Real-time steam generator state monitoring method based on artificial intelligence

Abstract

A steam generator state real-time monitoring method based on artificial intelligence belongs to the field of artificial intelligence, and comprises the following steps: S1, data acquisition and labeling; s2, performing sample generation by adopting a quantum generative adversarial network based on random projection embedding to realize data expansion; s3, inputting the expanded data into a feature extraction model to train the feature extraction model, and performing feature extraction by adopting a five-layer full-connection neural network; s4, inputting the data subjected to feature extraction into a feature dimension reduction model for feature dimension reduction model training, and realizing feature dimension reduction by adopting a self-encoding neural network algorithm based on locality preserving projection; s5, inputting the data subjected to dimension reduction into a classifier for training a classifier model; and S6, identifying and monitoring the state of the steam generator. According to the method, the problems of insufficient sample quantity and lack of data diversity can be solved, and the robustness when the model has noise or fuzzy classification boundary data is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a real-time monitoring method for a steam generator state based on artificial intelligence. Background Art

[0002] With the improvement of industrial automation and intelligence, the demand for efficiency and safe operation monitoring of energy equipment such as steam generators is increasing. As a key device for energy conversion, steam generators play a core role in many industrial processes. Real-time monitoring of their status is crucial to ensuring production safety, improving energy efficiency and reducing environmental impact. However, traditional monitoring methods rely on limited sensor data and manual monitoring, and often find it difficult to accurately capture the complex dynamic changes of equipment in real time, especially when faced with large-scale data processing and real-time response requirements. There are obvious limitations. In addition, conventional data processing methods are insufficient in samples, lack of data diversity, and the ability to process complex nonlinear data structures, which limits the effectiveness of monitoring systems in abnormality prediction and fault diagnosis.

[0003] In response to the above shortcomings, the Chinese invention patent with application number CN202311255147.0 proposes a drone detection method, a machine learning model training method and related equipment. The method includes: obtaining an initial image, using a target detection model to detect at least one first area in the initial image, obtaining a detection result of the target object in the first area, determining the first area where the target object whose detection result meets the preset requirements is located as the second area, using a classification model to perform binary classification on the second area, obtaining a second classification label of the target object in the second area, screening out the second area whose second classification label of the target object is the drone category, and determining the target object in the screened second area as a drone. This application first screens the initial image based on the target detection model, then accurately classifies it through the classification model, and performs secondary screening. The combination of the two screenings can effectively reduce the false alarm rate of drone detection. The Chinese invention patent with application number CN202110250990.4 proposes a processing method, device, storage medium and electronic device for a machine learning model, wherein the method includes: obtaining a configuration file and processing data, wherein the configuration file includes at least one processing process of at least one machine learning model; according to each processing process in the configuration file, calling the executor corresponding to each processing process to obtain the processing result of each processing process, wherein the executor executes the corresponding processing process based on the machine learning model and processing data corresponding to each processing process in the configuration file. In this embodiment, by setting the executors of each processing process of the machine learning model, the execution of each processing process is realized, and the execution result of each processing process is obtained. There is no need to write program code, which reduces the implementation difficulty of the machine learning model and improves the learning and research and development efficiency of the machine learning model. The Chinese invention patent with application number CN202011607378.X proposes a training method, server and computer device for a machine learning model, wherein the method includes: generating a homomorphic key pair, wherein the homomorphic key pair includes a public key and a private key; sending the public key to multiple data owner servers so that the multiple data owner servers encrypt their own data through the public key, and send the ciphertext data generated after encryption to the machine learning server, wherein the machine learning server performs model training and testing based on the ciphertext data provided by the multiple data owner servers to generate a ciphertext machine learning model; and receiving the ciphertext machine learning model sent by the machine learning server, and decrypting it according to the private key to form a plaintext machine learning model, thereby being able to protect the data provided by the data owner server, and being able to protect the plaintext machine learning model, with high security.

[0004] The above technical solution is innovative to a certain extent, but there are still the following problems that need to be further solved: 1. In traditional methods, data samples are usually limited by the actual collection situation, which often leads to insufficient sample quantity, lack of sufficient diversity and representativeness, and limits the generalization ability of the model.

[0005] 2. Traditional feature extraction methods may experience gradient vanishing or gradient explosion due to structural or algorithmic limitations, or fall into local optimal solutions due to insufficient algorithmic optimization strategies, affecting the training efficiency and performance of the model.

[0006] 3. Traditional feature dimensionality reduction technology may not be able to effectively maintain the main features and local structure of the data when processing complex and nonlinear data, resulting in the loss of important information in the reduced data, affecting the effect of subsequent processing.

[0007] 4. Traditional classification methods usually lack sufficient robustness when processing data with noise or unclear classification boundaries, resulting in unstable or inaccurate classification results, especially in complex monitoring scenarios. Summary of the invention

[0008] The purpose of the present invention is to propose a real-time monitoring method for steam generator status based on artificial intelligence, which can solve the problems of insufficient sample number and lack of data diversity in traditional methods, and can effectively avoid common problems of gradient disappearance, explosion or falling into local optimality, thereby enhancing the robustness of the model when processing data with noisy or fuzzy classification boundaries.

[0009] The technical solution adopted by the present invention is: The method for real-time monitoring of steam generator status based on artificial intelligence comprises the following steps: S1. Data collection and labeling: The collected data comes from the process data of the biomass steam generator, as well as the control status data and system operation status of the induced draft fan, blower, feeder, and circulation pump. The data is collected in real time through sensors and monitoring systems and stored in a structured JSON format. The collected data is manually labeled; S2, data expansion: Use quantum generative adversarial networks based on random projection embedding to generate samples and achieve data expansion; S3, feature extraction model training: the expanded data is input into the feature extraction model for training, and a 5-layer fully connected neural network is used for feature extraction; S4: Feature dimensionality reduction model training: input the feature-extracted data into the feature dimensionality reduction model for feature dimensionality reduction model training; S5, classifier model training: input the dimension-reduced data into the classifier to train the classifier model; S6. Steam generator status recognition and monitoring: The new steam generator status data is processed through the trained model to achieve real-time monitoring and status classification. Specifically, the collected original data is input into the trained feature extraction and feature dimension reduction model for feature processing, and then the processed features are input into the classifier model for classifier training to obtain the classification results.

[0010] Furthermore, in the data expansion of step S2, the acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. A quantum generative adversarial network based on random projection embedding is used to generate samples, thereby achieving data expansion; the quantum generative adversarial network based on random projection embedding maps high-dimensional data to low-dimensional quantum states through random projection, and then uses quantum mechanisms to accelerate the training process of the generative adversarial network, and designs the initialization process of the quantum state through random matrix theory, thereby improving the randomness and coverage of data mapping, and ensuring the diversity and practicality of generated data.

