dynamic adjustment based on the discriminant result difference, so that the discriminator can dynamically adjust its feedback strength according to the similarity between the generated samples and the real samples, thereby improving the discrimination efficiency and accuracy. The calculation method is expressed as: ; In the formula, is the basic perturbation strength, represents the output difference between the generated samples and the real samples in the discriminator. Preferably, is set to 0.005.

[0041] S304. During the adversarial training process, the generator and the discriminator are alternately updated in the way of training 100 rounds each. Through continuous confrontation and coordination, the two models are pushed towards the optimal direction, that is: the generator is committed to improving the quality of the generated samples to deceive the discriminator; the discriminator tries to distinguish between the generated samples and the real samples to improve the discrimination ability. The adversarial training of the generator and the discriminator is realized by minimizing the loss function, which is expressed as: ; ; In the formula, Represents the loss function of the generator, which updates the generator parameters by maximizing the probability that the generated samples are judged to be real; Represents the loss function of the discriminator, which maximizes the correct recognition of real samples while minimizing the probability of misidentifying generated samples; Represents the information entropy of the generated samples, which is used to evaluate the sample diversity; Represents the cosine similarity between real samples and generated samples, which is used to evaluate the closeness of the two in the feature space.

[0042] Furthermore, based on the measure of information entropy, the information entropy of the generated samples is used as a loss term for the generator to enhance the exploration ability of the model. The measure of information entropy enables the generator to explore more diverse sample distributions and avoid overly single generated samples. The cosine similarity ensures the closeness between the generated samples and real samples in the feature space, thereby reducing overfitting. For example, for the generated device running time or energy consumption data, it is both diverse in distribution and similar to the real data, further enhancing the data augmentation effect. The calculation method of the information entropy of the generated samples is expressed as: ; In the formula, is the weight of the control term, which adjusts the proportion of the information entropy term in the loss function; represents the probability distribution of the generated samples, which is estimated by the sample probabilities generated by the generator. Preferably, is set to 0.1.

[0043] Furthermore, the cosine similarity between real samples and generated samples is a regularization term based on sample similarity to reduce overfitting and improve the discrimination effect. The calculation method is expressed as: ; In the formula, is the weight of the regularization term, which is used to control the influence of the similarity measure in the loss function; is the L2 norm. Preferably, is set to 0.3; S305. Update the weight parameters of the generator and discriminator using the momentum-based gradient descent method, and update the bias parameters of the generator and discriminator using the gradient descent method. For the high-dimensional characteristics of industrial Internet of Things data, it can update the parameters of the generator and discriminator more efficiently. By optimizing and adjusting the momentum term of the weights, the model can still converge quickly under complex data characteristics, improving the training efficiency. The update method is expressed as: ; ; ; ; In the formula, represents a parameter update operation, represents the learning rate of the generative adversarial network; is the momentum term of the generator weights, is the momentum term of the discriminator weights; is the gradient of the generator's loss function with respect to the bias parameter of the generator, is the gradient of the discriminator's loss function with respect to the bias parameter of the discriminator. Preferably, is set to 0.03.

[0044] Furthermore, the update method of the momentum terms of the generator and discriminator weights is expressed as: ; ; In the formula, is the cumulative momentum term, is the momentum decay parameter; is the gradient of the generator's loss function with respect to the weight parameter of the generator, is the gradient of the discriminator's loss function with respect to the weight parameter of the discriminator. Preferably, is set to 0.95.

[0045] S306. Repeat the above steps iteratively until the preset stop iteration condition is satisfied, which indicates that the model training is completed. Through adversarial training, the generated production speed samples are not only diverse but also have the characteristics of real industrial distribution, effectively improving the prediction performance of the model. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0046] S4. Input the augmented industrial Internet of Things data into the neural network model for data mining; The augmented industrial Internet of Things data is mined through the neural network model for industrial Internet of Things data, that is, feature extraction is performed using the neural network. In the present invention, a 6-layer fully connected neural network is used for feature extraction. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, problems such as gradient disappearance, gradient explosion, or getting stuck in local optimal solutions may occur, affecting the stability of training and the performance of the model. The present invention uses a self-adjusting learning rate optimization algorithm to optimize the parameters of the neural network model to achieve the training of the neural network model. Specifically, the present invention adopts a self-adjusting learning rate mechanism that allows the network to dynamically adjust its weight update strategy according to the error of the previous training cycle, enhancing the model's response ability to initial conditions and parameter changes.

