A circulating fluidized bed boiler furnace outlet temperature control system and method

Abstract

This invention discloses a circulating fluidized bed boiler furnace outlet temperature control system and method, belonging to the field of equipment control technology. It addresses the technical problem that existing solutions cannot improve the adaptability and accuracy of furnace outlet temperature control through hybrid modeling. By using the output error of the mechanistic model as input, and fusing the predicted value of the mechanistic model with data-driven correction coefficients, a closed-loop logic of mechanistic modeling, error compensation, and dynamic fusion is achieved. Multi-step prediction over several seconds solves the problem of large time delay in the boiler system, effectively reducing the impact of traditional control lag and enabling control actions to adapt to disturbances in advance, achieving proactive control. Real-time capture of coal type switching through convolutional neural networks and rapid adaptive correction of control quantities through weight matrix adjustment improve dynamic adaptability. Rolling time-domain optimization combined with operating condition classification weights effectively improves the control accuracy of the furnace outlet temperature and reduces overshoot.

Description

Technical Field

[0001] This invention relates to the field of equipment control technology, specifically to a method for controlling the furnace outlet temperature of a circulating fluidized bed boiler. Background Technology

[0002] Controlling the furnace outlet temperature of a circulating fluidized bed (CFB) boiler is a key aspect of the boiler's safe, efficient, and environmentally friendly operation. The control objective is to maintain the furnace outlet flue gas temperature within a stable and reasonable range, typically between 850°C and 950°C.

[0003] Existing circulating fluidized bed (CFB) boiler furnace outlet temperature control schemes rely heavily on mechanistic or simplified identification models, which fail to accurately describe the highly nonlinear and multivariate coupling characteristics of CFB boilers, leading to significant temperature prediction errors. Furthermore, existing identification models, such as neural networks, depend on extensive historical data training, but changes in fuel characteristics during actual operation significantly reduce the model's generalization ability. Consequently, there is a deficiency in the ability to improve the adaptability and accuracy of furnace outlet temperature control through hybrid modeling methods. Summary of the Invention

[0004] The purpose of this invention is to provide a method for controlling the furnace outlet temperature of a circulating fluidized bed boiler, which solves the technical problem that existing solutions cannot improve the adaptability and accuracy of furnace outlet temperature control through hybrid modeling.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for controlling the furnace outlet temperature of a circulating fluidized bed boiler includes:

[0007] Step 1, Construct a dynamic model of mechanism-data fusion: Based on the laws of mass conservation and energy conservation, establish a furnace combustion heat transfer mechanism model, and define the state vector including furnace outlet temperature, bed temperature, coal feed rate, primary air volume and secondary air volume; use the error feedback type data-driven module to compensate for the deviation of the mechanism model in real time, and use the prediction error of the mechanism model and real-time operating parameters as input to output dynamic correction coefficients;

[0008] Step 2: Design an adaptive predictive controller based on a fusion model: Use the constructed mechanism data fusion model to predict the furnace outlet temperature change trend in the next few seconds, use a rolling time-domain optimization algorithm to generate an initial control sequence, use a convolutional neural network operating condition feature extractor to perform feature dimensionality reduction on real-time operating data, dynamically adjust the weight matrix of the predictive controller, and provide monitoring prompts on the implementation effect of correcting the weights of the objective function based on the calculated monitoring evaluation array.

[0009] Preferably, when constructing the dynamic mechanism model of the circulating fluidized bed boiler furnace based on the laws of mass conservation and energy conservation, a state vector is set. ;in, These are the furnace outlet temperature, average bed temperature, coal feed rate, primary air volume, and secondary air volume, respectively.

[0010] Preferably, the ratio of primary and secondary air to bed material circulation is correlated using the mass conservation equation to derive... and A system of third-order nonlinear differential equations.

[0011] Preferably, when constructing the error feedback type data-driven module, an improved long short-term memory network is designed based on the output of the mechanism model to achieve real-time deviation compensation.

[0012] Preferably, the mechanism model is fused with the output of the data-driven module to obtain a mechanism-data fusion model, involving the following expression: ;in, The furnace outlet temperature after fusion; is the adaptive compensation term for operating conditions; k is the dynamic correction coefficient output by the data-driven module.