[0011] The training process of the quantum generative adversarial network algorithm based on random projection embedding is as follows: S201, quantum generative adversarial network first maps high-dimensional data to low-dimensional quantum state space, and specifically uses random projection embedding method to initialize the quantum state of quantum generative adversarial network. The initial quantum state is defined as: ; In the formula, represents the initialized quantum state; Indicates The complex probability amplitude of the data points; Indicates The quantum ground state after the data points are mapped; N represents the total number of data samples; S202. During the training process of the quantum generator, a quantum state with a specific distribution is generated through quantum gate operations, and the parameters of the quantum generator are adjusted to minimize the difference between the generated data and the actual data. The parameter update method of the quantum generator is expressed as: ; In the formula, represents the parameters of the quantum generator; Indicates The parameters of the quantum generator for the iteration; Indicates The parameters of the quantum generator for the iteration; represents the learning rate of the quantum generator; represents the quantum information gain loss function; Denotes the quantum information gain loss function with respect to the parameter The gradient of For the Dynamic adjustment coefficient of iteration; Among them, the dynamic adjustment coefficient is dynamically adjusted according to the coefficient of variation of the data, and the adjustment method is expressed as: ; In the formula, is the first hyperparameter to be tuned; is the second tuning hyperparameter; is the coefficient of variation; The calculation method of the coefficient of variation is expressed as: ; in, and They are the standard deviation and mean of the current input batch data respectively; S203. The quantum state output by the quantum generator is converted into synthetic data for machine learning after inverse quantum Fourier transformation, which is expressed as: ; In the formula, represents the generated synthetic data; represents the inverse quantum Fourier transform function; S204. During the training process of the quantum discriminator, the quantum discriminator constructs a loss function of the quantum discriminator by comparing the quantum states of the real data and the synthetic data, and optimizes the parameters by minimizing the loss function of the quantum discriminator. The updating process of the parameters of the quantum discriminator is expressed as: ; In the formula, represents the parameters of the quantum discriminator; Indicates The parameters of the quantum discriminator for the iteration; Indicates The parameters of the quantum discriminator for the iteration; represents the learning rate of the quantum discriminator; represents the loss function of the quantum discriminator; The loss function of the quantum discriminator is expressed as The gradient of Among them, the calculation method of the loss function of the quantum discriminator is expressed as: ; In the formula, It is The labels of the samples; It is The quantum state of the samples; S205, repeating iterative steps S201-S204 until the preset stop iteration condition is met, which means that the model training is completed; after the data expansion model training is completed, the trained data expansion model is used to increase the number of samples.

[0012] Furthermore, in the feature extraction model training of step S3, the expanded data is input into the feature extraction model for feature extraction model training, and a 5-layer fully connected neural network is used for feature extraction; in the prior art, some schemes use neural networks for feature extraction, and in some neural network structures, gradient vanishing, gradient explosion or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a neural network based on feature extraction fruit fly optimization as a feature extraction model, the input layer receives the collected multi-dimensional sensor data, and the hidden layer performs deep feature extraction through a multi-layer neural network. The present invention uses an information bottleneck strategy to constrain the feature extraction process, ensuring that the information transmitted to the output layer is the most critical, thereby reducing the propagation of redundant information and effectively accelerating the learning process.

[0013] Specifically, a neural network algorithm based on feature extraction fruit fly optimization is adopted, and its training process is as follows: S301, initialize the weights and biases of the neural network. The weights and biases are initialized as follows: ; In the formula, For the neural network The weight of the layer; For the neural network Bias of the layer; Indicates compliance with a specific distribution; is the number of neurons in the l-1th layer; The mean is 0 and the standard deviation is For the first layer of neurons, the standard deviation of the initialized normal distribution is set to 0.01; S302, data is transmitted layer by layer through the network. In the process of transmitting from the hidden layer to the output layer, the information bottleneck layer is used to force the network to learn more useful features by limiting the amount of information. Specifically, the forward propagation process is expressed as: ; In the formula, For the neural network The linear transformation output of the layer; For the neural network The output of the layer; For the neural network The activation noise of the layer; is the ReLU activation function; is the scaling factor of the information bottleneck; express norm; The intensity of the activation noise is dynamically adjusted according to the value of the loss function, and the adjustment method is expressed as: ; In the formula, is the baseline noise factor; is the loss function of the neural network; is the noise sensitivity hyperparameter; is a random noise vector generated from a standard normal distribution; Scaling Factors for Information Bottlenecks Dynamically adjust based on the entropy change of the input data to increase the model's sensitivity to data changes. The calculation method is expressed as: ; In the formula, For target data Entropy of For the The entropy of the layer output data; The target data input to the information bottleneck layer represents the probability data of the input data corresponding to the prediction of the correct sample label, which is calculated by the preset Softmax function; Target data The calculation method of entropy is expressed as: ; In the formula, is the entropy calculation function; all the logs in this article have no base because: in computer programming languages, the log function calculates the natural logarithm by default, so the base is not written, and when explaining the principles of machine learning models in this field, it is usually defaulted to no base, because no matter what the base is, as long as any base is fixed, it will not affect the process and results of model training.

[0014] S303, using the loss function to evaluate the difference between the feature expression of the output layer and the target feature, the calculation method of the loss function is expressed as: ; In the formula, is the loss function of the neural network, is the real output; The output of the model is calculated by the preset Softmax function to obtain the predicted output; is the balance parameter; S304. In the error back propagation stage, the fruit fly optimization algorithm is used to evaluate the loss corresponding to each set of parameters, and the parameters are guided to be updated in the direction of minimum loss. The parameter update method is expressed as: ; In the formula, For the Tier The weight parameters of the neural network at the iteration, For the Tier The weight parameters of the neural network at the iteration, For the Tier The bias parameters of the neural network for the iteration, For the Tier The bias parameters of the neural network for the iteration; is the weight parameter adjustment value obtained based on the fruit fly algorithm; is the bias parameter adjustment value obtained based on the fruit fly algorithm; is the learning rate of the neural network; S305, repeating iterative steps S301-S304 until the preset stop iteration condition is met, indicating that the model training is completed.

[0015] Furthermore, in the feature dimension reduction model training of step S4, the data after feature extraction is input into the feature dimension reduction model for feature dimension reduction model training. The present invention adopts an autoencoder neural network based on local preservation projection to realize feature dimension reduction. The autoencoder neural network based on local preservation projection includes an encoder part and a decoder part. The encoder is used to map the high-dimensional feature space to the low-dimensional feature space, and the decoder attempts to restore the original data from the low-dimensional space, thereby maintaining the main features of the data while reducing the dimension. The present invention adopts a feature-based nonlinear hash coding strategy to retain the local structure of the data during the coding process, and maps the features to the hash space through a nonlinear function, thereby enhancing the ability of the model to process nonlinear data structures.

