Industrial furnace temperature prediction system based on inference human learning optimization algorithm

Abstract

The present invention relates to the technical field of intelligent prediction and control technology, and more specifically, to an industrial furnace temperature prediction system based on a reasoning human learning optimization algorithm, which is used to solve the problems that the prior art cannot accurately predict temperature changes under complex working conditions, cannot adaptively adjust under dynamic working conditions, and cannot achieve accurate prediction of temperature fields and optimization of combustion control. The present invention constructs a furnace temperature prediction model through a prediction model construction module to achieve high-precision prediction of furnace temperature. Through the combination of Bayesian reasoning and an adaptive reasoning network, the system can accurately predict temperature changes under complex working conditions, and respond quickly to achieve optimal combustion control. The system can adaptively adjust under dynamic conditions to achieve accurate prediction of temperature fields and optimization of combustion control. It has strong robustness and is suitable for a variety of high-temperature industrial scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent prediction and control technology, and more specifically, to an industrial furnace temperature prediction system based on a reasoning human learning optimization algorithm. Background Art

[0002] Precise control of industrial furnace temperature is a key link in industrial production, especially in high-temperature environments such as steel and petrochemicals. Accurate temperature prediction and control can significantly improve combustion efficiency and reduce energy consumption. However, current temperature prediction methods, such as those based on traditional neural networks and fuzzy control, usually lack sufficient prediction accuracy and response speed when facing complex and nonlinear working conditions, resulting in poor control effects.

[0003] The patent application with reference publication number CN118009322A discloses a furnace temperature control method and system for a municipal solid waste incineration process in collaboration with an edge, an edge, and a cloud, which relates to the field of furnace temperature control. The method and system obtain process data of a municipal solid waste incineration process in real time; process the collected process data, and establish a furnace temperature prediction model based on the received data, update the prediction model based on a self-correction mechanism, and send the established model to the edge; predict the furnace temperature through a furnace temperature prediction model established and updated online by the edge relying on the cloud, optimize the objective function using the gradient descent method, solve for the optimal control law, and adjust the execution equipment according to the calculated control law, thereby achieving stable and precise control of the furnace temperature in the municipal solid waste incineration process;

[0004] However, the above-mentioned reference patent achieves stable and precise control of furnace temperature through real-time data collection, edge processing, cloud modeling and self-correction, and issuance of optimized control laws, ensuring long-term and efficient combustion process and control performance. However, it cannot accurately predict temperature changes under complex working conditions, cannot adaptively adjust under dynamic working conditions, cannot accurately predict temperature fields and optimize combustion control, and cannot accurately determine the operating status of the furnace, cannot perform timely processing based on monitoring results, and cannot fully guarantee the safe operation of the furnace.

[0005] To this end, we propose an industrial furnace temperature prediction system based on reasoning human learning optimization algorithm to address the above problems. Summary of the invention

[0006] The purpose of the present invention is to provide an industrial furnace temperature prediction system based on an inference human learning optimization algorithm, which solves the problems that the prior art cannot accurately predict temperature changes under complex working conditions, cannot adaptively adjust under dynamic working conditions, cannot achieve accurate prediction of temperature fields and combustion control optimization, and cannot accurately determine the furnace operating status, cannot perform timely processing based on monitoring results, and cannot fully guarantee the safe operation of the furnace.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] An industrial furnace temperature prediction system based on reasoning human learning optimization algorithm, including a combustion control platform, a data collection and preprocessing module, a prediction model building module, a safety monitoring and alarm module, and a control output module;

[0009] The data collection and preprocessing module collects historical furnace combustion dynamic data, including combustion flow RL, air flow KL, oxygen concentration YN, furnace pressure LY and corresponding furnace temperature LW, and performs preprocessing operations on the collected data;

[0010] The prediction model building module builds a furnace temperature prediction model based on the Bayesian reasoning learning strategy and the adaptive reasoning network, inputs the preprocessed furnace combustion dynamic data into the furnace temperature prediction model, and outputs the furnace temperature prediction value;

[0011] Safety monitoring and alarm module, which monitors the key operating parameters of the furnace in real time, including furnace temperature, furnace pressure and flame intensity. When an abnormal situation is detected, an alarm is triggered immediately;

[0012] The control output module outputs adjustment instructions to the furnace operation system based on the prediction results and monitoring results, optimizes the fuel supply and air flow, and optimizes the furnace temperature control.

[0013] As a preferred embodiment of the present invention, the specific process of the data collection preprocessing for preprocessing the historical furnace combustion dynamic data is as follows:

[0014] Data cleaning: Use filters to process signals, remove electromagnetic interference and background noise, use interpolation to fill missing values, and delete records with many missing values;

[0015] Feature Engineering: Calculating the fuel-air ratio: ;

[0016] Calculate the excess air factor: , where AL 理论 is the theoretical air demand;

[0017] Data standardization: Convert the data to a distribution with a mean of 0 and a standard deviation of 1 using the following formula: , where x is the numerical sequence of the signal, μ is the mean, and σ is the standard deviation;

[0018] Divide the dataset: Divide the dataset into training, validation, and test sets with a distribution ratio of 70% training, 15% validation, and 15% testing.