[0047] Specifically, the neural network algorithm based on self - adjusting learning rate is as follows: S401. Initialize the weights and biases of the neural network. In one embodiment, the initialization method is expressed as: ; ; In the formula, is the initial value of the weights of the neural network; is the initial value of the biases of the neural network; is the mean offset of the weights of the neural network; is the mean offset of the weight biases of the neural network; is the standard deviation of the weight initialization of the neural network; represents a normal distribution with a mean of 0 and a standard deviation of the identity matrix, is the identity matrix. Preferably, is set to 0.01, is set to 0.05, is set to 0.001.

[0048] S402. During the training process, evaluate the output error of the neural network model. To solve the complex problems brought by the high - dimensional, non - linear and diverse industrial Internet of Things data, avoid overfitting or underfitting by self - adjusting the learning rate, help the neural network jump out of the local optimal solution, and at the same time effectively alleviate the problems of gradient disappearance and gradient explosion in traditional methods. For example, through the mechanism of dynamically adjusting the learning rate, the model can automatically optimize the learning rate according to the error change, so as to more flexibly adapt to the characteristics of industrial Internet of Things data. The dynamic update method of self - adjusting the learning rate is expressed as: ; In the formula, is the learning rate of the neural network at the th iteration; is the learning rate of the neural network at the th iteration; is the adjustment factor of the neural network, controlling the sensitivity of the learning rate change; is the error change amount of the neural network in the previous cycle.

[0049] Furthermore, according to the updated learning rate, update the weights, which is expressed as: ; In the formula, is the weight of the neural network at the th iteration; is the weight of the neural network at the th iteration; is the error of the neural network, which is calculated by the preset Softmax function for the output features of the last layer of the neural network; is the symbol of partial derivative.

[0050] S403. During the training process, overfitting is prevented by adopting a regularization strategy to ensure the generalization ability of the model. Specifically, regularization sparsity is performed by dynamically adjusting the number of nodes in the hidden layer of the neural network, which is expressed as: ; In the formula, is the number of nodes in the hidden layer of the neural network at the -th iteration; is the number of nodes in the hidden layer of the neural network at the -th iteration; is the node adjustment step size of the neural network hidden layer; is the sign function of the error change, which is used to determine whether to increase or decrease the number of nodes. That is, for , when is greater than zero, is 1, and when is less than or equal to zero,

[0051] is -1. In one embodiment, the node adjustment step size of the neural network hidden layer depends on a function of the network level, and the calculation method is expressed as: ; In the formula, is the rounding operation; is the influence coefficient for adjusting the adjustment step size; is the depth of the current network layer. Preferably, takes the value of 6, is set to 2.5.

[0052] S404. Repeat the above steps iteratively until the preset stop iteration condition is met, which indicates that the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0053] S5. Input the mined data into the pre-constructed classifier model for classification prediction to obtain classification data; the classifier model is a graph neural network model based on a bidirectional graph matching strategy; Data makes decisions through a classifier model, that is, the predicted class of the data is obtained using the classifier model. The present invention uses a graph neural network model based on a bidirectional graph matching strategy as the classification algorithm model. Traditional graph neural networks usually aggregate information of neighbor nodes only in a single direction. The present invention aggregates information both from global to local and from local to global through bidirectional graph matching to more comprehensively capture the complex relationships between nodes.

[0054] Specifically, the training process of the graph neural network algorithm based on the bidirectional graph matching strategy is as follows: S501. Construct an initial graph according to the data features after feature extraction. Each node represents a device, and the edge represents the connection between devices or data similarity. Let the dataset after feature extraction be . The construction of the graph is based on the similarity between devices, and the construction method is expressed as: ; In the formula, represents the edge weight between the th node and the th node in the graph; is the bandwidth parameter of the Gaussian kernel, which is used to control the sensitivity of the edge weight; is the th sample in the dataset after feature extraction, that is, the th node in the graph; is the th sample in the dataset after feature extraction, that is, the th node in the graph; represents the exponential function with the natural constant as the base; is the L2 norm. Preferably, is set to 0.1.

[0055] S502. Use the features of the nodes, that is, the feature vectors after feature extraction of each sample, and convert these features into low-dimensional vectors through the embedding layer, which is expressed as: ; In the formula, is the embedding vector of the th node, and are the weight and bias of the embedding layer respectively; is the ReLU activation function.

[0056] S503. Perform two-way information aggregation. Traditional graph neural network models only aggregate neighbor information unidirectionally. However, industrial Internet of Things data usually has the characteristic of coexistence of global and local dependencies. Through the two-way graph matching strategy, the present invention aggregates information from two directions, from global to local and from local to global, and can capture the complex topological relationships between devices more comprehensively. For example, in an industrial production line, a device may be jointly affected by neighboring devices (local) and the overall state of the entire production line (global). Through two-way information aggregation, the model can effectively capture these hierarchical dependencies and improve the accuracy of classification. In the global-to-local aggregation process, information is propagated throughout the graph, and each node aggregates information from all its neighbors, emphasizing the influence of the global context, which is expressed as: ; In the formula, represents the edge weight between the th node and the th node in the graph; Moreover, in the local-to-global aggregation process, each node only aggregates information from its direct neighbors to capture local features, which is expressed as: ; In the formula, and respectively represent the feature vectors after global-to-local and local-to-global aggregation in the th iteration, and respectively represent the feature vectors after global-to-local and local-to-global aggregation in the th iteration, is the weight based on the attention mechanism in the th iteration, represents the set of nodes.