[0016] Specifically, the autoencoder neural network algorithm based on local preservation projection is used to achieve feature dimensionality reduction. The training process is as follows: S401, randomly initialize the weights and biases of the autoencoder neural network, and the initialization method is expressed as: ; In the formula, is the weight parameter of the autoencoder neural network, is the bias parameter of the autoencoder neural network, is the initialization standard deviation of the autoencoder neural network; It is the input layer dimension of the autoencoder neural network, that is, the feature dimension after feature extraction; is the hidden layer dimension of the autoencoder neural network; To generate Standard normal distribution of intervals; generate dimensional zero vector; S402, the input feature data after dimensionality reduction is processed by the encoder part to form an encoded low-dimensional representation. The forward propagation process is expressed as: ; In the formula, represents the encoded feature representation; represents the attention weight; is the dot product symbol; is the Sigmoid activation function; is the weight matrix The transpose of It is the data after feature dimension reduction that is input into the autoencoder neural network; The attention weight is weighted by the input features to highlight the features that are more important to the current monitoring task. The calculation method is expressed as: ; In the formula, is a learnable weight matrix; is the Softmax function; S403, in the hidden layer of the autoencoder neural network, a local preservation projection strategy is adopted to optimize the output of the hidden layer neurons according to the adjacency relationship of the data to maintain the local relationship between the data points, which is expressed as: ; In the formula, It is a local preservation loss term, which is used to constrain the training process of hidden layer neurons; It is Samples and The similarity weight between samples; express norm; is the bandwidth parameter of the kernel function; S404, mapping the locally preserved low-dimensional features to the hash space through a nonlinear function, optimizing the dimensionality reduction effect by improving the separability of the data, and enhancing the feature expression ability of the model, which is expressed as: ; In the formula, It is the hash-encoded feature; It is the mapping matrix from feature space to hash space; It is the bias term in the mapping process; is a symbolic function used to generate a binary hash code; The calculation method is expressed as: ; In the formula, express The transpose of is the regularization coefficient to ensure numerical stability; represents the identity matrix; S405, by calculating the output layer and the original input The reconstruction error between , the gradient descent method is used to update the parameters of the autoencoder neural network and , optimize the performance of the entire network, and update the model parameters as follows: ; In the formula, is the total loss function; for The transpose of is the reconstructed input data; is the regularization parameter; is the learning rate of the autoencoder neural network; is the Sigmoid activation function; Indicates parameter update operation; The total loss function Substituting into the update formula of weights and biases, we get the update method of weights and biases, which is expressed as: ; In the formula, the gradient term and Represent the partial derivatives of the reconstruction error with respect to weight and bias, respectively. and represents the partial derivatives of the local preservation loss with respect to weights and biases; Weight Matrix The update is performed by gradient descent, and the update method is expressed as: ; In the formula, represents the partial derivative of the total loss with respect to the attention weight; represents the partial derivative of the attention weight with respect to the weight matrix; S406: Repeat iterative steps S401-S405 until the preset stop iteration condition is met, indicating that the model training is completed.

[0017] Furthermore, in the classifier model training of step S5, the dimension-reduced data is input into the classifier to train the classifier model. The present invention adopts a random forest algorithm based on sparse subspace learning as a classifier. The random forest algorithm uses multiple decision trees to construct an integrated learning framework, and enhances the generalization ability and accuracy of the model by constructing each tree in a randomly selected subspace. The present invention adopts fuzzy variable constraints on the basis of the random forest algorithm to make the model more robust when processing data with noise or fuzzy classification boundaries. By adopting fuzzy logic in the node splitting process of the decision tree, the process of feature selection and classification decision is optimized, thereby better adapting to the complex situation in steam generator state monitoring.

[0018] Specifically, the random forest algorithm based on sparse subspace learning is used for training, and the process is as follows: S501, initialize the number of decision trees and the number of features per tree , the number of features of each tree is initialized as: ; In the formula, The total number of data features after the input features are reduced in dimension; S502, randomly extract from the original feature set Features are selected randomly from the training samples to form the training data for each tree; for each tree, decision nodes are recursively constructed. At each node, fuzzy logic is used to evaluate and select the contribution of each feature to sample classification, allowing partial overlap and uncertainty to increase the model's ability to handle complex and fuzzy data. The calculation method of information gain based on fuzzy logic at node splitting is expressed as: ; In the formula, is the information gain based on fuzzy logic, which is used to select the split feature se and split threshold te; It is the maximum value operation under the conditions of split feature se and split threshold te; is the fuzzy entropy under the threshold te; For the sample weights; nf is the total number of input samples in the current batch; is the membership function of the category under the threshold te; For sample The total membership of Among them, fuzzy entropy represents fuzzy membership, and the calculation method is expressed as: ; In the formula, The sample belongs to the threshold te. The fuzzy probability of each category is is the set of all categories; S503, in each splitting process of the decision tree, the fuzzy membership is calculated, and the splitting features and points are selected according to the fuzzy membership to optimize the structure of the tree. The calculation method is expressed as: ; In the formula, is the steepness parameter that controls the membership; For sample Kernel function under threshold te; is the smoothness parameter that controls the membership; Kernel Function The calculation method is expressed as: ; In the formula, is the bandwidth parameter of the adaptive kernel function; The calculation method of the bandwidth parameter of the adaptive kernel function is expressed as: ; In the formula, is the regulating factor; is the median calculation function, Representation sample and its neighborhood samples The median of the Euclidean distance between ; S504, during the growth process of the decision tree, split each node until a stop condition is met; S505: After all decision trees stop classifying, it means that the model training is completed, and the output of the trained model is output as the final classification result by voting.

[0019] Further, in step S202, the quantum information gain loss function The calculation is performed using the Wasserstein loss based on quantum states, and the calculation method is expressed as: ; In the formula, Express expectations, and represent the real and generated quantum states respectively; is an interpolation between the real and generated quantum states; represents the discriminant function of the quantum discriminator; is the balance coefficient; for Norm.

[0020] Furthermore, in step S203, the generated synthetic data Adjustment is made through hybrid kernel density estimation, using the kernel function to more accurately simulate the target data distribution, expressed as: ; In the formula, represents the mixed kernel density function; Indicates The mixing weight of each core; Indicates The bandwidth of the kernel function; indicates nuclear center; Representation data No. Features Represents the generated synthetic data The characteristic dimension of Among them, the calculation method of the mixed kernel density function is expressed as: ; In the formula, and It’s data No. The mean and standard deviation of each feature; is the location parameter of the Laplace kernel; is a small positive number that prevents the denominator from being zero.