[0019] As a preferred embodiment of the present invention, the specific process of the prediction model building module building the furnace temperature prediction model is as follows:

[0020] The adaptive reasoning network structure consists of the following parts:

[0021] Input layer: Determine the input layer dimension d equal to 6, and convert the preprocessed historical furnace combustion dynamic data into a corresponding set of feature vectors x, x = {x1, x2, ..., x t}, where x t ∈R d is the eigenvector at time t, R d Represents d-dimensional real number space, taking x as input data;

[0022] The spatial features are extracted through the convolution layer. The specific steps are as follows:

[0023] Input: x∈R t×d , where R t×d represents a real matrix with t rows and d columns;

[0024] Convolution operation: C i =f(W c *x i +b c ), where W c ∈R k×d is the convolution kernel, k is the convolution kernel size, b c is the bias, * indicates the convolution operation, and f is the activation function;

[0025] Output: C∈R t＇×d＇ , where C represents the feature matrix output by the convolutional layer, and t' and d' depend on the convolution parameters.

[0026] As a preferred embodiment of the present invention, the time feature is extracted by a recurrent layer, and the specific steps are as follows:

[0027] Input: C∈R t＇×d＇ ;

[0028] Cycle layer update formula: i t =σ(W ii ·x t +b ii +W hi ·h t-1 +b hi );

[0029] f t =σ(W if ·x t +b if +W hf ·h t-1 +bhf );

[0030] g t =tanh(W ig ·x t +b ig +W hg ·h t-1 +b hg );

[0031] o t =σ(W io ·x t +b io +W ho ·h t-1 +b ho );

[0032] c t =f t *c t-1 +i t *g t ;

[0033] h t =o t *tanh(c t );

[0034] where i t 、f t , g t , o t 、c t 、h t They are input gate, forget gate, unit state, output gate, unit state update, hidden state update, W ii , W if , W ig , W io They are the input weights of the input gate, forget gate, unit state, and output gate, respectively. hi , W hf , W hg , W ho are the weights from the hidden state to the input gate, the forget gate, the candidate state, and the output gate, respectively. ii 、b if 、b ig 、b io They are the bias of the input gate, forget gate, candidate state, and output gate, respectively, and b hi 、b hf 、b hg 、b ho They are the biases from hidden state to input gate, forget gate, candidate state, and output gate, σ is the sigmoid function, tanh is the hyperbolic tangent function, and x trepresents the input vector at time step t, h t-1 represents the hidden state at the previous time step t-1, c t-1 represents the cell state at the previous time step t-1;

[0035] Output: Z t , the output of the recurrent layer is the hidden state h t , that is, Z t =h t ;

[0036] Reasoning optimization layer: embeds reasoning learning operators and combines Bayesian reasoning learning strategies to perform parameter optimization;

[0037] Output layer: output furnace temperature prediction value.

[0038] As a preferred implementation of the present invention, the core of the Bayesian reasoning learning strategy is an update mechanism based on Bayesian theory, and the specific steps are as follows:

[0039] Initialization parameters: Assume that the initial parameter of the model is θ0 and its prior distribution is p(θ0);

[0040] Data observation: At time t, the real-time data D of the furnace is observed t , D t ={x t， y t};

[0041] Parameter update: According to Bayes' theorem, update the posterior distribution of the parameters:

[0042] ;

[0043] Where p(D t |θ t ) is the likelihood function, p(θ t-1 |D t-1 ) is the posterior distribution of the parameters at the previous time step, p(D t ) is the normalization constant;

[0044] Parameter sampling: From the posterior distribution p(θ t |D t ) to obtain the new parameter θ t , calculated by the following formula: t =MCMC(p(θ t |D t )), where MCMC stands for Markov Chain Monte Carlo method, which is used to obtain the posterior distribution p(θ t |D t ) in the sample;

[0045] Model prediction: using updated parameters θt Carry out furnace temperature prediction.

[0046] As a preferred embodiment of the present invention, a furnace temperature prediction model is constructed, and the specific steps are as follows:

[0047] Input layer: receives preprocessed furnace combustion dynamic data x t ;

[0048] Convolutional layer: extract spatial features C;

[0049] Recurrent layer: extracting temporal features Z t ;

[0050] Reasoning optimization layer: Update model parameters θ in combination with Bayesian reasoning learning strategy t ;

[0051] Output layer: Use the updated parameters θ t Carry out furnace temperature prediction and output furnace temperature prediction value , the calculation formula of the furnace temperature prediction model is: .

[0052] As a preferred embodiment of the present invention, the furnace temperature prediction model training process is as follows:

[0053] Data preparation: Collect furnace combustion dynamic data and perform preprocessing;

[0054] Initialize model parameters: Randomly initialize the parameters of the convolutional layer, recurrent layer, and inference optimization layer;

[0055] Forward propagation: The input data is used to extract spatial features through the convolution layer, and the spatial features are used to extract temporal features through the circulation layer. The temporal features are input into the inference optimization layer, and the model parameters are updated in combination with the Bayesian reasoning learning strategy. The updated parameters are used to predict the furnace temperature.

[0056] Loss function: Define the loss function. The loss function calculation formula is as follows:

[0057] ;

[0058] Where N is the number of samples, is the true value, is the predicted value;

[0059] Back propagation: Calculate the gradient of the loss function with respect to the model parameters and use the optimization algorithm Adam to update the model parameters;

[0060] Iterative training: Repeat the forward propagation and backpropagation process until the model converges or reaches a predetermined number of training rounds.

[0061] As a preferred embodiment of the present invention, the specific process of the safety monitoring alarm module monitoring the key operating parameters of the furnace is as follows:

[0062] Obtain the key operating parameters of the furnace, generate a monitoring cycle, and divide the monitoring cycle JC into multiple monitoring periods {jc1, jc2, …, jcn}, that is, JC={jc1, jc2, …, jcn};

[0063] Obtain the furnace temperature change rate of the furnace in multiple monitoring periods, the furnace temperature change rate represents the ratio between the furnace temperature change amount and the length of the corresponding time period, and calculate the arithmetic average of the multiple furnace temperature change rates obtained, and record the arithmetic average of the multiple furnace temperature change rates as the average furnace temperature change rate PLW;

[0064] Obtaining the furnace pressure change rate of the furnace in multiple monitoring periods, the furnace pressure change rate represents the ratio between the furnace pressure change amount and the length of the corresponding time period, and calculating the arithmetic average of the multiple furnace pressure change rates obtained, and recording the arithmetic average of the multiple furnace pressure change rates as the average furnace pressure change rate PLY;

[0065] The furnace flame intensity change rate of the furnace in multiple monitoring time periods is obtained. The flame intensity change rate represents the ratio between the change in furnace flame intensity and the length of the corresponding time period. The arithmetic mean of the multiple furnace flame intensity change rates obtained is calculated and recorded as the average furnace flame intensity change rate PLH.

[0066] As a preferred embodiment of the present invention, the average furnace temperature change rate PLW, the average furnace pressure change rate PLY and the average furnace flame intensity change rate PLH are obtained, and the operation monitoring evaluation coefficient YJP is calculated by the following formula:

[0067] YJP=g1*PLW+g2*PLY+g3*PLH;

[0068] Among them, g1+g2+g3=1, g1, g2, g3 are weight coefficients of average furnace temperature change rate, average furnace pressure change rate, and average furnace flame intensity change rate, respectively. The operation monitoring assessment coefficient YJP is compared with the preset operation monitoring assessment coefficient threshold:

[0069] If the operation monitoring assessment coefficient YJP is less than the preset operation monitoring assessment coefficient threshold, it indicates that the furnace is operating normally and no signal is generated;

[0070] If the operation monitoring and evaluation coefficient YJP is greater than or equal to the preset operation monitoring and evaluation coefficient threshold, it indicates that there is an abnormality in the furnace operation, and an alarm signal is generated and sent to the combustion control platform.

[0071] As a preferred embodiment of the present invention, the specific process of the control output module adjusting the furnace operation system is as follows:

[0072] Obtain prediction results and monitoring results. The prediction result is the predicted value of furnace temperature, and the monitoring result is the furnace operation status. The monitoring result has a higher priority than the prediction result.

[0073] When the monitoring result shows that the furnace operation is abnormal, the prediction result is ignored and the furnace operation system is adjusted according to the monitoring result. After receiving the alarm signal, the combustion control platform immediately outputs the adjustment instruction to the furnace operation system. The adjustment instruction content is as follows:

[0074] Automatically shut down fuel supply, automatically start air cooling and water cooling systems to reduce furnace temperature, and automatically adjust air flow supply;

[0075] When the monitoring result shows that the furnace is operating normally and the predicted furnace temperature is higher than the preset furnace temperature threshold, the combustion control platform immediately outputs an adjustment instruction to the furnace operation system. The adjustment instruction content is as follows:

[0076] Automatically reduce fuel supply, automatically start the air cooling system to reduce furnace temperature, and automatically increase air flow supply;

[0077] When the monitoring result shows that the furnace is operating normally and the predicted furnace temperature is lower than the preset furnace temperature threshold, the combustion control platform immediately outputs an adjustment instruction to the furnace operation system. The adjustment instruction content is as follows:

[0078] Automatically increase fuel supply, automatically shut down air cooling and water cooling systems, and automatically reduce air flow supply.

[0079] Compared with the prior art, the advantages of the present invention are:

[0080] (1) In the present invention, a furnace temperature prediction model is constructed by a prediction model construction module to achieve high-precision prediction of the furnace temperature. Through the combination of Bayesian reasoning and adaptive reasoning network, the system can accurately predict temperature changes under complex working conditions and respond quickly to achieve optimal combustion control. The system can adaptively adjust under dynamic working conditions to achieve accurate prediction of the temperature field and optimization of combustion control. It has strong robustness and is suitable for a variety of high-temperature industrial scenarios.

[0081] (2) In the present invention, the key parameters of the furnace are monitored in real time through the safety monitoring and alarm module, the monitoring accuracy is guaranteed by calculating the change rate and the evaluation coefficient, the weight is dynamically adjusted according to the historical data, and it is flexibly adapted to different environments. The alarm signal is automatically generated and sent to the control platform, which reduces manual intervention and improves operational efficiency. The comprehensive evaluation of multiple parameters improves the monitoring reliability and stability, and comprehensively guarantees the safe operation of the furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is a system block diagram of the present invention;

[0083] Figure 2 This is a flow chart of a method for constructing a furnace temperature prediction model in Embodiment 1 of the present invention;

[0084] Figure 3 This is a logical flow diagram of the second embodiment of the present invention. DETAILED DESCRIPTION

[0085] The following will combine the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. All other embodiments obtained by ordinary technicians in this field without creative work based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0086] Embodiment 1: Figure 1 and Figure 2 As shown, the industrial furnace temperature prediction system based on the inference human learning optimization algorithm proposed in the present invention includes a combustion control platform, a data collection and preprocessing module, a prediction model building module, a safety monitoring and alarm module, and a control output module;

[0087] The industrial furnace temperature prediction system based on the inference human learning optimization algorithm in the present invention introduces the Bayesian inference learning strategy into human learning and algorithms to form an inference human learning optimization algorithm, and combines the inference human learning optimization algorithm with a neural network to propose an adaptive inference neural network. A furnace temperature prediction model is constructed based on the Bayesian inference learning strategy and the adaptive inference network to predict the industrial furnace temperature.

[0088] The combustion control platform is connected to the prediction model building module and the safety monitoring alarm module in two-way communication, and the combustion control platform is connected to the data collection preprocessing module and the control output module in one-way communication;

[0089] The data collection and preprocessing module collects historical furnace combustion dynamic data, including combustion flow RL, air flow KL, oxygen concentration YN, furnace pressure LY and corresponding furnace temperature LW, and performs preprocessing operations on the collected data;

[0090] Data collection preprocessing The specific process of preprocessing the historical furnace combustion dynamic data is as follows:

[0091] Data cleaning: Use filters to process signals, remove electromagnetic interference and background noise, use interpolation to fill missing values, and delete records with many missing values;

[0092] Feature Engineering: Calculating the fuel-air ratio: ;

[0093] Calculate the excess air factor: , where AL 理论 is the theoretical air demand;

[0094] Data standardization: Convert the data to a distribution with a mean of 0 and a standard deviation of 1 using the following formula: , where x is the numerical sequence of the signal, μ is the mean, and σ is the standard deviation;

[0095] Divide the dataset: Divide the dataset into training set, validation set and test set, with the allocation ratio of 70% training, 15% validation and 15% test;

[0096] The data collection preprocessing module significantly improves the quality and availability of furnace combustion dynamic data through preprocessing operations such as data cleaning, feature engineering, data standardization and data set partitioning. Data cleaning removes noise and missing values ​​and improves data accuracy; feature engineering extracts useful features and enhances model prediction capabilities; data standardization accelerates model convergence and improves prediction accuracy; data set partitioning verifies model generalization capabilities and avoids overfitting. These operations not only improve model performance, but also simplify complexity, accelerate training and verification, and provide a reliable data foundation for subsequent model applications.

[0097] The prediction model building module builds a furnace temperature prediction model based on the Bayesian reasoning learning strategy and the adaptive reasoning network, inputs the preprocessed furnace combustion dynamic data into the furnace temperature prediction model, and outputs the furnace temperature prediction value;

[0098] The specific process of constructing the furnace temperature prediction model in the prediction model building module is as follows:

[0099] The adaptive reasoning network structure consists of the following parts:

[0100] Input layer: Determine the input layer dimension d equal to 6, and convert the preprocessed historical furnace combustion dynamic data into a corresponding set of feature vectors x, x = {x1, x2, ..., x t}, where x t ∈R d is the eigenvector at time t, R d Represents d-dimensional real number space, taking x as input data;

[0101] The spatial features are extracted through the convolution layer. The specific steps are as follows:

[0102] Input: x∈R t×d , where R t×d represents a real matrix with t rows and d columns;

[0103] Convolution operation: C i=f(W c *x i +b c ), where W c ∈R k×d is the convolution kernel, k is the convolution kernel size, b c is the bias, * indicates the convolution operation, and f is the activation function;

[0104] Output: C∈R t＇×d＇ , where C represents the feature matrix output by the convolutional layer, and t' and d' depend on the convolution parameters;

[0105] The time features are extracted through the recurrent layer. The specific steps are as follows:

[0106] Input: C∈R t＇×d＇ ;

[0107] Cycle layer update formula: i t =σ(W ii ·x t +b ii +W hi ·h t-1 +b hi );

[0108] f t =σ(W if ·x t +b if +W hf ·h t-1 +b hf );

[0109] g t =tanh(W ig ·x t +b ig +W hg ·h t-1 +b hg );

[0110] o t =σ(W io ·x t +b io +W ho ·h t-1 +b ho );

[0111] c t =f t *c t-1 +i t *g t ;

[0112] h t =o t*tanh(c t );

[0113] where i t 、f t , g t , o t 、c t 、h t They are input gate, forget gate, unit state, output gate, unit state update, hidden state update, W ii , W if , W ig , W io They are the input weights of the input gate, forget gate, unit state, and output gate, respectively. hi , W hf , W hg , W ho are the weights from the hidden state to the input gate, the forget gate, the candidate state, and the output gate, respectively. ii 、b if 、b ig 、b io They are the bias of the input gate, forget gate, candidate state, and output gate, respectively, and b hi 、b hf 、b hg 、b ho They are the biases from hidden state to input gate, forget gate, candidate state, and output gate, σ is the sigmoid function, tanh is the hyperbolic tangent function, and x t represents the input vector at time step t, h t-1 represents the hidden state at the previous time step t-1, c t-1 represents the cell state at the previous time step t-1;

[0114] Output: Z t , the output of the recurrent layer is the hidden state h t , that is, Z t =h t ;

[0115] Reasoning optimization layer: embeds reasoning learning operators and combines Bayesian reasoning learning strategies to perform parameter optimization;

[0116] Output layer: output furnace temperature prediction value;

[0117] The core of the Bayesian reasoning learning strategy is the update mechanism based on Bayesian theory. The specific steps are as follows:

[0118] Initialization parameters: Assume that the initial parameter of the model is θ0 and its prior distribution is p(θ0);

[0119] Data observation: At time t, the real-time data D of the furnace is observedt , D t ={x t， y t};

[0120] Parameter update: According to Bayes' theorem, update the posterior distribution of the parameters:

[0121] ;

[0122] Where p(D t |θ t ) is the likelihood function, p(θ t-1 |D t-1 ) is the posterior distribution of the parameters at the previous time step, p(D t ) is the normalization constant;

[0123] Parameter sampling: From the posterior distribution p(θ t |D t ) to obtain the new parameter θ t , calculated by the following formula: t =MCMC(p(θ t |D t )), where MCMC stands for Markov Chain Monte Carlo method, which is used to obtain the posterior distribution p(θ t |D t ) in the sample;

[0124] Model prediction: using updated parameters θ t Carry out furnace temperature prediction;

[0125] Construct a furnace temperature prediction model. The specific steps are as follows:

[0126] Input layer: receives preprocessed furnace combustion dynamic data x t ;

[0127] Convolutional layer: extract spatial features C;

[0128] Recurrent layer: extracting temporal features Z t ;

[0129] Reasoning optimization layer: Update model parameters θ in combination with Bayesian reasoning learning strategy t ;

[0130] Output layer: Use the updated parameters θ t Carry out furnace temperature prediction and output furnace temperature prediction value , the calculation formula of the furnace temperature prediction model is: ;

[0131] The furnace temperature prediction model training process is as follows:

[0132] Data preparation: Collect furnace combustion dynamic data and perform preprocessing;

[0133] Initialize model parameters: Randomly initialize the parameters of the convolutional layer, recurrent layer, and inference optimization layer;

[0134] Forward propagation: The input data is used to extract spatial features through the convolution layer, and the spatial features are used to extract temporal features through the circulation layer. The temporal features are input into the inference optimization layer, and the model parameters are updated in combination with the Bayesian reasoning learning strategy. The updated parameters are used to predict the furnace temperature.

[0135] Loss function: Define the loss function. The loss function calculation formula is as follows:

[0136] ;

[0137] Where N is the number of samples, is the true value, is the predicted value;

[0138] Back propagation: Calculate the gradient of the loss function with respect to the model parameters and use the optimization algorithm Adam to update the model parameters;

[0139] Iterative training: Repeat the forward propagation and back propagation process until the model converges or reaches the predetermined number of training rounds;

[0140] The prediction model building module combines Bayesian reasoning and adaptive reasoning networks. This system can accurately predict temperature changes and respond quickly to achieve optimal combustion control. It has intelligent adaptability and strong robustness and is suitable for various high-temperature industrial scenarios such as steelmaking, electric power and chemical industry. The system uses historical data and real-time data to update parameters to ensure the accuracy and real-time performance of the prediction model, and uses the Markov chain Monte Carlo method to perform automated parameter sampling and optimization to reduce manual intervention and improve operational efficiency. Multi-level feature extraction includes convolutional layer extraction of spatial features, cyclic layer extraction of temporal features, and inference optimization layer combined with Bayesian reasoning for parameter optimization to ensure that the model fully captures the furnace operation status. In addition, back propagation is performed through the Adam optimization algorithm to accelerate model convergence and improve training efficiency.

[0141] Embodiment 2: The technical solution of the embodiment of the present invention is different from that of embodiment 1 in that:

[0142] like Figure 1 and Figure 3 As shown, the safety monitoring and alarm module monitors the key operating parameters of the furnace in real time, including furnace temperature, furnace pressure and flame intensity. When an abnormal situation is detected, an alarm is triggered immediately;

[0143] The specific process of the safety monitoring and alarm module monitoring the key operating parameters of the furnace is as follows:

[0144] Obtain the key operating parameters of the furnace, generate a monitoring cycle, and divide the monitoring cycle JC into multiple monitoring periods {jc1, jc2, …, jcn}, that is, JC={jc1, jc2, …, jcn};

[0145] Obtain the furnace temperature change rate of the furnace in multiple monitoring periods, the furnace temperature change rate represents the ratio between the furnace temperature change amount and the length of the corresponding time period, and calculate the arithmetic average of the multiple furnace temperature change rates obtained, and record the arithmetic average of the multiple furnace temperature change rates as the average furnace temperature change rate PLW;

[0146] Obtaining the furnace pressure change rate of the furnace in multiple monitoring periods, the furnace pressure change rate represents the ratio between the furnace pressure change amount and the length of the corresponding time period, and calculating the arithmetic average of the multiple furnace pressure change rates obtained, and recording the arithmetic average of the multiple furnace pressure change rates as the average furnace pressure change rate PLY;

[0147] Obtain the furnace flame intensity change rate of the furnace in multiple monitoring time periods, the flame intensity change rate represents the ratio between the change amount of the furnace flame intensity and the length of the corresponding time period, and calculate the arithmetic average of the multiple furnace flame intensity change rates obtained, and record the arithmetic average of the multiple furnace flame intensity change rates as the average furnace flame intensity change rate PLH;

[0148] The average furnace temperature change rate PLW, the average furnace pressure change rate PLY and the average furnace flame intensity change rate PLH are obtained, and the operation monitoring evaluation coefficient YJP is calculated by the following formula:

[0149] YJP=g1*PLW+g2*PLY+g3*PLH;

[0150] Among them, g1+g2+g3=1, g1, g2, g3 are the weight coefficients of average furnace temperature change rate, average furnace pressure change rate, and average furnace flame intensity change rate respectively;

[0151] By analyzing historical data, we can understand the influence of weight coefficients g1, g2 and g3 on the furnace operation status. According to historical data, we can determine the specific values ​​of g1, g2 and g3 to ensure that the calculated operation monitoring and evaluation coefficient YJP can accurately reflect the actual operation status of the furnace.

[0152] Compare the operation monitoring assessment coefficient YJP with the preset operation monitoring assessment coefficient threshold:

[0153] If the operation monitoring assessment coefficient YJP is less than the preset operation monitoring assessment coefficient threshold, it indicates that the furnace is operating normally and no signal is generated;

[0154] If the operation monitoring and evaluation coefficient YJP is greater than or equal to the preset operation monitoring and evaluation coefficient threshold, it indicates that there is an abnormality in the furnace operation, and an alarm signal is generated and sent to the combustion control platform;

[0155] The safety monitoring and alarm module can monitor the key operating parameters of the furnace in real time and detect abnormal situations in time to improve safety. It can accurately reflect the actual operating status of the furnace by calculating the change rate and evaluation coefficient to ensure the accuracy of the monitoring results. It can dynamically adjust the weight coefficient according to historical data and flexibly adapt to different operating environments and needs. It can automatically generate alarm signals and send them to the control platform, reducing manual intervention and improving operational efficiency. It can improve the reliability and stability of the monitoring results through comprehensive evaluation of multiple parameters, providing comprehensive protection for the safe operation of the furnace.

[0156] The control output module outputs adjustment instructions to the furnace operation system based on the prediction results and monitoring results to optimize the fuel supply and air flow and the furnace temperature control;

[0157] The specific process of controlling the output module to adjust the furnace operation system is as follows:

[0158] Obtain prediction results and monitoring results. The prediction result is the predicted value of furnace temperature, and the monitoring result is the furnace operation status. The monitoring result has a higher priority than the prediction result.

[0159] When the monitoring result shows that the furnace operation is abnormal, the prediction result is ignored and the furnace operation system is adjusted according to the monitoring result. After receiving the alarm signal, the combustion control platform immediately outputs the adjustment instruction to the furnace operation system. The adjustment instruction content is as follows:

[0160] Automatically shut down fuel supply, automatically start air cooling and water cooling systems to reduce furnace temperature, and automatically adjust air flow supply;

[0161] When the monitoring result shows that the furnace is operating normally and the predicted furnace temperature is higher than the preset furnace temperature threshold, the combustion control platform immediately outputs an adjustment instruction to the furnace operation system. The adjustment instruction content is as follows:

[0162] Automatically reduce fuel supply, automatically start the air cooling system to reduce furnace temperature, and automatically increase air flow supply;

[0163] When the monitoring result shows that the furnace is operating normally and the predicted furnace temperature is lower than the preset furnace temperature threshold, the combustion control platform immediately outputs an adjustment instruction to the furnace operation system. The adjustment instruction content is as follows:

[0164] Automatically increase fuel supply, automatically shut down air cooling and water cooling systems, and automatically reduce air flow supply;

[0165] The control output module responds quickly to monitoring results and takes immediate measures to ensure furnace safety when abnormal furnace operation is detected. It intelligently adjusts fuel supply and air flow to ensure that the furnace temperature is maintained within the optimal operating range and improves combustion efficiency. The automation function reduces manual intervention, automatically generates and sends adjustment instructions, and improves operational efficiency and consistency. The module can dynamically adjust the adjustment strategy according to different operating conditions and prediction results, adapt to various operating environments and requirements, and ensure the stability and reliability of the system. In addition, it effectively prevents and responds to abnormal conditions in furnace operation, improving the safety and stability of the system.

[0166] The working principle of the present invention: when in use, a furnace temperature prediction model is constructed through the prediction model construction module to achieve high-precision prediction of the furnace temperature. Through the combination of Bayesian reasoning and adaptive reasoning network, the system can accurately predict temperature changes under complex working conditions and respond quickly to achieve optimal combustion control. The system can adaptively adjust under dynamic working conditions to achieve accurate prediction of the temperature field and optimization of combustion control. It has strong robustness and is suitable for a variety of high-temperature industrial scenarios. It can also monitor key furnace parameters in real time through the safety monitoring and alarm module, ensure monitoring accuracy by calculating the rate of change and the evaluation coefficient, dynamically adjust the weight according to historical data, flexibly adapt to different environments, automatically generate alarm signals and send them to the control platform, reduce manual intervention, and improve operating efficiency. The comprehensive evaluation of multiple parameters improves monitoring reliability and stability, and comprehensively guarantees the safe operation of the furnace.

[0167] The above are only preferred specific implementation modes of the present invention; however, the protection scope of the present invention is not limited thereto; any technician familiar with the technical field within the technical scope disclosed by the present invention; any equivalent replacement or change based on the technical solution and improved concept of the present invention shall be covered within the protection scope of the present invention.

Claims

1. An industrial furnace temperature prediction system based on reasoning human learning optimization algorithm, characterized in that, It includes combustion control platform, data collection and preprocessing module, prediction model building module, safety monitoring and alarm module and control output module; The data collection and preprocessing module collects historical furnace combustion dynamic data, including combustion flow RL, air flow KL, oxygen concentration YN, furnace pressure LY and corresponding furnace temperature LW, and performs preprocessing operations on the collected data; The prediction model building module builds a furnace temperature prediction model based on the Bayesian reasoning learning strategy and the adaptive reasoning network, inputs the preprocessed furnace combustion dynamic data into the furnace temperature prediction model, and outputs the furnace temperature prediction value; The specific process of constructing the furnace temperature prediction model by the prediction model construction module is as follows: The adaptive reasoning network structure consists of the following parts: Input layer: Determine the input layer dimension d equal to 6, and convert the preprocessed historical furnace combustion dynamic data into a corresponding set of feature vectors x, x = {x1, x2, ..., x t }, where x t ∈R d is the eigenvector at time t, R d Represents d-dimensional real number space, taking x as input data; The spatial features are extracted through the convolution layer. The specific steps are as follows: Input: x∈R t×d , where R t×d represents a real matrix with t rows and d columns; Convolution operation: C i =f(W c *x i +b c ), where W c ∈R k×d is the convolution kernel, k is the convolution kernel size, b c is the bias, * indicates the convolution operation, and f is the activation function; Output: C∈R t＇×d＇ , where C represents the feature matrix output by the convolutional layer, and t' and d' depend on the convolution parameters; The core of the Bayesian reasoning learning strategy is the update mechanism based on Bayesian theory. The specific steps are as follows: Initialization parameters: Assume that the initial parameter of the model is θ0 and its prior distribution is p(θ0); Data observation: At time t, the real-time data D of the furnace is observed t , D t ={x t， y t }; Parameter update: According to Bayes' theorem, update the posterior distribution of the parameters: ； Where p(D t |θ t ) is the likelihood function, p(θ t-1 |D t-1 ) is the posterior distribution of the parameters at the previous time step, p(D t ) is the normalization constant; Parameter sampling: From the posterior distribution p(θ t |D t ) to obtain the new parameter θ t , calculated by the following formula: t =MCMC(p(θ t |D t )), where MCMC stands for Markov Chain Monte Carlo method, which is used to obtain the posterior distribution p(θ t |D t ) in the sample; Model prediction: using updated parameters θ t Carry out furnace temperature prediction; Safety monitoring and alarm module, which monitors the key operating parameters of the furnace in real time, including furnace temperature, furnace pressure and flame intensity. When an abnormal situation is detected, an alarm is triggered immediately; The control output module outputs adjustment instructions to the furnace operation system based on the prediction results and monitoring results, optimizes the fuel supply and air flow, and optimizes the furnace temperature control.

2. The industrial furnace temperature prediction system based on reasoning human learning optimization algorithm according to claim 1 is characterized in that: The specific process of data collection preprocessing for preprocessing historical furnace combustion dynamic data is as follows: Data cleaning: Use filters to process signals, remove electromagnetic interference and background noise, use interpolation to fill missing values, and delete records with many missing values; feature Engineering: Calculate the fuel-air ratio: ; Calculate the excess air factor: , where AL 理论 is the theoretical air demand; Data standardization: Convert the data to a distribution with a mean of 0 and a standard deviation of 1 using the following formula: , where x is the numerical sequence of the signal, μ is the mean, and σ is the standard deviation; Divide the dataset: Divide the dataset into training, validation, and test sets with a distribution ratio of 70% training, 15% validation, and 15% testing.

3. The industrial furnace temperature prediction system based on reasoning human learning optimization algorithm according to claim 1 is characterized in that: The time features are extracted through the recurrent layer. The specific steps are as follows: Input: C∈R t＇×d＇ ; Cycle layer update formula: i t =σ(W ii ·x t +b ii +W hi ·h t-1 +b hi ); f t =σ(W if ·x t +b if +W hf ·h t-1 +b hf ); g t =tanh(W ig ·x t +b ig +W hg ·h t-1 +b hg ); o t =σ(W io ·x t +b io +W ho ·h t-1 +b ho ); c t =f t *c t-1 +i t *g t ; h t =o t *tanh(c t ); where i t 、f t , g t , o t 、c t 、h t They are input gate, forget gate, unit state, output gate, unit state update, hidden state update, W ii , W if , W ig , W io They are the input weights of the input gate, forget gate, unit state, and output gate, respectively. hi , W hf , W hg , W ho are the weights from the hidden state to the input gate, the forget gate, the candidate state, and the output gate, respectively. ii , b if , b ig , b io They are the bias of the input gate, forget gate, candidate state, and output gate, respectively, and b hi , b hf , b hg , b ho They are the biases from hidden state to input gate, forget gate, candidate state, and output gate, σ is the sigmoid function, tanh is the hyperbolic tangent function, and x t represents the input vector at time step t, h t-1 represents the hidden state at the previous time step t-1, c t-1 represents the cell state at the previous time step t-1; Output: Z t , the output of the recurrent layer is the hidden state h t , that is, Z t =h t ; Reasoning optimization layer: embeds reasoning learning operators and combines Bayesian reasoning learning strategies to perform parameter optimization; Output layer: output furnace temperature prediction value.

4. The industrial furnace temperature prediction system based on reasoning human learning optimization algorithm according to claim 1 is characterized in that: Construct a furnace temperature prediction model. The specific steps are as follows: Input layer: receives preprocessed furnace combustion dynamic data x t ; Convolutional layer: extract spatial features C; Recurrent layer: extracting temporal features Z t ; Reasoning optimization layer: Update model parameters θ in combination with Bayesian reasoning learning strategy t ; Output layer: Use the updated parameters θ t Carry out furnace temperature prediction and output furnace temperature prediction value , the calculation formula of the furnace temperature prediction model is: .

5. The industrial furnace temperature prediction system based on reasoning human learning optimization algorithm according to claim 4 is characterized in that: The furnace temperature prediction model training process is as follows: Data preparation: Collect furnace combustion dynamic data and perform preprocessing; Initialize model parameters: Randomly initialize the parameters of the convolutional layer, recurrent layer, and inference optimization layer; Forward propagation: The input data is used to extract spatial features through the convolution layer, and the spatial features are used to extract temporal features through the circulation layer. The temporal features are input into the inference optimization layer, and the model parameters are updated in combination with the Bayesian reasoning learning strategy. The updated parameters are used to predict the furnace temperature. Loss function: Define the loss function. The loss function calculation formula is as follows: ； Where N is the number of samples, is the true value, is the predicted value; Back propagation: Calculate the gradient of the loss function with respect to the model parameters and use the optimization algorithm Adam to update the model parameters; Iterative training: Repeat the forward propagation and backpropagation process until the model converges or reaches a predetermined number of training rounds.

6. The industrial furnace temperature prediction system based on reasoning human learning optimization algorithm according to claim 1 is characterized in that: The specific process of the safety monitoring alarm module monitoring the key operating parameters of the furnace is as follows: Obtain the key operating parameters of the furnace, generate a monitoring cycle, and divide the monitoring cycle JC into multiple monitoring periods {jc1, jc2, …, jcn}, that is, JC={jc1, jc2, …, jcn}; Obtain the furnace temperature change rate of the furnace in multiple monitoring periods, the furnace temperature change rate represents the ratio between the furnace temperature change amount and the length of the corresponding time period, and calculate the arithmetic average of the multiple furnace temperature change rates obtained, and record the arithmetic average of the multiple furnace temperature change rates as the average furnace temperature change rate PLW; Obtaining the furnace pressure change rate of the furnace in multiple monitoring periods, the furnace pressure change rate represents the ratio between the furnace pressure change amount and the length of the corresponding time period, and calculating the arithmetic average of the multiple furnace pressure change rates obtained, and recording the arithmetic average of the multiple furnace pressure change rates as the average furnace pressure change rate PLY; The furnace flame intensity change rate of the furnace in multiple monitoring time periods is obtained. The flame intensity change rate represents the ratio between the change in furnace flame intensity and the length of the corresponding time period. The arithmetic mean of the multiple furnace flame intensity change rates obtained is calculated and recorded as the average furnace flame intensity change rate PLH.

7. The industrial furnace temperature prediction system based on the reasoning human learning optimization algorithm according to claim 6 is characterized in that: The average furnace temperature change rate PLW, the average furnace pressure change rate PLY and the average furnace flame intensity change rate PLH are obtained, and the operation monitoring evaluation coefficient YJP is calculated by the following formula: YJP=g1*PLW+g2*PLY+g3*PLH; Among them, g1+g2+g3=1, g1, g2, g3 are weight coefficients of average furnace temperature change rate, average furnace pressure change rate, and average furnace flame intensity change rate, respectively. The operation monitoring assessment coefficient YJP is compared with the preset operation monitoring assessment coefficient threshold: If the operation monitoring assessment coefficient YJP is less than the preset operation monitoring assessment coefficient threshold, it indicates that the furnace is operating normally and no signal is generated; If the operation monitoring and evaluation coefficient YJP is greater than or equal to the preset operation monitoring and evaluation coefficient threshold, it indicates that there is an abnormality in the furnace operation, and an alarm signal is generated and sent to the combustion control platform.

8. The industrial furnace temperature prediction system based on reasoning human learning optimization algorithm according to claim 1 is characterized in that: The specific process of the control output module adjusting the furnace operation system is as follows: Obtain prediction results and monitoring results. The prediction result is the predicted value of furnace temperature, and the monitoring result is the furnace operation status. The monitoring result has a higher priority than the prediction result. When the monitoring result shows that the furnace operation is abnormal, the prediction result is ignored and the furnace operation system is adjusted according to the monitoring result. After receiving the alarm signal, the combustion control platform immediately outputs the adjustment instruction to the furnace operation system. The adjustment instruction content is as follows: Automatically shut down fuel supply, automatically start air cooling and water cooling systems to reduce furnace temperature, and automatically adjust air flow supply; When the monitoring result shows that the furnace is operating normally and the predicted furnace temperature is higher than the preset furnace temperature threshold, the combustion control platform immediately outputs an adjustment instruction to the furnace operation system. The adjustment instruction content is as follows: Automatically reduce fuel supply, automatically start the air cooling system to reduce furnace temperature, and automatically increase air flow supply; When the monitoring result shows that the furnace is operating normally and the predicted furnace temperature is lower than the preset furnace temperature threshold, the combustion control platform immediately outputs an adjustment instruction to the furnace operation system. The adjustment instruction content is as follows: Automatically increase fuel supply, automatically shut down air cooling and water cooling systems, and automatically reduce air flow supply.

Images