[0057] Furthermore, the calculation method of the weight based on the attention mechanism in the th iteration is expressed as: ; In the formula, is the LeakyReLU activation function, is the parameter vector in the attention mechanism, used to calculate the attention weight; is transpose; is the weight matrix used for node feature transformation in the attention mechanism; represents the embedding vector of the th iteration of the th node; represents the th iteration of the The embedding vector of a node; Indicates the th iteration of the embedding vector of the node; Indicates the vector concatenation operation.

[0058] S504. Fuse the features obtained by the two aggregation strategies. Further, use a conditional random field layer to process the fused node features. The conditional random field layer takes into account the dependencies between nodes and realizes the final classification of IoT devices by maximizing the posterior probability, optimizing the classification decision process. By maximizing the posterior probability, the dependencies between nodes are considered. For example, in an industrial cold chain logistics monitoring scenario, the temperature control states of devices often show strong dependencies (such as a group of connected refrigeration units). The conditional random field layer models this dependency relationship to make the classification more accurate, expressed as: ; ; In the formula, is the fused feature vector, and are the weights and biases of the fusion layer, represents the conditional probability of each node category given the graph structure and node features. The loss function is to minimize this conditional probability; and respectively represent the feature vectors after global-to-local and local-to-global aggregation; where represents the feature vector after global-to-local aggregation, represents the feature vector after local-to-global aggregation; is the true label matrix of the sample; represents the conditional random field layer function.

[0059] Further, the calculation method of the weights of the fusion layer is expressed as: ; In the formula, is an orthogonal matrix, representing the basis of the feature space; is a diagonal matrix, and its diagonal elements represent the scaling coefficients in each direction, used to adjust the sensitivity of feature fusion; is the transpose.

[0060] S505. Repeat the above steps iteratively until the preset stop iteration condition is satisfied, which means the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0061] S6. Based on the classified data, determine the data analysis results of the industrial Internet of Things cloud platform, specifically: input the industrial Internet of Things data of each node into the trained distributed federated learning model, and then perform data expansion and data mining, and further input the processed features into the classifier model to train the classifier, and then obtain the classification results. In this embodiment, the classification categories include the marked categories including different operating states of the production line equipment, specifically including: "normal operation", "early failure", "mid-failure", and "late failure".

[0062] In order to verify the effect of this method, Figure 3 As shown in the figure, in order to verify the necessity of the two-way information aggregation strategy and the conditional random field fusion layer, the experiment compared the complete model with the simplified version without key components. Five model configurations were compared horizontally: the complete model, the basic model without two-way aggregation, the simplified model with only global aggregation, the simplified model with only local aggregation, and the model without conditional random field fusion. (a) is a comparative experiment of the five models in classification accuracy. The horizontal axis represents the five models (the complete model, the basic model without two-way aggregation, the simplified model with only global aggregation, the simplified model with only local aggregation, and the model without conditional random field fusion), and the vertical axis represents the classification accuracy; (b) is a comparative experiment of the five models in training time. The horizontal axis represents the five models (the complete model, the basic model without two-way aggregation, the simplified model with only global aggregation, the simplified model with only local aggregation, and the model without conditional random field fusion), and the vertical axis represents the training time (in hours). The experimental results show that the red column of the complete model is significantly higher than other blue column configurations in terms of accuracy, and the performance drops most significantly after removing the bidirectional aggregation, proving that the complementarity of global and local information plays a key role in modeling the topological relationship of the Industrial Internet of Things. Although the training time of the complete model in the right curve graph increases slightly, its accuracy improvement far exceeds the time cost, reflecting the advantage of bidirectional aggregation in feature learning efficiency.

[0063] like Figure 4As shown in the figure, by simulating the dynamic changes of device connection density and signal strength in a real industrial environment, the adaptability of different algorithms in complex scenarios is evaluated. The abscissa uses the product of device density and connection strength as a quantitative index for scene complexity, and the ordinate uses the comprehensive evaluation index F1 value. The red curve of this technology shows a gentle downward trend during the process of increasing complexity, while the dotted curve of the traditional method shows an accelerating downward trend, especially forming an obvious performance gap in the high-complexity region. The gradient-filled area between the curves represents the stability interval of the algorithm performance. The narrow filled band of this technology indicates its strong robustness to environmental changes. When the abscissa exceeds the critical value, the fluctuation amplitude of the traditional method increases significantly, while this technology effectively suppresses the performance oscillation caused by complex topologies through a bidirectional attention mechanism to adaptively adjust the information weight.

[0064] As Figure 5 shown, aiming at the common noise interference problem in industrial sensor data, by adding Gaussian noise with different intensities, the feature stability of the algorithm is tested. The abscissa represents the noise standard deviation, and the ordinate shows the classification accuracy. The four colored curves respectively represent this technology and three mainstream graph neural network methods. As the noise increases, the red curve of this technology shows a slow linear downward trend, while the curves of other methods show a cliff-like drop in the medium-high noise range. The divergent trend between the curves intuitively shows the dual advantages of this technology in feature cleaning and error correction.

Claims

1. A data analysis method for an industrial Internet of Things cloud platform, characterized in that It includes the following steps: Obtain the industrial Internet of Things data of each node; Input the industrial Internet of Things data of each node into a pre-constructed distributed federated learning model, and upload the trained model parameters to the central server; Use the generative adversarial network algorithm based on quantum state iteration to augment the obtained industrial Internet of Things data; Input the augmented industrial Internet of Things data into a neural network model for data mining; Input the mined data into a pre-constructed classifier model for classification prediction to obtain classification data; The classifier model is a graph neural network model based on a bidirectional graph matching strategy; Based on the classification data, determine the data analysis result of the industrial Internet of Things cloud platform.

2. The data analysis method of an industrial Internet of Things cloud platform according to claim 1, wherein, During the model training process of distributed federated learning, the federated averaging algorithm is used as the model aggregation algorithm, and the federated averaging algorithm realizes the collaborative training of local models through multiple global iterations.

3. A data analysis method for an industrial Internet of Things cloud platform according to claim 1, characterized in that The specific method of using the generative adversarial network algorithm based on quantum state iteration is as follows: Randomly initialize the parameters of the generator and discriminator; Update the generator. The generator receives random noise as input and generates new industrial Internet of Things data samples; Update the discriminator. The discriminator receives the generated samples from the generator and the samples in the real industrial Internet of Things dataset, evaluates the two types of samples, and outputs the probability that it is a real sample. When the probability that the generated sample is judged as a real sample by the discriminator is in the interval [45%, 55%], it indicates that the generated sample and the real sample are similar, and the perturbation factor of the discriminator is increased to further optimize the quality of the generated sample; During the adversarial training process, the generator and discriminator are alternately updated in the way of training 100 rounds each; Use the gradient descent method based on momentum to update the weight parameters of the generator and discriminator, and use the gradient descent method to update the bias parameters of the generator and discriminator.

4. The data analysis method of an industrial Internet of Things cloud platform according to claim 1, wherein, The neural network model uses a neural network algorithm based on self-adjusting learning rate.

5. A method for data analysis of an industrial Internet of Things cloud platform according to claim 1, characterized in that The specific method of the graph neural network model based on the bidirectional graph matching strategy is as follows: Construct an initial graph according to the data features after feature extraction; For each feature vector after sample feature extraction, convert the feature into a low-dimensional vector through the embedding layer; Through the bidirectional graph matching strategy, perform bidirectional information aggregation from two directions: from global to local and from local to global; Fuse the aggregated features, and use the conditional random field layer to process the fused node features.

6. The data analysis method of an industrial Internet of Things cloud platform according to claim 3, characterized in that, The generator adopts a probability amplitude dynamic adjustment mechanism of quantum state during the generation process.

7. A method for data analysis of an industrial Internet of Things cloud platform according to claim 1, characterized in that, The industrial Internet of Things data of each node comes from the industrial Internet of Things cloud platform, covering data from sensors, machine equipment, and industrial control systems on the production line.

8. The data analysis method of an industrial Internet of Things cloud platform according to claim 5, wherein The construction of the initial graph is based on the similarity between devices, and the construction method is expressed as: ； In the formula, represents the edge weight between the -th node and the -th node in the graph; is the bandwidth parameter of the Gaussian kernel, which is used to control the sensitivity of the edge weight; is the -th sample in the dataset after feature extraction, that is, the -th node in the graph; is the -th sample in the dataset after feature extraction, that is, the -th node in the graph; represents the exponential function with the natural constant as the base; is the L2 norm.

9. A data analysis method for an industrial Internet of Things cloud platform according to claim 5, characterized in that The conditional random field layer considers the dependency relationship between nodes and realizes the final classification of Internet of Things devices by maximizing the posterior probability, optimizing the classification decision-making process.

Images