[0021] Further, in step S304, the calculation methods of the weight parameter adjustment amount and the bias parameter adjustment amount obtained based on the fruit fly algorithm are respectively expressed as: ; In the formula, is the search radius of the fruit fly algorithm; Generates random numbers from a standard normal distribution in the interval (0,1); is a symbolic function; Represents the loss function Gradients with respect to weights; Represents the loss function Gradient with respect to the bias.

[0022] Further, in step S502, the fuzzy probability The calculation method is expressed as: ; In the formula, It is a sample Under the threshold te, it belongs to The fuzzy membership of the categories; It is a sample Fuzzy membership under threshold te.

[0023] Beneficial effects of the present invention: 1. The present invention realizes the generation and expansion of data samples through a quantum generative adversarial network based on random projection embedding, which solves the problems of insufficient sample quantity and lack of data diversity in traditional methods.

[0024] 2. The present invention utilizes a fully connected neural network combined with an information bottleneck strategy and a fruit fly optimization algorithm for feature extraction, effectively avoiding common problems of gradient vanishing, explosion, or falling into local optimum.

[0025] 3. The present invention utilizes an autoencoder neural network based on local preserving projection to perform feature dimensionality reduction, and improves the model's ability to process nonlinear data by maintaining the local relationship between data points.

[0026] 4. The present invention utilizes the random forest algorithm based on sparse subspace learning combined with fuzzy logic to enhance the robustness of the model when processing data with noise or fuzzy classification boundaries. DETAILED DESCRIPTION

[0027] The method of the present invention is based on the existing device hardware. The monitored steam generator is a vertical single-turn spiral tube gas steam generator, which uses natural gas as fuel and is mainly used for steam production; the burner of the steam generator is installed at the bottom, using full premixed low-nitrogen combustion technology, and the center of the burner is equipped with a copper structure with a metal mesh, extending to the upper part of the steam generator; the steam generator body is designed as a single-turn spiral fin tube, which has a compact and simple structure and a small water capacity. The water volume in the boiler is less than 30L, and no additional monitoring and inspection is required. The small water capacity brings the advantages of rapid heating and low shutdown heat loss. The heat exchange tube of the steam generator is a single-turn spiral fin tube. The spiral fins are fixed to the pipe by high-frequency welding. After welding, the fins are flattened and the flue gas passes through the baffle on the outside of the spiral fin tube. The baffle evenly distributes the flue gas, slows down the flow rate, improves the heat exchange efficiency, and thus reduces the exhaust temperature.

[0028] The principle of the steam generator is that the burner generates high-temperature flue gas, which passes through the spiral fin tube and is affected by the smoke baffle. The flue gas temperature drops to about 200°C, and then enters the condenser for further cooling. The exhaust temperature drops to fifty or sixty degrees and is finally discharged from the exhaust port.

[0029] A monitoring system is provided on the steam generator, which includes the following main modules: 1. Data acquisition and transmission module: collects the operation data of the steam generator in real time through a variety of sensors for subsequent algorithm analysis; 2. Data processing and expansion module: Generate more diverse data samples through data expansion to ensure sufficient data during training to improve the generalization ability of the model; 3. Feature extraction and dimensionality reduction module: through feature extraction and feature dimensionality reduction, to maintain the local structural relationship of the data; 4. State classification module: Identify and classify different states to realize multi-state identification of steam generators; 5. Real-time monitoring and alarm module: Display the status information of the steam generator in real time, automatically trigger an alarm to prompt the operator in the event of a fault, and record all alarm information.

[0030] On this basis, a real-time monitoring method for steam generator status based on artificial intelligence was established, which includes the following steps: S1. Data collection and annotation The collected data comes from the process data of the biomass steam generator, as well as the control status data and system operation status of the induced draft fan, blower, feeder, and circulation pump. The data is collected in real time through sensors and monitoring systems and stored in a structured JSON format. In this embodiment, the attributes of the collected steam generator status data include: Ta represents steam temperature in degrees (°C); Pa represents steam pressure in Pascal (Pa); Fa represents fuel flow in kilograms per hour (kg / h); Aa represents air flow in cubic meters per hour (m³ / h); Va represents steam flow rate in meters per second (m / s); Ea represents heat output in joules (J); Ca represents chemical composition ratio in percentage (%); Da represents dew point temperature in degrees (°C); Oa represents oxygen content in percentage (%); Ba represents combustion efficiency in percentage (%).

[0031] It should be noted that this embodiment is only used to illustrate one data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds.

[0032] The collected data is manually labeled; in this embodiment, the labeled categories include: normal working state, initial fault state, mid-fault state, severe fault state, and working stop state, a total of 5 categories.

[0033] S2. Data expansion In the task of the present invention, the acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. A quantum generative adversarial network based on random projection embedding is used to generate samples, thereby realizing data expansion; the quantum generative adversarial network based on random projection embedding maps high-dimensional data to low-dimensional quantum states through random projection, and then uses quantum mechanisms to accelerate the training process of the generative adversarial network, and designs the initialization process of the quantum state through random matrix theory, thereby improving the randomness and coverage of data mapping, and ensuring the diversity and practicality of generated data.

[0034] The training process of the quantum generative adversarial network algorithm based on random projection embedding is as follows: S201, quantum generative adversarial network first maps high-dimensional data to low-dimensional quantum state space, and specifically uses random projection embedding method to initialize the quantum state of quantum generative adversarial network. The initial quantum state is defined as: ; In the formula, represents the initialized quantum state; Indicates The complex probability amplitude of the data points; Indicates The quantum ground state after the data points are mapped; N represents the total number of data samples; S202. During the training process of the quantum generator, a quantum state with a specific distribution is generated through quantum gate operations, and the parameters of the quantum generator are adjusted to minimize the difference between the generated data and the actual data. The parameter update method of the quantum generator is expressed as: ; In the formula, represents the parameters of the quantum generator; Indicates The parameters of the quantum generator for the iteration; Indicates The parameters of the quantum generator for the iteration; represents the learning rate of the quantum generator; represents the quantum information gain loss function; Denotes the quantum information gain loss function with respect to the parameter The gradient of For the The dynamic adjustment coefficient of the iteration; preferably, Set to 0.01.

[0035] Among them, the dynamic adjustment coefficient is dynamically adjusted according to the coefficient of variation of the data, and the adjustment method is expressed as: ; In the formula, is the first hyperparameter to be tuned; is the second tuning hyperparameter; is the coefficient of variation; preferably, Set to 0.1, Set to 0.5.

[0036] The calculation method of the coefficient of variation is expressed as: ; in, and They are the standard deviation and mean of the current input batch data respectively; Quantum Information Gain Loss Function The calculation is performed using the Wasserstein loss based on quantum states, and the calculation method is expressed as: ; In the formula, Express expectations, and represent the real and generated quantum states respectively; is an interpolation between the real and generated quantum states; represents the discriminant function of the quantum discriminator; is the balance coefficient; for Preferably, Set to 0.3.

[0037] S203. The quantum state output by the quantum generator is converted into synthetic data for machine learning after inverse quantum Fourier transformation, which is expressed as: ; In the formula, represents the generated synthetic data; represents the inverse quantum Fourier transform function; Generated synthetic data Adjustment is made through hybrid kernel density estimation, using the kernel function to more accurately simulate the target data distribution, expressed as: ; In the formula, represents the mixed kernel density function; Indicates The mixing weight of each core; Indicates The bandwidth of the kernel function; indicates nuclear center; Representation data No. Features Represents the generated synthetic data The characteristic dimension of Among them, the calculation method of the mixed kernel density function is expressed as: ; In the formula, and It’s data No. The mean and standard deviation of each feature; is the location parameter of the Laplace kernel; is a small positive number that prevents the denominator from being zero. Preferably, Set to 0.01.

[0038] S204. During the training process of the quantum discriminator, the quantum discriminator constructs a loss function of the quantum discriminator by comparing the quantum states of the real data and the synthetic data, and optimizes the parameters by minimizing the loss function of the quantum discriminator. The updating process of the parameters of the quantum discriminator is expressed as: ; In the formula, represents the parameters of the quantum discriminator; Indicates The parameters of the quantum discriminator for the iteration; Indicates The parameters of the quantum discriminator for the iteration; represents the learning rate of the quantum discriminator; represents the loss function of the quantum discriminator; The loss function of the quantum discriminator is expressed as Preferably, Set to 0.01.

[0039] Among them, the calculation method of the loss function of the quantum discriminator is expressed as: ; In the formula, It is The labels of the samples; It is The quantum state of the samples; S205. Repeat iterative steps S201-S204 until a preset stop iteration condition is met, indicating that the model training is completed; in this embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0040] After the data expansion model training is completed, the trained data expansion model is used to increase the number of samples; in this embodiment, assuming that the original collected samples are 800, and the data expansion model expands and generates 200 samples, the expanded data set contains 1000 samples.

[0041] S3. Feature extraction model training The expanded data is input into the feature extraction model for training the feature extraction model, and a 5-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, gradient vanishing, gradient explosion or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a neural network based on feature extraction fruit fly optimization as a feature extraction model, the input layer receives the collected multi-dimensional sensor data, and the hidden layer performs deep feature extraction through a multi-layer neural network. The present invention uses an information bottleneck strategy to constrain the feature extraction process to ensure that the information transmitted to the output layer is the most critical, thereby reducing the propagation of redundant information and effectively accelerating the learning process.

[0042] Specifically, a neural network algorithm based on feature extraction fruit fly optimization is adopted, and its training process is as follows: S301, initialize the weights and biases of the neural network. The weights and biases are initialized as follows: ; In the formula, For the neural network The weight of the layer; For the neural network Bias of the layer; Indicates compliance with a specific distribution; is the number of neurons in the previous layer; The mean is 0 and the standard deviation is For the first layer of neurons, the standard deviation of the initialized normal distribution is set to 0.01; S302, data is transmitted layer by layer through the network. In the process of transmitting from the hidden layer to the output layer, the information bottleneck layer is used to force the network to learn more useful features by limiting the amount of information. Specifically, the forward propagation process is expressed as: ; In the formula, For the neural network The linear transformation output of the layer; For the neural network The output of the layer; For the neural network The activation noise of the layer; is the ReLU activation function; is the scaling factor of the information bottleneck; express norm; The intensity of the activation noise is dynamically adjusted according to the value of the loss function, and the adjustment method is expressed as: ; In the formula, is the baseline noise factor; is the loss function of the neural network; is the noise sensitivity hyperparameter; is a random noise vector generated from a standard normal distribution; preferably, Set to 0.002, Set to 2.

[0043] Scaling Factors for Information Bottlenecks Dynamically adjust based on the entropy change of the input data to increase the model's sensitivity to data changes. The calculation method is expressed as: ; In the formula, For target data Entropy of For the The entropy of the layer output data; is the target data input to the information bottleneck layer; Target data The calculation method of entropy is expressed as: ; In the formula, It is the entropy calculation function.

[0044] S303, using the loss function to evaluate the difference between the feature expression of the output layer and the target feature, the calculation method of the loss function is expressed as: ; In the formula, is the loss function of the neural network, is the real output; The output of the model is calculated by the preset Softmax function to obtain the predicted output; is a balance parameter; preferably, Set to 0.3.

[0045] S304. In the error back propagation stage, the fruit fly optimization algorithm is used to evaluate the loss corresponding to each set of parameters, and the parameters are guided to be updated in the direction of minimum loss. The parameter update method is expressed as: ; In the formula, For the Tier The weight parameters of the neural network at the iteration, For the Tier The weight parameters of the neural network at the iteration, For the Tier The bias parameters of the neural network for the iteration, For the Tier The bias parameters of the neural network for the iteration; is the weight parameter adjustment value obtained based on the fruit fly algorithm; is the bias parameter adjustment value obtained based on the fruit fly algorithm; is the learning rate of the neural network; preferably, Set to 0.01.

[0046] The calculation methods of the weight parameter adjustment and bias parameter adjustment based on the fruit fly algorithm are respectively expressed as: ; In the formula, is the search radius of the fruit fly algorithm; Generates random numbers from a standard normal distribution in the interval (0,1); is a symbolic function; Represents the loss function Gradients with respect to weights; Represents the loss function Gradient with respect to the bias.

[0047] S305. Repeat iterative steps S301-S304 until a preset stop iteration condition is met, indicating that the model training is completed; in this embodiment, the preset stop iteration condition is to reach a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0048] S4: Feature Dimensionality Reduction Model Training The data after feature extraction is input into the feature dimension reduction model for feature dimension reduction model training. The present invention uses an autoencoder neural network based on local preservation projection to achieve feature dimension reduction. The autoencoder neural network based on local preservation projection includes an encoder part and a decoder part. The encoder is used to map the high-dimensional feature space to the low-dimensional feature space, and the decoder attempts to restore the original data from the low-dimensional space, thereby maintaining the main features of the data while reducing the dimension. The present invention uses a feature-based nonlinear hash coding strategy to retain the local structure of the data during the coding process, and maps the features to the hash space through a nonlinear function, thereby enhancing the ability of the model to process nonlinear data structures.

[0049] Specifically, the autoencoder neural network algorithm based on local preservation projection is used to achieve feature dimensionality reduction. The training process is as follows: S401, randomly initialize the weights and biases of the autoencoder neural network, and the initialization method is expressed as: ; In the formula, is the weight parameter of the autoencoder neural network, is the bias parameter of the autoencoder neural network, is the initialization standard deviation of the autoencoder neural network; It is the input layer dimension of the autoencoder neural network, that is, the feature dimension after feature extraction; is the hidden layer dimension of the autoencoder neural network; To generate Standard normal distribution of intervals; generate -dimensional zero vector; preferably, Set to 100.

[0050] S402, the input feature data after dimensionality reduction is processed by the encoder part to form an encoded low-dimensional representation. The forward propagation process is expressed as: ; In the formula, represents the encoded feature representation; represents the attention weight; is the dot product symbol; is the Sigmoid activation function; is the weight matrix The transpose of It is the data after feature dimension reduction that is input into the autoencoder neural network; The attention weight is weighted by the input features to highlight the features that are more important to the current monitoring task. The calculation method is expressed as: ; In the formula, is a learnable weight matrix; is the Softmax function; S403, in the hidden layer of the autoencoder neural network, a local preservation projection strategy is adopted to optimize the output of the hidden layer neurons according to the adjacency relationship of the data to maintain the local relationship between the data points, which is expressed as: ; In the formula, It is a local preservation loss term, which is used to constrain the training process of hidden layer neurons; It is Samples and The similarity weight between samples; express norm; is the bandwidth parameter of the kernel function; preferably, Set to 0.1.

[0051] S404, mapping the locally preserved low-dimensional features to the hash space through a nonlinear function, optimizing the dimensionality reduction effect by improving the separability of the data, and enhancing the feature expression ability of the model, which is expressed as: ; In the formula, It is the hash-encoded feature; It is the mapping matrix from feature space to hash space; It is the bias term in the mapping process; is a symbolic function used to generate a binary hash code; The calculation method is expressed as: ; In the formula, express The transpose of is the regularization coefficient to ensure numerical stability; represents the identity matrix; S405, by calculating the output layer and the original input The reconstruction error between , the gradient descent method is used to update the parameters of the autoencoder neural network and , optimize the performance of the entire network, and update the model parameters as follows: ; In the formula, is the total loss function; for The transpose of is the reconstructed input data; is the regularization parameter; is the learning rate of the autoencoder neural network; is the Sigmoid activation function; Indicates parameter update operation; The total loss function Substituting into the update formula of weights and biases, we get the update method of weights and biases, which is expressed as: ; In the formula, the gradient term and Represent the partial derivatives of the reconstruction error with respect to weight and bias, respectively. and represents the partial derivatives of the local preservation loss with respect to weights and biases; Weight Matrix The update is performed by gradient descent, and the update method is expressed as: ; In the formula, represents the partial derivative of the total loss with respect to the attention weight; represents the partial derivative of the attention weight with respect to the weight matrix; S406. Repeat iterative steps S401-S405 until a preset stop iteration condition is met, indicating that the model training is completed; in this embodiment, the preset stop iteration condition is to reach a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0052] S5. Classifier model training The reduced-dimensional data is input into the classifier to train the classifier model. The present invention adopts a random forest algorithm based on sparse subspace learning as a classifier. The random forest algorithm uses multiple decision trees to build an integrated learning framework, and enhances the generalization ability and accuracy of the model by building each tree in a randomly selected subspace. The present invention adopts fuzzy variable constraints on the basis of the random forest algorithm to make the model more robust when processing data with noise or fuzzy classification boundaries. By adopting fuzzy logic in the node splitting process of the decision tree, the process of feature selection and classification decision-making is optimized, thereby better adapting to the complex situation in steam generator state monitoring.

[0053] Specifically, the random forest algorithm based on sparse subspace learning is used for training, and the process is as follows: S501, initialize the number of decision trees and the number of features per tree In this embodiment, the number of decision trees is initialized to 100, and the number of features of each tree is initialized to: ; In the formula, The total number of data features after the input features are reduced in dimension; S502, randomly extract from the original feature set Features are selected randomly from the training samples to form the training data for each tree; for each tree, decision nodes are recursively constructed. At each node, fuzzy logic is used to evaluate and select the contribution of each feature to sample classification, allowing partial overlap and uncertainty to increase the model's ability to handle complex and fuzzy data. The calculation method of information gain based on fuzzy logic at node splitting is expressed as: ; In the formula, is the information gain based on fuzzy logic, which is used to select the split feature se and split threshold te; It is the maximum value operation under the conditions of split feature se and split threshold te; is the fuzzy entropy under the threshold te; For the sample weights; nf is the total number of input samples in the current batch; is the membership function of the category under the threshold te; For sample The total membership of Among them, fuzzy entropy represents fuzzy membership, and the calculation method is expressed as: ; In the formula, The sample belongs to the threshold te. The fuzzy probability of each category is is the set of all categories; Fuzzy Probability The calculation method is expressed as: ; In the formula, It is a sample Under the threshold te, it belongs to The fuzzy membership of the categories; It is a sample Fuzzy membership under threshold te.

[0054] S503, in each splitting process of the decision tree, the fuzzy membership is calculated, and the splitting features and points are selected according to the fuzzy membership to optimize the structure of the tree. The calculation method is expressed as: ; In the formula, is the steepness parameter that controls the membership; For sample Kernel function under threshold te; is a smoothness parameter controlling the membership; preferably, Set to 1, Set to 3.14.

[0055] Kernel Function The calculation method is expressed as: ; In the formula, is the bandwidth parameter of the adaptive kernel function; The calculation method of the bandwidth parameter of the adaptive kernel function is expressed as: ; In the formula, is the regulating factor; is the median calculation function, Representation sample and its neighborhood samples The median of the Euclidean distance between Preferably, Set to 0.5.

[0056] S504: During the growth of the decision tree, each node is split until a stop condition is met. In this embodiment, the stop condition is when the number of samples under a node is less than a threshold. or reach the maximum depth of the decision tree Preferably, Set to 5, Set to 4.

[0057] S505: After all decision trees stop classifying, the model training is completed, and the output of the trained model is voted to output the final classification result; S6. Steam generator status identification and monitoring The new steam generator status data is processed by the trained model to achieve real-time monitoring and status classification. Specifically, the collected raw data is input into the trained feature extraction and feature dimension reduction model for feature processing, and then the processed features are input into the classifier model for classifier training to obtain the classification results. In this embodiment, the classification categories include: normal working state, initial fault state, mid-fault state, serious fault state, and work stop state, a total of 5 categories.

Claims

1. A real-time monitoring method for steam generator status based on artificial intelligence, characterized in that: The steps include: S1. Data collection and labeling: The collected data comes from the process data of the biomass steam generator, as well as the control status data and system operation status of the induced draft fan, blower, feeder, and circulation pump. The data is collected in real time through sensors and monitoring systems and stored in a structured JSON format. The collected data is manually labeled; S2. Data expansion: Use quantum generative adversarial networks based on random projection embedding to generate samples and achieve data expansion; S3, feature extraction model training: the expanded data is input into the feature extraction model for training, and a 5-layer fully connected neural network is used for feature extraction; S4: Feature dimensionality reduction model training: input the feature-extracted data into the feature dimensionality reduction model for feature dimensionality reduction model training; S5, classifier model training: input the dimension-reduced data into the classifier to train the classifier model; S6. Steam generator status recognition and monitoring: The new steam generator status data is processed through the trained model to achieve real-time monitoring and status classification. Specifically, the collected original data is input into the trained feature extraction and feature dimension reduction model for feature processing, and then the processed features are input into the classifier model for classifier training to obtain the classification results.

2. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 1, characterized in that: In the data expansion of step S2, the training process of sample generation is as follows: S201, quantum generative adversarial network first maps high-dimensional data to low-dimensional quantum state space, and specifically uses random projection embedding method to initialize the quantum state of quantum generative adversarial network. The initial quantum state is defined as: ； In the formula, represents the initialized quantum state; Indicates The complex probability amplitude of the data points; Indicates The quantum ground state after the data points are mapped; N represents the total number of data samples; S202. During the training process of the quantum generator, a quantum state with a specific distribution is generated through quantum gate operations, and the parameters of the quantum generator are adjusted to minimize the difference between the generated data and the actual data. The parameter update method of the quantum generator is expressed as: ； In the formula, represents the parameters of the quantum generator; Indicates The parameters of the quantum generator for the iteration; Indicates The parameters of the quantum generator for the iteration; represents the learning rate of the quantum generator; represents the quantum information gain loss function; Denotes the quantum information gain loss function with respect to the parameter The gradient of For the Dynamic adjustment coefficient of iteration; Among them, the dynamic adjustment coefficient is dynamically adjusted according to the coefficient of variation of the data, and the adjustment method is expressed as: ； In the formula, is the first hyperparameter to be tuned; is the second tuning hyperparameter; is the coefficient of variation; The calculation method of the coefficient of variation is expressed as: ； in, and They are the standard deviation and mean of the current input batch data respectively; S203. The quantum state output by the quantum generator is converted into synthetic data for machine learning after inverse quantum Fourier transformation, which is expressed as: ； In the formula, represents the generated synthetic data; represents the inverse quantum Fourier transform function; S204. During the training process of the quantum discriminator, the quantum discriminator constructs a loss function of the quantum discriminator by comparing the quantum states of the real data and the synthetic data, and optimizes the parameters by minimizing the loss function of the quantum discriminator. The updating process of the parameters of the quantum discriminator is expressed as: ； In the formula, represents the parameters of the quantum discriminator; Indicates The parameters of the quantum discriminator for the iteration; Indicates The parameters of the quantum discriminator for the iteration; represents the learning rate of the quantum discriminator; represents the loss function of the quantum discriminator; The loss function of the quantum discriminator is expressed as The gradient of Among them, the calculation method of the loss function of the quantum discriminator is expressed as: ； In the formula, It is The labels of the samples; It is The quantum state of the samples; S205, repeating iterative steps S201-S204 until the preset stop iteration condition is met, which means that the model training is completed; after the data expansion model training is completed, the trained data expansion model is used to increase the number of samples.

3. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 1, characterized in that: In the feature extraction model training of step S3, a neural network algorithm based on feature extraction fruit fly optimization is adopted, and its training process is as follows: S301, initialize the weights and biases of the neural network. The weights and biases are initialized as follows: ； In the formula, For the neural network The weight of the layer; For the neural network Bias of the layer; Indicates compliance with a specific distribution; is the number of neurons in the l-1th layer; The mean is 0 and the standard deviation is For the first layer of neurons, the standard deviation of the initialized normal distribution is set to 0.01; S302, data is transmitted layer by layer through the network. In the process of transmitting from the hidden layer to the output layer, the information bottleneck layer is used to force the network to learn more useful features by limiting the amount of information. Specifically, the forward propagation process is expressed as: ； In the formula, For the neural network The linear transformation output of the layer; For the neural network The output of the layer; For the neural network The activation noise of the layer; is the ReLU activation function; is the scaling factor of the information bottleneck; express norm; The intensity of the activation noise is dynamically adjusted according to the value of the loss function, and the adjustment method is expressed as: ； In the formula, is the baseline noise factor; is the loss function of the neural network; is the noise sensitivity hyperparameter; is a random noise vector generated from a standard normal distribution; Scaling Factors for Information Bottlenecks Dynamically adjust based on the entropy change of the input data to increase the model's sensitivity to data changes. The calculation method is expressed as: ； In the formula, For target data Entropy of For the The entropy of the layer output data; The target data input to the information bottleneck layer represents the probability data of the input data corresponding to the prediction of the correct sample label, which is calculated by the preset Softmax function; Target data The calculation method of entropy is expressed as: ； In the formula, is the entropy calculation function; S303, using the loss function to evaluate the difference between the feature expression of the output layer and the target feature, the calculation method of the loss function is expressed as: ； In the formula, is the loss function of the neural network, is the real output; The output of the model is calculated by the preset Softmax function to obtain the predicted output; is the balance parameter; S304. In the error back propagation stage, the fruit fly optimization algorithm is used to evaluate the loss corresponding to each set of parameters, and the parameters are guided to be updated in the direction of minimum loss. The parameter update method is expressed as: ； In the formula, For the Tier The weight parameters of the neural network at the iteration, For the Tier The weight parameters of the neural network at the iteration, For the Tier The bias parameters of the neural network for the iteration, For the Tier The bias parameters of the neural network for the iteration; is the weight parameter adjustment value obtained based on the fruit fly algorithm; is the bias parameter adjustment value obtained based on the fruit fly algorithm; is the learning rate of the neural network; S305, repeating iterative steps S301-S304 until the preset stop iteration condition is met, indicating that the model training is completed.

4. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 1, characterized in that: In the feature dimensionality reduction model training of step S4, the feature dimensionality reduction is achieved by using the autoencoder neural network algorithm based on local preservation projection. The training process is as follows: S401, randomly initialize the weights and biases of the autoencoder neural network, and the initialization method is expressed as: ； In the formula, is the weight parameter of the autoencoder neural network, is the bias parameter of the autoencoder neural network, is the initialization standard deviation of the autoencoder neural network; It is the input layer dimension of the autoencoder neural network, that is, the feature dimension after feature extraction; is the hidden layer dimension of the autoencoder neural network; To generate Standard normal distribution of intervals; generate dimensional zero vector; S402, the input feature data after dimensionality reduction is processed by the encoder part to form an encoded low-dimensional representation. The forward propagation process is expressed as: ； In the formula, represents the encoded feature representation; represents the attention weight; is the dot product symbol; is the Sigmoid activation function; is the weight matrix The transpose of It is the data after feature dimension reduction that is input into the autoencoder neural network; The attention weight is weighted by the input features to highlight the features that are more important to the current monitoring task. The calculation method is expressed as: ； In the formula, is a learnable weight matrix; is the Softmax function; S403, in the hidden layer of the autoencoder neural network, a local preservation projection strategy is adopted to optimize the output of the hidden layer neurons according to the adjacency relationship of the data to maintain the local relationship between the data points, which is expressed as: ； In the formula, It is a local preservation loss term, which is used to constrain the training process of hidden layer neurons; It is Samples and The similarity weight between samples; express norm; is the bandwidth parameter of the kernel function; S404, mapping the locally preserved low-dimensional features to the hash space through a nonlinear function, optimizing the dimensionality reduction effect by improving the separability of the data, and enhancing the feature expression ability of the model, which is expressed as: ； In the formula, It is the hash-encoded feature; It is the mapping matrix from feature space to hash space; It is the bias term in the mapping process; is a symbolic function used to generate a binary hash code; The calculation method is expressed as: ； In the formula, express The transpose of is the regularization coefficient; represents the identity matrix; S405, by calculating the output layer and the original input The reconstruction error between , the gradient descent method is used to update the parameters of the autoencoder neural network and , optimize the performance of the entire network, and update the model parameters as follows: ； In the formula, is the total loss function; for The transpose of is the reconstructed input data; is the regularization parameter; is the learning rate of the autoencoder neural network; is the Sigmoid activation function; Indicates parameter update operation; The total loss function Substituting into the update formula of weights and biases, we get the update method of weights and biases, which is expressed as: ； In the formula, the gradient term and Represent the partial derivatives of the reconstruction error with respect to weight and bias, respectively. and represents the partial derivatives of the local preservation loss with respect to weights and biases; Weight Matrix The update is performed by gradient descent, and the update method is expressed as: ； In the formula, represents the partial derivative of the total loss with respect to the attention weight; represents the partial derivative of the attention weight with respect to the weight matrix; S406: Repeat iterative steps S401-S405 until the preset stop iteration condition is met, indicating that the model training is completed.

5. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 1, characterized in that: In the classifier model training of step S5, the random forest algorithm based on sparse subspace learning is used for training, and the process is as follows: S501, initialize the number of decision trees and the number of features per tree , the number of features of each tree is initialized as: ； In the formula, The total number of data features after the input features are reduced in dimension; S502, randomly extract from the original feature set Features are selected randomly from the training samples to form the training data for each tree; for each tree, decision nodes are recursively constructed. At each node, fuzzy logic is used to evaluate and select the contribution of each feature to sample classification, allowing partial overlap and uncertainty to increase the model's ability to handle complex and fuzzy data. The calculation method of information gain based on fuzzy logic at node splitting is expressed as: ； In the formula, is the information gain based on fuzzy logic, which is used to select the split feature se and split threshold te; It is the maximum value operation under the conditions of split feature se and split threshold te; is the fuzzy entropy under the threshold te; For the sample weights; nf is the total number of input samples in the current batch; is the membership function of the category under the threshold te; For sample The total membership of Among them, fuzzy entropy represents fuzzy membership, and the calculation method is expressed as: ; In the formula, The sample belongs to the threshold te. The fuzzy probability of each category is is the set of all categories; S503, in each splitting process of the decision tree, the fuzzy membership is calculated, and the splitting features and points are selected according to the fuzzy membership to optimize the structure of the tree. The calculation method is expressed as: ; In the formula, is the steepness parameter that controls the membership; For sample Kernel function at threshold te; is the smoothness parameter that controls the membership; Kernel Function The calculation method is expressed as: ； In the formula, is the bandwidth parameter of the adaptive kernel function; The calculation method of the bandwidth parameter of the adaptive kernel function is expressed as: ； In the formula, is the regulating factor; is the median calculation function, Representation sample and its neighborhood samples The median of the Euclidean distance between ; S504, during the growth process of the decision tree, split each node until a stop condition is met; S505: After all decision trees stop classifying, it means that the model training is completed, and the output of the trained model is output as the final classification result by voting.

6. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 2, characterized in that: In step S202, the quantum information gain loss function The calculation is performed using the Wasserstein loss based on quantum states, and the calculation method is expressed as: ； In the formula, Express expectations, and represent the real and generated quantum states respectively; is the interpolation between the real and generated quantum states; represents the discriminant function of the quantum discriminator; is the balance coefficient; for Norm.

7. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 2, characterized in that: In step S203, the generated synthetic data Adjustment is made through hybrid kernel density estimation, using the kernel function to more accurately simulate the target data distribution, expressed as: ； In the formula, represents the mixed kernel density function; Indicates The mixing weight of each kernel; Indicates The bandwidth of the kernel function; indicates nuclear center; Representation data No. Features Represents the generated synthetic data The characteristic dimension of Among them, the calculation method of the mixed kernel density function is expressed as: ； In the formula, and It’s data No. The mean and standard deviation of each feature; is the location parameter of the Laplace kernel; is a small positive number that prevents the denominator from being zero.

8. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 3, characterized in that: In step S304, the calculation methods of the weight parameter adjustment amount and the bias parameter adjustment amount obtained based on the fruit fly algorithm are respectively expressed as: ； In the formula, is the search radius of the fruit fly algorithm; Generates random numbers from a standard normal distribution in the interval (0, 1); is a symbolic function; Represents the loss function Gradients with respect to weights; Represents the loss function Gradient with respect to bias.

9. The method for real-time monitoring of steam generator status based on artificial intelligence according to claim 5, characterized in that: In step S502, the fuzzy probability The calculation method is expressed as: ； In the formula, It is a sample Under the threshold te, it belongs to The fuzzy membership of the categories; It is a sample Fuzzy membership under threshold te.

Images