[0013] Preferably, when using the constructed mechanism data fusion model to predict the trend of furnace outlet temperature change in the next few seconds, the current state vector and real-time operating parameters are input, and the furnace outlet temperature prediction sequence for the next N times is output; the furnace outlet temperature prediction sequence is then used to generate the initial control sequence of secondary air volume and return valve opening through rolling time domain optimization.

[0014] Preferably, a convolutional neural network operating condition feature extractor is introduced to dynamically adjust and optimize the weight matrix of the objective function based on real-time operating data, so that the control sequence can adapt to coal type switching or load changes.

[0015] Preferably, when providing regulatory feedback on the implementation effect of correcting and optimizing the objective function weights, the regulatory evaluation array after correcting and optimizing the objective function weights is obtained and analyzed;

[0016] If all elements in the regulatory assessment array are 0, then the existing correction and optimization scheme will be maintained;

[0017] If there are non-zero elements in the regulatory assessment array, prompts will be given to add, delete, or modify the existing correction and optimization scheme rules.

[0018] Preferably, the regulatory assessment array includes a first element and a second element.

[0019] A circulating fluidized bed boiler furnace outlet temperature control system includes:

[0020] The hybrid model construction data processing module constructs a mechanism-data fusion dynamic model: Based on the laws of mass conservation and energy conservation, a furnace combustion heat transfer mechanism model is established, and the state vector is defined to include furnace outlet temperature, bed temperature, coal feed rate, primary air volume and secondary air volume; the error feedback type data-driven module compensates for the mechanism model deviation in real time, using the mechanism model prediction error and real-time operating parameters as input, and outputs dynamic correction coefficients;

[0021] The adaptive predictive regulatory analysis and control module designs an adaptive predictive controller based on a fusion model: it uses a constructed mechanism data fusion model to predict the furnace outlet temperature change trend in the next few seconds, uses a rolling time-domain optimization algorithm to generate an initial control sequence, uses a convolutional neural network operating condition feature extractor to perform feature dimensionality reduction on real-time operating data, dynamically adjusts the weight matrix of the predictive controller, and provides regulatory prompts on the implementation effect of correcting the weights of the objective function based on the calculated regulatory evaluation array.

[0022] Compared to existing solutions, the beneficial effects achieved by this invention are:

[0023] This invention uses the output error of the mechanistic model as input and integrates the predicted value of the mechanistic model with the data-driven correction coefficient to achieve a closed-loop logic of mechanistic modeling, error compensation, and dynamic fusion, which can ensure that the model has both physical interpretability and data adaptability.

[0024] This invention addresses the large time delay problem in boiler systems through multi-step prediction over several seconds, effectively reducing the impact of lag in traditional control and enabling control actions to adapt to disturbances in advance, achieving proactive control. It uses a convolutional neural network to capture coal type switching in real time and adjusts the weight matrix to quickly correct control variables, improving dynamic adaptability. By combining rolling time-domain optimization with operating condition classification weights, it effectively improves the control accuracy of furnace outlet temperature and reduces overshoot. Furthermore, by using constraints to limit bed temperature and control range in real time, it avoids the risk of coking or flameout due to coal type fluctuations, enhancing safety. Attached Figure Description

[0025] The invention will now be further described with reference to the accompanying drawings.

[0026] Figure 1 This is a flowchart illustrating the steps involved in implementing a circulating fluidized bed boiler furnace outlet temperature control method according to the present invention.

[0027] Figure 2 This is a block diagram of a circulating fluidized bed boiler furnace outlet temperature control system according to the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Example 1, such as Figure 1 As shown, the present invention is a method for controlling the furnace outlet temperature of a circulating fluidized bed boiler, comprising:

[0030] Step 1, Constructing a Mechanism-Data Fusion Dynamic Model: Based on the laws of mass and energy conservation, establish a furnace combustion heat transfer mechanism model, defining a state vector including furnace outlet temperature, bed temperature, coal feed rate, primary air volume, and secondary air volume; use an error feedback-type data-driven module to compensate for mechanism model deviations in real time, taking the mechanism model prediction error and real-time operating parameters as input, and outputting dynamic correction coefficients; specific steps include:

[0031] Based on the laws of mass and energy conservation, when constructing a dynamic mechanism model of the furnace of a circulating fluidized bed boiler, a state vector is set. ;in, These are the furnace outlet temperature, average bed temperature, coal feed rate, primary air volume, and secondary air volume, respectively.

[0032] The Arrhenius equation describes the combustion reaction rate of coal particles, and the relevant expression is:

[0033] Where A is the pre-exponential factor, E is the activation energy, and R is the gas constant;

[0034] Based on the one-dimensional heat transfer equation of a fluidized bed, an energy conservation relationship is established, involving the following expressions:

[0035] ;in, For bed heat capacity, Combustion releases heat. For heat loss in the furnace, This refers to the heat transfer between the bed and the heated surface;

[0036] By relating the ratio of primary and secondary air to the bed material circulation rate using the mass conservation equation, the following derivation is made. and The system of third-order nonlinear differential equations involves the following expressions:

[0037]

[0038] in, This refers to the bed material circulation rate; The temperature of the bed; k1 represents the furnace outlet temperature; k1, k2, and k3 are all material circulation coefficients; k4 is the reaction rate coefficient. hA represents the lower heating value of coal; hA represents the overall heat transfer coefficient of the bed. For ambient reference temperature; The return material temperature; The specific heat capacity of the flue gas; The density of the flue gas; This refers to the residence time of the flue gas. The sensible heat carried away by the circulating material;

[0039] It needs to be explained that, Combustion efficiency Related to its own nonlinearity, Influence of material sensible heat term , Simultaneously affected and The coupling effect;

[0040] Primary and secondary air ratio passed and Item adjustment Indirectly change and ;

[0041] The third-order nonlinear differential equation system needs to be solved by numerical methods, such as the Runge-Kutta method, which can reflect the dynamic temperature response under changes in coal type and air volume disturbance. The specific steps are not described here.

[0042] When constructing an error feedback type data-driven module, an improved long short-term memory network is designed based on the output of the mechanism model to achieve real-time deviation compensation;

[0043] The input layer uses a mechanistic model to predict errors. Real-time operating parameters are used as input; among them, To measure the outlet temperature, These are predicted values ​​from the mechanistic model; real-time operating parameters include the calorific value Q of the coal type and particle size distribution. Bed pressure ;

[0044] Introducing error gradient gating into the hidden layers of Long Short-Term Memory (LSTM) networks controls the rate of change of error. As a gating signal , The sampling time interval, When the error value from the previous time step is mapped to the historical error weight adjustment factor g(t) through a nonlinear transformation, if |e|>5℃, or When the temperature is greater than 2℃ / s, g(t) increases, which strengthens the weight of recent error.

[0045] When |e| < 1℃ and When <0.5℃ / s, g(t) approaches 0.5, balancing the weights of historical and recent errors;

[0046] The expression for the gating signal is: ;in, This is the sensitivity coefficient, with a value range of [0.8, 1.2]. This is a threshold parameter, with a value range of [1, 1.5].

[0047] The output layer uses the hyperbolic tangent activation function, and the range of the correction coefficient k is limited to [-0.2, 0.2] to avoid overcompensation;

[0048] The mechanism model is fused with the output of the data-driven module to obtain the mechanism-data fusion model, which involves the following expressions: ;in, The furnace outlet temperature after fusion; This is the adaptive compensation term based on operating conditions, obtained by weighting the coal type correction coefficient and the load fluctuation coefficient; the coal type correction coefficient is based on air-dried basis moisture content from industrial analysis data. Air-dried ash content The load fluctuation coefficient is calculated by taking the ratio of the current load to the rated load; k is the dynamic correction coefficient output by the data-driven module.

[0049] The fusion result is analyzed using a Kalman filter. State estimation is performed to suppress sensor noise; among which, the filter gain The gain is dynamically adjusted based on the variance of e, and is reduced when the variance of e is less than 0.5℃².

[0050] It should be noted that by ensuring the consistency of physical laws through the mechanistic model, the data-driven module corrects unmodeled dynamics such as coal type changes and equipment aging in real time through error feedback. Compared with the existing single mechanistic model for prediction under coal type switching conditions, it can effectively reduce prediction errors.

[0051] An improved error gradient gating mechanism for long short-term memory networks reduces compensation latency, achieves dynamic response speed optimization, and solves the lag problem of traditional data-driven models.

[0052] By integrating Kalman filtering with adaptive compensation terms based on operating conditions, the control accuracy can be effectively improved and the robustness of the system can be enhanced compared to existing single models.

[0053] In this embodiment of the invention, by taking the output error of the mechanism model as input, the fusion is achieved based on the predicted value of the mechanism model and the data-driven correction coefficient, realizing the closed-loop logic of mechanism modeling, error compensation and dynamic fusion, which can ensure that the model has both physical interpretability and data adaptability.

[0054] Step 2, design an adaptive predictive controller based on a fusion model: Utilize the constructed mechanistic data fusion model to predict the furnace outlet temperature change trend over the next few seconds; employ a rolling time-domain optimization algorithm to generate the initial control sequence; use a convolutional neural network operating condition feature extractor to perform feature dimensionality reduction on the real-time operating data; dynamically adjust the weight matrix of the predictive controller; and provide monitoring prompts based on the calculated monitoring evaluation array to assess the effectiveness of correcting the objective function weights. Specific steps include:

[0055] When using the constructed mechanistic data fusion model to predict the furnace outlet temperature change trend over several seconds (specifically, 5 seconds), the current state vector and real-time operating parameters are input; the state vector includes... , , , , Real-time operating parameters include air-dried basis moisture content. Air-dried ash content and current load ;

[0056] When making predictions, the third-order nonlinear differential equations of the fusion model are numerically integrated using the fourth-order Runge-Kutta method, with a time step of 0.5 seconds.

[0057] Predict the time domain N=10 steps, corresponding to 5 seconds; output the predicted furnace outlet temperature sequence for the next N time steps. ;

[0058] It should be noted that the data processing and analysis used in the mechanistic data fusion model for prediction is based on existing conventional technical solutions, and the specific implementation steps will not be elaborated here.

[0059] The predicted furnace outlet temperature sequence is used to generate the initial control sequence for the secondary air volume and the return valve opening through rolling time-domain optimization. The objective function involved is:

[0060] ;

[0061] Where k is the step index in the prediction time domain, k=1,2,3,…,N; These are all weighting coefficients, and the specific values ​​are not limited. The future furnace outlet temperature predicted by the fusion model at step k; This is the setpoint for the furnace outlet temperature; Let be the rate of change of the secondary air volume in the k-th step, which is the difference between the secondary air volume in the current step and the previous step. The rate of change of the return valve opening at step k is the difference between the opening at the current step and the previous step.

[0062] When setting upper and lower limits for control quantities and establishing safety constraints for bed temperature, , , This indicates the opening degree of the return valve in a circulating fluidized bed boiler, i.e., the degree to which the return valve is open. To avoid coking;

[0063] An improved particle swarm optimization algorithm is used to solve for the optimal control sequence in 50 iterations. Execute the control quantity at the current moment The next moment, the scrolling is repeated for optimization;

[0064] A convolutional neural network operating condition feature extractor is introduced, and the weight matrix of the objective function is dynamically adjusted and optimized based on real-time operating data, so that the control sequence can adapt to coal type switching or load changes.

[0065] The input layer of the convolutional neural network's operational feature extractor is used to input real-time operational data, consisting of 100 sets of historical data windows, including... Prediction error, air-dried basis moisture Air-dried ash content Load fluctuations ;

[0066] By using two convolutional layers and one max pooling layer, the input data is reduced to a 1×8 feature vector representing the working conditions. The convolutional kernels of the two convolutional layers are 3×3, and the activation function is ReLU; the stride of the max pooling layer is 2. These are: coal type stability index, load fluctuation intensity, temperature prediction error amplitude, error change rate index, bed temperature stability index, secondary air-load matching degree, return material system dynamic response index, and comprehensive operating condition complexity. The comprehensive operating condition complexity is obtained by weighted summation of the first seven features and is used to quantify the overall complexity of the operating condition.

[0067] eigenvectors Mapped to weight adjustment coefficients through fully connected layers W and b are training parameters, and σ is the Sigmoid function. Adjust and optimize the weights of the objective function:

[0068] ;

[0069] when hour, Increase the priority of temperature tracking to ensure that the greater the fluctuation in coal type, the higher the priority of temperature tracking.

[0070] when hour, , Reducing the value allows for greater adjustment of the control quantity, thus speeding up the response.

[0071] in, When the increase is reached, a piecewise linear-exponential mixed function is used to calculate the increase. :

[0072] ;in, This is the initial weight, with a default value of 10; These are the linear scaling factor and the exponential enhancement factor, respectively, with values ​​of 0.5 and 0.1.

[0073] , When decreasing, the decrease is calculated using the inverse hyperbolic tangent function. , :

[0074] Where j = 2, 3; , They share the same function, only their coefficients differ; All have different initial weights, with default values ​​of 1 and 0.5 respectively; