Method and system for identifying operation state of compact five-return type biomass steam boiler

Abstract

A compact five-return-stroke biomass steam boiler operation state identification method and system belong to the technical field of data identification, and the method comprises the following steps: collecting operation data of a biomass steam boiler, and marking the collected data; training the data by adopting a multilayer extreme learning machine algorithm to obtain a state recognition model of the biomass steam boiler; training the local models of the plurality of clients through a federal learning architecture; and performing future state prediction on the running state data of the biomass steam boiler in the continuous time period by adopting a multi-layer neural network. According to the method, federal learning is adopted, and the model is jointly trained through a plurality of distributed client boiler devices or servers, so that data privacy is prevented from being leaked. The multi-layer extreme learning machine based on quantum population evolution simulates selection, crossover and variation processes in classical population evolution through superposition and interference effects of quantum bits, and the classification performance is effectively improved. The multi-layer neural network is adopted to predict the time series data, and the accuracy of state recognition is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of data identification, and in particular relates to an operating status identification method and system for a compact five-pass biomass steam boiler. Background Art

[0002] Biomass steam boiler is a steam boiler that uses biomass energy such as wood chips, straw, rice husks, etc. as fuel. As an important clean energy boiler equipment, biomass steam boiler has broad prospects for use in energy production and industrial applications. Traditional biomass steam boilers have certain technical bottlenecks in structural design, operating efficiency and intelligent management. First of all, traditional biomass steam boilers often adopt a more complex and bulky structural design, resulting in a larger boiler size, more metal consumption, and lower thermal efficiency. In addition, since the boiler needs to continuously and efficiently transfer heat to water or steam, traditional boiler designs often have problems such as insufficient heat exchange area and incomplete combustion, resulting in low thermal efficiency of the boiler and increased energy consumption and operating costs. Therefore, designing a more compact, efficient and easy-to-maintain boiler has become an important direction for improving the performance of biomass steam boilers. The biomass particles are sent to the furnace through a screw feeder from the silo and fall onto the chain grate. There is a rear arch at the rear of the chain grate. The biomass particles fall onto the chain grate and burn on the grate. The high-temperature flue gas passes through the rear arch from front to back, which is the first return of the flue gas. After turning 180°, it goes from back to front, a total of 5 return trips. The heating pipe is composed of a group of curved water pipes, which are interlocked with each other on the left and right, making the entire boiler structure relatively simple and compact, with a larger heat exchange area, a long flue gas flow, more complete combustion, and more complete heat exchange. At the same time, the water volume is relatively small, which makes the startup temperature rise fast and the heat loss small when the furnace is shut down. Compared with traditional biomass boilers, the compact five-pass biomass steam boiler is more compact, simpler, has low metal consumption, high thermal efficiency, and is relatively easy to clean. The air bag on the upper chassis is used for gas separation to improve the quality of steam. Therefore, the five-pass biomass steam boiler will make the flue gas inside the boiler go through five heat exchange processes when it flows through the heat exchange surface, which helps to fully recover and utilize the heat of the exhaust gas after combustion and improve thermal efficiency. Each return can effectively increase the heat exchange area of ​​the boiler, ensure that the heat energy is more fully utilized, and reduce energy waste.

[0003] In terms of operation status monitoring and fault detection, with the popularization of biomass boilers, the operation monitoring system of boilers has become increasingly important. Traditional monitoring systems have the following disadvantages: First, they rely on centralized data collection and analysis methods, which increases the burden of data transmission and poses the risk of privacy leakage during data transmission; Second, these traditional monitoring methods usually rely on data from a single device or node, and cannot effectively integrate the operation data from multiple boiler devices for global optimization, which makes the operation status diagnosis of the boiler often lag behind and difficult to find potential fault hazards, thus affecting the safety and stability of the boiler; Third, although the traditional extreme learning machine algorithm can be trained quickly in some cases, it is easy to fall into the local optimal solution, affecting the classification effect, and the recognition accuracy of the boiler operation status is low; Fourth, the traditional prediction method has poor processing effect on the time series data of the boiler operation status, and cannot fully explore the nonlinear characteristics in the data, and the prediction accuracy of the boiler operation status is low. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a compact five-pass biomass steam boiler operating status identification method that can achieve data privacy protection, collaboration between different boiler equipment, and effectively improve classification performance.

[0005] The technical solution adopted to solve the above technical problems is: a compact five-pass biomass steam boiler operation state identification method, comprising the following steps: S1, collecting the operation data of the biomass steam boiler, annotating the collected data, and the annotated categories include normal state A, minor fault B, serious fault C and abnormal state D; S2, using a multi-layer extreme learning machine algorithm to train the data in S1 to obtain a state identification model of the biomass steam boiler, and optimizing the weights and bias parameters of the model based on the multi-layer extreme learning machine algorithm of quantum swarm evolution; S3, training the local models of multiple clients through a federated learning architecture, each client trains the local model through its local data, and uploads the model parameter updates to the central model, and the central model is aggregated and updated and transmitted back to each client; S4, based on the trained model, the operation state of the steam boiler is identified, and the operation state of the biomass steam boiler at each moment is output. The identification results include normal state A, minor fault B, serious fault C and abnormal state D, and a multi-layer neural network is used to predict the future state of the operation state data of the biomass steam boiler in a continuous time period.

[0006] Furthermore, the operation data of the biomass steam boiler in S1 includes temperature Ra, pressure Da, flow rate Ea, vibration Fa, fuel consumption rate Ga, boiler efficiency Ha, heat loss Ia, flue gas emission Ja, load rate Ka, and water level La.

[0007] Furthermore, the multi-layer extreme learning machine algorithm in S2 includes an input layer, multiple hidden layers and an output layer.

[0008] Furthermore, the training process of the multi-layer extreme learning machine algorithm based on quantum swarm evolution in S2 includes the following steps: S201, initializing the initial weights and biases of the multi-layer extreme learning machine by quantum state representation, setting the initial quantum state to: , In the formula, represents the initial quantum state, Represents the amplitude coefficient of the quantum state, satisfying the normalization condition: , represents the quantum ground state, Represents the number of qubits, which is equal to the dimension of the input data features of the multi-layer extreme learning machine; The initial weights and biases of the multi-layer extreme learning machine are obtained through quantum measurement, and the mapping relationship between quantum states and classical data is: , In the formula, represents the weights of the multi-layer extreme learning machine, represents the function mapping after quantum measurement, Represents the input features of the multi-layer extreme learning machine, and the corresponding weight matrix is ​​obtained after quantum mapping. represents the conjugation of the initial quantum bit state; Function mapping after quantum measurement The initial weights of the multi-layer extreme learning machine are obtained through quantum measurement. The specific mapping process is obtained by calculating the inner product of the quantum state and the input feature. At the same time, the quantum state is weighted to assign a different influence to each sample. The calculation method is expressed as: , In the formula, Indicates the number of samples input to the multi-layer extreme learning machine in the current batch, represents the i-th input feature of the multi-layer extreme learning machine, represents the amplitude coefficient of the quantum bit state; S202, using quantum gate operation to update the quantum state of each parameter, and updating the quantum state through the quantum population evolution method to simulate the selection, crossover and mutation process in the classical population evolution. Each iteration of the quantum state update is realized through quantum gate operation, and the quantum gate operation is expressed as: , In the formula, represents the quantum bit state of the tth iteration, represents the quantum gate operation matrix, represents the quantum bit state of the t-1th iteration, The parameters of the quantum gate represent the rotation angle of the quantum bit. Through multiple quantum gate operations, the quantum bit state is continuously updated to find the global optimal solution. S203. In each iteration, the quantum state is converted into classical information, i.e., the weights and bias values ​​of the multi-layer extreme learning machine, through quantum measurement, and its performance is evaluated on the training data set. The output of the extreme learning machine is expressed as: , In the formula, represents the prediction output of the multi-layer extreme learning machine, Represents the activation function of the multi-layer extreme learning machine. The activation function of the multi-layer extreme learning machine adopts the Sigmoid activation function. represents the weight of the multilayer extreme learning machine at the t-th iteration, obtained by performing quantum measurement operations on the quantum state at the t-th iteration, represents the weights of the multi-layer extreme learning machine; The way a quantum measurement operation converts a quantum bit state into classical information is expressed as: , In the formula, represents the weight of the multilayer extreme learning machine at the tth iteration The i-th element of represents the conjugation of the quantum bit state at the tth iteration; S204, adopting a quantum weight dynamic adjustment strategy to enhance the dynamic adaptability of the quantum population evolution process. The weight update process of the multi-layer extreme learning machine depends on the input features and quantum states. The weight of the multi-layer extreme learning machine is dynamically adjusted by calculating an adaptive factor that depends on the fitness value of all solutions in the population. The calculation method is expressed as: , In the formula, Represents the mean of the current group fitness, characterizing the performance of the entire group. represents the exponential function, Represents the adjustment factor, which controls the sensitivity of the adaptive factor. Set to 2, represents the adaptive factor; S205, the dynamic constraint function is continuously updated as the algorithm iteration process proceeds. The dynamic constraint function represents the constraint strength of the solution space in the current optimization stage, and the calculation method is expressed as: , In the formula, represents the dynamic constraint function of the tth iteration. As the iteration proceeds, the constraint strength gradually increases and the amplitude of the variation gradually decreases, ensuring the stability and concentration of the variation in the later stages. represents the first parameter of the constraint function, represents the second parameter of the constraint function, Indicates the current iteration number; S206. In order to maintain the stability of the algorithm, each iteration will select a certain proportion of the optimal solutions from the current population and retain their quantum states to the next generation to ensure that excellent characteristics can be inherited, expressed as: , In the formula, represents the quantum bit state of the t+1th iteration, Indicates that the solution with the smallest fitness is selected as the retained quantum state and enters the t+1th iteration. represents the fitness of the ith solution, represents a tuning parameter that controls the strength of preserving solutions, represents the penalty term of the solution, indicating the degree of failure of the solution, Indicates variable asynchronous length. Set to 0.1, represents the mutation matrix of the ith qubit in the tth generation, represents the dynamic constraint function of the tth generation. In the early stage of optimization, the constraint function is small, allowing the qubits to mutate in a larger range and promote population diversity. In the later stage of optimization, the constraint function approaches , and the variation range gradually decreases, making the algorithm focus on the refined search for the global optimal solution; The penalty term of the solution is used to avoid overfitting. It quantifies the difference between the model output and the true label to ensure that the selected solution has better generalization ability. The calculation method is expressed as: , In the formula, represents the input features of the i-th sample, express norm; S207. The weight matrix of the multi-layer extreme learning machine depends on the inner product of the input data and the quantum state, and is updated according to the influence of the interference effect on the weight. The update method is expressed as: , In the formula, represents the weight of the multilayer extreme learning machine at the t+1th iteration, represents the weight of the multilayer extreme learning machine at the t-th iteration, obtained by performing quantum measurement operations on the quantum state at the t-th iteration, Represents the learning rate of the multi-layer extreme learning machine, which controls the step size of weight update. Set to 0.01, represents the quantum interference term; The quantum interference term plays an optimization role in the parameter update process of the multi-layer extreme learning machine. Through the interference coefficient, each round of iteration in the quantum group evolution process can find the optimal solution more accurately. The calculation method is expressed as: , In the formula, The interference coefficient of the kth feature represents the interference strength of each quantum gate and is a training parameter; Based on the gradient information and the fitness of the model, the weights of the multi-layer extreme learning machine are dynamically optimized through local perturbations. The weight update increment calculation method of the multi-layer extreme learning machine is expressed as: , In the formula, represents the weight update increment term of the t-th iteration of the multilayer extreme learning machine, represents the learning rate of the incremental term of weight update, Set to 0.2, represents the partial derivative of the fitness function with respect to the weights of the multilayer extreme learning machine at the t-th iteration, represents the adjustment factor used to balance the effect of the fitness function on weight updates, Set to 0.4, represents the perturbation intensity factor, which is used to control the impact of the perturbation on the weight update. Set to 0.3, Represents a random perturbation term, which is used to generate local randomness, avoid falling into local minima during training, and enhance exploration; S208, repeat the above steps until a preset stop iteration condition is met. The preset stop iteration condition is reaching a preset maximum number of iterations, which means that the model training is completed.

[0009] Furthermore, in S203, the fitness of each quantum solution is evaluated, that is, the performance of the model on the training set is calculated, and the fitness function is: , In the formula, Represents the fitness function of the multi-layer extreme learning machine. The smaller the fitness value calculated according to the fitness function, the better the classification effect of the solution. Indicates the number of samples input to the multi-layer extreme learning machine in the current batch, represents the predicted value of the multi-layer extreme learning machine for the i-th sample, represents the true label of the i-th sample.

[0010] Furthermore, in order to improve the flexibility of the mutation operation in S205, the mutation matrix of each qubit is calculated, and the calculation method is expressed as: , In the formula, Indicates the type of variation, Set to 4, Represents the adaptive mutation adjustment factor, which is used to adjust the impact of the k-th mutation type on the current qubit. represents the kth mutation matrix, which is randomly generated and represents a certain mutation method for quantum bits; The adaptive mutation adjustment factor is dynamically adjusted according to the distribution of the current group fitness, and the calculation method is expressed as: , In the formula, Represents the adjustment parameter of the adaptive mutation adjustment factor, which controls the sensitivity of each mutation mode. Set to 2, represents the fitness of the i-th qubit under the k-th mutation mode, Represents the mean fitness of the kth mutation method.

[0011] Furthermore, the federated learning architecture in S3 includes multiple clients, each client trains a local model through its local data and interacts with the central model, and model parameters are exchanged and updated between the client and the central model.

[0012] Furthermore, the loss function calculation method of the multi-layer neural network in S4 is expressed as: , In the formula, represents the loss function of a multi-layer neural network, The dimension representing the steam boiler operating status data at continuous moments, Represents the multi-layer neural network input at the tth time point Response, The characteristic value at the t-th time point corresponds to the operating state of the steam boiler, such as normal state A, minor fault B, serious fault C, and abnormal state D; Considering the multi-layer structure and activation of the multi-layer neural network, the response calculation is expressed as: , In the formula, represents the activation function of the output layer, and Represent the weights and biases of the output layer of the multi-layer neural network, Represents the first The manifold mapping function of the layer, Represents the first time point of the multilayer neural network The output of the layer; The random manifold learning strategy is used to identify the nonlinear structure in the steam boiler operating status data at continuous moments, and then more complex data patterns are extracted. Manifold learning is performed on each dimension of the output of the multi-layer neural network in order to discover the nonlinear structure of the data, which is expressed as: , In the formula, Indicates The manifold mapping function of the layer realizes manifold learning through Taylor series. represents the first parameter of manifold learning, Represents the Taylor expansion of the nonlinear activation function, which is used to fit the characteristics of highly nonlinear data. represents the second parameter of manifold learning, represents the third parameter of manifold learning, the first parameter of manifold learning, the second parameter of manifold learning and the third parameter of manifold learning are all training parameters; The nonlinear activation function is expressed using Taylor series expansion as follows: , In the formula, represents the series of Taylor expansion, z represents the input of the nonlinear activation function, corresponding to , a represents the expansion point of the Taylor series, represents the nth derivative of the function at point a, represents the factorial of the nth order of Taylor expansion; Will Bring in In the expression of , we can get: , In the formula, the nonlinear activation function is set to the Sigmoid activation function; The learning rate is dynamically adjusted according to the loss function of the multi-layer neural network. The way to dynamically adjust the learning rate is expressed as: , In the formula, represents the learning rate of the adjusted multi-layer neural network, represents the learning rate of the multi-layer neural network before adjustment, represents the adjustment coefficient of the dynamic learning rate, Indicates the change in the loss function of the current iteration; Adjust the weights and bias parameters of the multi-layer neural network. The adjustment method is expressed as: , In the formula, Represents the updated multi-layer neural network The weight of the layer, represents the first The weight of the layer, Represents the updated multi-layer neural network The bias of the layer, represents the first The bias of the layer, represents the sign of partial derivative; Repeat the above steps, and the preset stopping condition is to reach the preset maximum number of iterations, until the preset stopping condition is met, which means that the multi-layer neural network model training is completed.

[0013] Compact five-pass biomass steam boiler operation status identification system, including: The data acquisition module is used to collect the operating data of the steam boiler, including the boiler's temperature Ra, pressure Da, flow Ea, vibration Fa, fuel consumption rate Ga, boiler efficiency Ha, heat loss Ia, flue gas emission Ja, load rate Ka, and water level La; A data storage module, used for storing collected operation data; The data labeling module is used to manually label the data and label the operating status of the boiler as normal status A, minor fault B, serious fault C, and abnormal status D; The model training module trains the labeled data based on the multi-layer extreme learning machine algorithm to obtain the steam boiler state recognition model; The federated learning module is used to implement model training and parameter exchange of multiple clients, support the federated learning architecture, and ultimately obtain a global model; The state recognition module performs real-time steam boiler operation state recognition based on the trained model and outputs the boiler state category.

[0014] The beneficial effects of the present invention are as follows: (1) The present invention adopts federated learning to jointly train the model through multiple distributed client boiler devices or servers, while ensuring that data privacy is not leaked. There is no need to exchange data between the clients, and the model parameters are aggregated on the central server, thereby achieving data privacy protection. It can collaborate between different boiler devices, avoiding the security risks of data aggregation in the traditional centralized training process, and achieving more flexible and efficient model training.

[0015] (2) The present invention adopts a multi-layer extreme learning machine based on quantum population evolution. Through the superposition and interference effects of quantum bits, it can simulate the selection, crossover and mutation processes in classical population evolution, effectively improve the classification performance, and avoid the local optimality and overfitting problems common in traditional machine learning models.

[0016] (3) The present invention adopts a multi-layer neural network to predict time series data, which improves the accuracy of state recognition. By using the manifold learning strategy and Taylor expansion to analyze the nonlinear activation function, the complex nonlinear structure in the boiler operation status data can be modeled, the potential laws of the data can be effectively extracted, and the prediction accuracy can be enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a structural diagram of the federated learning architecture of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] The five-pass biomass steam boiler operation status identification system includes: The data acquisition module is used to collect the operating data of the steam boiler, including the boiler's temperature Ra, pressure Da, flow Ea, vibration Fa, fuel consumption rate Ga, boiler efficiency Ha, heat loss Ia, flue gas emission Ja, load rate Ka, and water level La.

[0020] The data storage module is used to store the collected operation data.

[0021] The data labeling module is used to manually label the data and mark the operating status of the boiler as normal status A, minor fault B, serious fault C and abnormal status D.

[0022] The model training module trains the labeled data based on the multi-layer extreme learning machine algorithm to obtain the steam boiler state recognition model.

[0023] The federated learning module is used to implement model training and parameter exchange for multiple clients, support the federated learning architecture, and ultimately obtain a global model.

[0024] The state recognition module performs real-time steam boiler operation state recognition based on the trained model and outputs the boiler state category.

[0025] The operating state identification method of a compact five-pass biomass steam boiler comprises the following steps: S1, collect the operation data of the biomass steam boiler and mark the collected data. The marked categories include normal state A, minor fault B, serious fault C and abnormal state D.

[0026] The operation data of the biomass steam boiler include temperature Ra, pressure Da, flow rate Ea, vibration Fa, fuel consumption rate Ga, boiler efficiency Ha, heat loss Ia, flue gas emission Ja, load rate Ka, and water level La.

[0027] S2, uses a multi-layer extreme learning machine algorithm to train the data in S1 to obtain the state recognition model of the biomass steam boiler, and optimizes the weights and bias parameters of the model based on the multi-layer extreme learning machine algorithm based on quantum swarm evolution.

[0028] The multi-layer extreme learning machine algorithm includes an input layer, multiple hidden layers and an output layer. Quantum swarm evolution uses quantum states to represent the weights and bias parameters of the multi-layer extreme learning machine, so that the multi-layer extreme learning machine algorithm can find the global optimal solution through the superposition and interference of quantum states during the iteration process, thereby avoiding the problem that traditional extreme learning machines are prone to falling into local optimality and improving classification performance.

[0029] The training process of the multi-layer extreme learning machine algorithm based on quantum swarm evolution includes the following steps: S201. In this embodiment, the initial weights and biases of the multi-layer extreme learning machine are initialized by quantum state representation, and the initial quantum state is set to: , In the formula, represents the initial quantum state, Represents the amplitude coefficient of the quantum state, satisfying the normalization condition: , represents the quantum ground state, Represents the number of qubits, which is equal to the dimension of the input data features of the multi-layer extreme learning machine.

[0030] Traditional extreme learning machine algorithms usually use random initialization to initialize the parameters of the extreme learning machine.

[0031] The initial weights and biases of the multi-layer extreme learning machine are obtained through quantum measurement, and the mapping relationship between quantum states and classical data is: , In the formula, represents the weights of the multi-layer extreme learning machine, represents the function mapping after quantum measurement, Represents the input features of the multi-layer extreme learning machine, and the corresponding weight matrix is ​​obtained after quantum mapping. represents the conjugation of the initial quantum state.

[0032] Function mapping after quantum measurement The initial weights of the multi-layer extreme learning machine are obtained through quantum measurement. The specific mapping process is obtained by calculating the inner product of the quantum state and the input feature. At the same time, the quantum state is weighted to assign a different influence to each sample. The calculation method is expressed as: , In the formula, Indicates the number of samples input to the multi-layer extreme learning machine in the current batch, represents the i-th input feature of the multi-layer extreme learning machine, represents the amplitude coefficient of the quantum state.

[0033] S202, using quantum gate operation to update the quantum state of each parameter, and updating the quantum state through the quantum population evolution method to simulate the selection, crossover and mutation process in the classical population evolution. Each iteration of the quantum state update is realized through quantum gate operation, and the quantum gate operation is expressed as: , In the formula, represents the quantum bit state of the tth iteration, represents the quantum gate operation matrix, represents the quantum bit state of the t-1th iteration, The parameters of the quantum gate represent the rotation angle of the quantum bit. Through multiple operations of the quantum gate, the quantum bit state is continuously updated to find the global optimal solution.

[0034] S203. In each iteration, the quantum state is converted into classical information, i.e., the weights and bias values ​​of the multi-layer extreme learning machine, through quantum measurement, and its performance is evaluated on the training data set. The output of the extreme learning machine is expressed as: , In the formula, represents the prediction output of the multi-layer extreme learning machine, Represents the activation function of the multi-layer extreme learning machine. The activation function of the multi-layer extreme learning machine adopts the Sigmoid activation function. represents the weight of the multilayer extreme learning machine at the t-th iteration, obtained by performing quantum measurement operations on the quantum state at the t-th iteration, Represents the weights of a multi-layer extreme learning machine.

[0035] The way a quantum measurement operation converts a quantum bit state into classical information is expressed as: , In the formula, represents the weight of the multilayer extreme learning machine at the tth iteration The i-th element of represents the conjugation of the qubit state at the tth iteration.

[0036] In order to evaluate the quality of each quantum solution, its fitness is calculated, that is, the performance of the model on the training set. The fitness function is: , In the formula, Represents the fitness function of the multi-layer extreme learning machine. The smaller the fitness value calculated according to the fitness function, the better the classification effect of the solution. Indicates the number of samples input to the multi-layer extreme learning machine in the current batch, represents the predicted value of the multi-layer extreme learning machine for the i-th sample, represents the true label of the i-th sample.

[0037] S204, adopting quantum weight dynamic adjustment strategy to enhance the dynamic adaptability of quantum group evolution process, avoiding overfitting and local optimal problems in weight update process, the weight update process of multi-layer extreme learning machine depends on input features and quantum states, and dynamically adjusts the weight of multi-layer extreme learning machine by calculating adaptive factors that depend on the fitness values ​​of all solutions in the group, and the calculation method is expressed as: , In the formula, Represents the mean of the current group fitness, characterizing the performance of the entire group. represents the exponential function, Represents the adjustment factor, which controls the sensitivity of the adaptive factor. Set to 2, Represents the adaptive factor.

[0038] S205. Dynamically adjust the mutation strategy according to the constraint changes in the current optimization process, and make precise adjustments according to the optimization requirements at different stages. In the early stage of the evolution process, focus on diversity, and in the later stage, enhance the focus on the global optimal solution, thereby accelerating the convergence of the multi-layer extreme learning machine and improving the accuracy of the solution. The dynamic constraint function is continuously updated as the algorithm iterative process proceeds. The dynamic constraint function represents the constraint strength of the solution space in the current optimization stage, and the calculation method is expressed as: , In the formula, represents the dynamic constraint function of the tth iteration. As the iteration proceeds, the constraint strength gradually increases and the amplitude of the variation gradually decreases, ensuring the stability and concentration of the variation in the later stages. represents the first parameter of the constraint function, represents the second parameter of the constraint function, Indicates the current iteration number; In order to improve the flexibility of mutation operations, the mutation matrix of each qubit is calculated. The calculation method is expressed as: , In the formula, Indicates the type of variation, Set to 4, Represents the adaptive mutation adjustment factor, which is used to adjust the impact of the k-th mutation type on the current qubit. Represents the kth mutation matrix, which is randomly generated and represents a certain mutation method for quantum bits.

[0039] The adaptive mutation adjustment factor is dynamically adjusted according to the distribution of the current group fitness, and the calculation method is expressed as: , In the formula, Represents the adjustment parameter of the adaptive mutation adjustment factor, which controls the sensitivity of each mutation mode. Set to 2, represents the fitness of the i-th qubit under the k-th mutation mode, Represents the mean fitness of the kth mutation method.

[0040] S206. In order to maintain the stability of the algorithm, each iteration will select a certain proportion of the optimal solutions from the current population and retain their quantum states to the next generation to ensure that excellent characteristics can be inherited, expressed as: , In the formula, represents the quantum bit state of the t+1th iteration, Indicates that the solution with the smallest fitness is selected as the retained quantum state and enters the t+1th iteration. represents the fitness of the ith solution, represents a tuning parameter that controls the strength of preserving solutions, represents the penalty term of the solution, indicating the degree of failure of the solution, Indicates variable asynchronous length. Set to 0.1, represents the mutation matrix of the ith qubit in the tth generation, Represents the dynamic constraint function of the tth generation. In the early stage of optimization, the constraint function is small, allowing the qubits to mutate in a larger range and promote population diversity. In the later stage of optimization, the constraint function approaches , and the variation amplitude gradually decreases, allowing the algorithm to focus on the refined search for the global optimal solution.

[0041] The penalty term of the solution is used to avoid overfitting. It quantifies the difference between the model output and the true label to ensure that the selected solution has better generalization ability. The calculation method is expressed as: , In the formula, represents the input features of the i-th sample, express norm; S207. The weight matrix of the multi-layer extreme learning machine depends on the inner product of the input data and the quantum state, and is updated according to the influence of the interference effect on the weight. The update method is expressed as: , In the formula, represents the weight of the multilayer extreme learning machine at the t+1th iteration, represents the weight of the multilayer extreme learning machine at the t-th iteration, obtained by performing quantum measurement operations on the quantum state at the t-th iteration, Represents the learning rate of the multi-layer extreme learning machine, which controls the step size of weight update. Set to 0.01, represents the quantum interference term.

[0042] The quantum interference term plays an optimization role in the parameter update process of the multi-layer extreme learning machine. Through the interference coefficient, each round of iteration in the quantum group evolution process can find the optimal solution more accurately. The calculation method is expressed as: , In the formula, represents the interference coefficient of the kth feature, represents the interference strength of each quantum gate, and is a training parameter.

[0043] Based on the gradient information and the fitness of the model, the weights of the multi-layer extreme learning machine are dynamically optimized through local perturbations. The weight update increment calculation method of the multi-layer extreme learning machine is expressed as: , In the formula, represents the weight update increment term of the t-th iteration of the multilayer extreme learning machine, represents the learning rate of the incremental term of weight update, Set to 0.2, represents the partial derivative of the fitness function with respect to the weights of the multilayer extreme learning machine at the t-th iteration, represents the adjustment factor used to balance the effect of the fitness function on weight updates, Set to 0.4, represents the perturbation intensity factor, which is used to control the impact of the perturbation on the weight update. Set to 0.3, Represents a random perturbation term, which is used to generate local randomness, avoid falling into local minima during training, and enhance exploration; S208. Repeat the above steps until a preset stop iteration condition is met. The preset stop iteration condition is reaching a preset maximum number of iterations. The preset maximum number of iterations is set to 1000 times, which means that the model training is completed.

[0044] S3 trains the local models of multiple clients through a federated learning architecture. Each client trains the local model with its local data and uploads the model parameter updates to the central model. The central model is aggregated, updated, and sent back to each client.

[0045] The federated learning architecture includes multiple clients. Each client trains a local model through its local data and interacts with the central model. Model parameters are exchanged and updated between the client and the central model. The federated learning architecture of this embodiment has multiple clients, labeled as client 0, client 1, client 2, and client 3, etc. These clients represent independent nodes in the federated learning network. Each client has its own local data, such as Figure 1 shown.

[0046] S4, based on the trained model, the operating status of the steam boiler is identified, and the operating status of the biomass steam boiler at each moment is output. The identification results include normal state A, minor fault B, serious fault C and abnormal state D. A multi-layer neural network is used to predict the future state of the operating status data of the biomass steam boiler in continuous time periods.

[0047] The loss function calculation method of the multi-layer neural network is expressed as: , In the formula, represents the loss function of a multi-layer neural network, The dimension representing the steam boiler operating status data at continuous moments, Represents the multi-layer neural network input at the tth time point Response, It represents the characteristic value at the t-th time point, corresponding to the operating state of the steam boiler, such as normal state A, minor fault B, serious fault C and abnormal state D.

[0048] Considering the multi-layer structure and activation of the multi-layer neural network, the response calculation is expressed as: , In the formula, represents the activation function of the output layer, and Represent the weights and biases of the output layer of the multi-layer neural network, Represents the first The manifold mapping function of the layer, Represents the first time point of the multilayer neural network The output of the layer.

[0049] The random manifold learning strategy is used to identify the nonlinear structure in the steam boiler operating status data at continuous moments, and then more complex data patterns are extracted. Manifold learning is performed on each dimension of the output of the multi-layer neural network in order to discover the nonlinear structure of the data, which is expressed as: , In the formula, Indicates The manifold mapping function of the layer realizes manifold learning through Taylor series. represents the first parameter of manifold learning, Represents the Taylor expansion of the nonlinear activation function, which is used to fit the characteristics of highly nonlinear data. represents the second parameter of manifold learning, represents the third parameter of manifold learning. The first parameter of manifold learning, the second parameter of manifold learning and the third parameter of manifold learning are all training parameters.

[0050] The nonlinear activation function is expressed using Taylor series expansion as follows: , In the formula, represents the series of Taylor expansion, z represents the input of the nonlinear activation function, corresponding to , a represents the expansion point of the Taylor series, represents the nth derivative of the function at point a, represents the nth order factorial of the Taylor expansion.

[0051] Will Bring in In the expression of , we can get: , In the formula, the nonlinear activation function is set to the Sigmoid activation function.

[0052] The learning rate is dynamically adjusted according to the loss function of the multi-layer neural network. If it is detected that the model is not sensitive to new data changes, the learning rate is increased, otherwise it is reduced. The way to dynamically adjust the learning rate is expressed as: , In the formula, represents the learning rate of the adjusted multi-layer neural network, represents the learning rate of the multi-layer neural network before adjustment, represents the adjustment coefficient of the dynamic learning rate, Indicates the change in the loss function of the current iteration.

[0053] Adjust the weights and bias parameters of the multi-layer neural network. The adjustment method is expressed as: , In the formula, Represents the updated multi-layer neural network The weight of the layer, represents the first The weight of the layer, Represents the updated multi-layer neural network The bias of the layer, represents the first The bias of the layer, Represents the sign of partial derivative.

[0054] Repeat the above steps, and the preset stop iteration condition is to reach the preset maximum number of iterations, which is set to 1000 times, until the preset stop iteration condition is met, which means that the multi-layer neural network model training is completed.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A method for identifying the operating status of a compact five-pass biomass steam boiler, characterized in that: The following steps are involved: S1, collect the operation data of the biomass steam boiler and mark the collected data. The marked categories include normal state A, minor fault B, serious fault C and abnormal state D; S2, using a multi-layer extreme learning machine algorithm to train the data in S1 to obtain a state recognition model of a biomass steam boiler, and optimizing the weights and bias parameters of the model using a multi-layer extreme learning machine algorithm based on quantum swarm evolution; S3, trains the local models of multiple clients through the federated learning architecture. Each client trains the local model with its local data and uploads the model parameter updates to the central model. The central model is aggregated, updated and sent back to each client. S4, based on the trained model, the operating status of the steam boiler is identified, and the operating status of the biomass steam boiler at each moment is output. The identification results include normal state A, minor fault B, serious fault C and abnormal state D. A multi-layer neural network is used to predict the future state of the operating status data of the biomass steam boiler in continuous time periods.

2. The method for identifying the operating status of a compact five-pass biomass steam boiler according to claim 1 is characterized in that: The operation data of the biomass steam boiler in S1 include temperature Ra, pressure Da, flow Ea, vibration Fa, fuel consumption rate Ga, boiler efficiency Ha, heat loss Ia, flue gas emission Ja, load rate Ka, and water level La.

3. The method for identifying the operating status of a compact five-pass biomass steam boiler according to claim 1 is characterized in that: The multi-layer extreme learning machine algorithm in S2 includes an input layer, multiple hidden layers and an output layer.

4. The method for identifying the operating status of a compact five-pass biomass steam boiler according to claim 1, characterized in that: The training process of the multi-layer extreme learning machine algorithm based on quantum swarm evolution in S2 includes the following steps: S201, initializing the initial weights and biases of the multi-layer extreme learning machine by quantum state representation, setting the initial quantum state to: ， In the formula, represents the initial quantum state, Represents the amplitude coefficient of the quantum state, satisfying the normalization condition: , represents the quantum ground state, Represents the number of qubits, which is equal to the dimension of the input data features of the multi-layer extreme learning machine; The initial weights and biases of the multi-layer extreme learning machine are obtained through quantum measurement, and the mapping relationship between quantum states and classical data is: ， In the formula, represents the weights of the multi-layer extreme learning machine, represents the function mapping after quantum measurement, Represents the input features of the multi-layer extreme learning machine, and the corresponding weight matrix is ​​obtained after quantum mapping. represents the conjugation of the initial quantum bit state; Function mapping after quantum measurement The initial weights of the multi-layer extreme learning machine are obtained through quantum measurement. The specific mapping process is obtained by calculating the inner product of the quantum state and the input feature. At the same time, the quantum state is weighted to assign a different influence to each sample. The calculation method is expressed as: ， In the formula, Indicates the number of samples input to the multi-layer extreme learning machine in the current batch, represents the i-th input feature of the multi-layer extreme learning machine, represents the amplitude coefficient of the quantum bit state; S202, using quantum gate operation to update the quantum state of each parameter, and updating the quantum state through the quantum population evolution method to simulate the selection, crossover and mutation process in the classical population evolution. Each iteration of the quantum state update is realized through quantum gate operation, and the quantum gate operation is expressed as: ， In the formula, represents the quantum bit state of the tth iteration, represents the quantum gate operation matrix, represents the quantum bit state of the t-1th iteration, The parameters of the quantum gate represent the rotation angle of the quantum bit. Through multiple quantum gate operations, the quantum bit state is continuously updated to find the global optimal solution. S203. In each iteration, the quantum state is converted into classical information, i.e., the weights and bias values ​​of the multi-layer extreme learning machine, through quantum measurement, and its performance is evaluated on the training data set. The output of the extreme learning machine is expressed as: ， In the formula, represents the prediction output of the multi-layer extreme learning machine, Represents the activation function of the multi-layer extreme learning machine. The activation function of the multi-layer extreme learning machine adopts the Sigmoid activation function. represents the weight of the multilayer extreme learning machine at the t-th iteration, obtained by performing quantum measurement operations on the quantum state at the t-th iteration, represents the weights of the multi-layer extreme learning machine; The way a quantum measurement operation converts a quantum bit state into classical information is expressed as: ， In the formula, represents the weight of the multilayer extreme learning machine at the tth iteration The i-th element of represents the conjugation of the quantum bit state at the tth iteration; S204, adopting a quantum weight dynamic adjustment strategy to enhance the dynamic adaptability of the quantum population evolution process. The weight update process of the multi-layer extreme learning machine depends on the input features and quantum states. The weight of the multi-layer extreme learning machine is dynamically adjusted by calculating an adaptive factor that depends on the fitness value of all solutions in the population. The calculation method is expressed as: ， In the formula, Represents the mean of the current group fitness, characterizing the performance of the entire group. represents the exponential function, Represents the adjustment factor, which controls the sensitivity of the adaptive factor. Set to 2, represents the adaptive factor; S205, the dynamic constraint function is continuously updated as the algorithm iteration process proceeds. The dynamic constraint function represents the constraint strength of the solution space in the current optimization stage, and the calculation method is expressed as: ， In the formula, represents the dynamic constraint function of the tth iteration. As the iteration proceeds, the constraint strength gradually increases and the amplitude of the variation gradually decreases, ensuring the stability and concentration of the variation in the later stages. represents the first parameter of the constraint function, represents the second parameter of the constraint function, Indicates the current iteration number; S206. In order to maintain the stability of the algorithm, each iteration will select a certain proportion of the optimal solutions from the current population and retain their quantum states to the next generation to ensure that excellent characteristics can be inherited, expressed as: ， In the formula, represents the quantum bit state of the t+1th iteration, Indicates that the solution with the smallest fitness is selected as the retained quantum state and enters the t+1th iteration. represents the fitness of the ith solution, represents a tuning parameter that controls the strength of the retained solution, represents the penalty term of the solution, indicating the degree of failure of the solution, Indicates variable asynchronous length. Set to 0.1, represents the mutation matrix of the ith qubit in the tth generation, represents the dynamic constraint function of the tth generation. In the early stage of optimization, the constraint function is small, allowing the qubits to mutate in a larger range and promote population diversity. In the later stage of optimization, the constraint function approaches , and the variation range gradually decreases, making the algorithm focus on the refined search for the global optimal solution; The penalty term of the solution is used to avoid overfitting. It quantifies the difference between the model output and the true label to ensure that the selected solution has better generalization ability. The calculation method is expressed as: ， In the formula, represents the input features of the i-th sample, express norm; S207. The weight matrix of the multi-layer extreme learning machine depends on the inner product of the input data and the quantum state, and is updated according to the influence of the interference effect on the weight. The update method is expressed as: ， In the formula, represents the weight of the multilayer extreme learning machine at the t+1th iteration, represents the weight of the multilayer extreme learning machine at the t-th iteration, obtained by performing quantum measurement operations on the quantum state at the t-th iteration, Represents the learning rate of the multi-layer extreme learning machine, which controls the step size of weight update. Set to 0.01, represents the quantum interference term; The quantum interference term plays an optimization role in the parameter update process of the multi-layer extreme learning machine. Through the interference coefficient, each round of iteration in the quantum group evolution process can find the optimal solution more accurately. The calculation method is expressed as: ， In the formula, represents the interference coefficient of the kth feature, represents the interference strength of each quantum gate, and is a training parameter; Based on the gradient information and the fitness of the model, the weights of the multi-layer extreme learning machine are dynamically optimized through local perturbations. The weight update increment calculation method of the multi-layer extreme learning machine is expressed as: ， In the formula, represents the weight update increment term of the t-th iteration of the multilayer extreme learning machine, represents the learning rate of the weight update increment term, Set to 0.2, represents the partial derivative of the fitness function with respect to the weights of the multilayer extreme learning machine at the t-th iteration, represents the adjustment factor used to balance the effect of the fitness function on weight updates, Set to 0.4, represents the perturbation intensity factor, which is used to control the impact of the perturbation on the weight update. Set to 0.3, Represents a random perturbation term, which is used to generate local randomness, avoid falling into local minima during training, and enhance exploration; S208, repeat the above steps until a preset stop iteration condition is met. The preset stop iteration condition is reaching a preset maximum number of iterations, which means that the model training is completed.

5. The method for identifying the operating status of a compact five-pass biomass steam boiler according to claim 4 is characterized in that: In S203, the fitness of each quantum solution is evaluated, that is, the performance of the model on the training set is calculated. The fitness function is: ， In the formula, Represents the fitness function of the multi-layer extreme learning machine. The smaller the fitness value calculated according to the fitness function, the better the classification effect of the solution. Indicates the number of samples input to the multi-layer extreme learning machine in the current batch, represents the predicted value of the multi-layer extreme learning machine for the i-th sample, represents the true label of the i-th sample.

6. The method for identifying the operating status of a compact five-pass biomass steam boiler according to claim 4 is characterized in that: In order to improve the flexibility of the mutation operation in S205, the mutation matrix of each qubit is calculated, and the calculation method is expressed as: ， In the formula, Indicates the type of variation, Set to 4, Represents the adaptive mutation adjustment factor, which is used to adjust the impact of the k-th mutation type on the current qubit. represents the kth mutation matrix, which is randomly generated and represents a certain mutation method for quantum bits; The adaptive mutation adjustment factor is dynamically adjusted according to the distribution of the current group fitness, and the calculation method is expressed as: ， In the formula, Represents the adjustment parameter of the adaptive mutation adjustment factor, which controls the sensitivity of each mutation mode. Set to 2, represents the fitness of the i-th qubit under the k-th mutation mode, Represents the mean fitness of the kth mutation method.

7. The method for identifying the operating status of a compact five-pass biomass steam boiler according to claim 1, characterized in that: The federated learning architecture in S3 includes multiple clients, each of which trains a local model through its local data and interacts with the central model. Model parameters are exchanged and updated between the client and the central model.

8. The method for identifying the operating status of a compact five-pass biomass steam boiler according to claim 1, characterized in that: The loss function calculation method of the multi-layer neural network in S4 is expressed as: ， In the formula, represents the loss function of a multi-layer neural network, The dimension representing the steam boiler operating status data at continuous moments, Represents the multi-layer neural network input at the tth time point Response, The characteristic value at the t-th time point corresponds to the operating state of the steam boiler, such as normal state A, minor fault B, serious fault C, and abnormal state D; Considering the multi-layer structure and activation of the multi-layer neural network, the response calculation is expressed as: ， In the formula, represents the activation function of the output layer, and Represent the weights and biases of the output layer of the multi-layer neural network, Represents the first The manifold mapping function of the layer, Represents the first time point of the multilayer neural network The output of the layer; The random manifold learning strategy is used to identify the nonlinear structure in the steam boiler operating status data at continuous moments, and then more complex data patterns are extracted. Manifold learning is performed on each dimension of the output of the multi-layer neural network in order to discover the nonlinear structure of the data, which is expressed as: ， In the formula, Indicates The manifold mapping function of the layer realizes manifold learning through Taylor series. represents the first parameter of manifold learning, Represents the Taylor expansion of the nonlinear activation function, which is used to fit the characteristics of highly nonlinear data. represents the second parameter of manifold learning, represents the third parameter of manifold learning, the first parameter of manifold learning, the second parameter of manifold learning and the third parameter of manifold learning are all training parameters; The nonlinear activation function is expressed using Taylor series expansion as follows: ， In the formula, represents the series of Taylor expansion, z represents the input of the nonlinear activation function, corresponding to , a represents the expansion point of the Taylor series, represents the nth derivative of the function at point a, represents the factorial of the nth order of Taylor expansion; Will Bring in In the expression of , we can get: ， In the formula, the nonlinear activation function is set to the Sigmoid activation function; The learning rate is dynamically adjusted according to the loss function of the multi-layer neural network. The way to dynamically adjust the learning rate is expressed as: ， In the formula, represents the learning rate of the adjusted multi-layer neural network, represents the learning rate of the multi-layer neural network before adjustment, represents the adjustment coefficient of the dynamic learning rate, Indicates the change in the loss function of the current iteration; Adjust the weights and bias parameters of the multi-layer neural network. The adjustment method is expressed as: ， In the formula, Represents the updated multi-layer neural network The weight of the layer, represents the first The weight of the layer, Represents the updated multi-layer neural network The bias of the layer, represents the first The bias of the layer, represents the sign of partial derivative; Repeat the above steps, and the preset stopping condition is to reach the preset maximum number of iterations, until the preset stopping condition is met, which means that the multi-layer neural network model training is completed.

9. A compact five-pass biomass steam boiler operation status identification system, applying any one of claims 1 to 8 of the compact five-pass biomass steam boiler operation status identification method, comprising: The data acquisition module is used to collect the operating data of the steam boiler, including the boiler's temperature Ra, pressure Da, flow Ea, vibration Fa, fuel consumption rate Ga, boiler efficiency Ha, heat loss Ia, flue gas emission Ja, load rate Ka, and water level La; A data storage module, used for storing collected operation data; The data labeling module is used to manually label the data and label the operating status of the boiler as normal status A, minor fault B, serious fault C, and abnormal status D; The model training module trains the labeled data based on the multi-layer extreme learning machine algorithm to obtain the steam boiler state recognition model; The federated learning module is used to implement model training and parameter exchange of multiple clients, support the federated learning architecture, and ultimately obtain a global model; The state recognition module performs real-time steam boiler operation state recognition based on the trained model and outputs the boiler state category.

Images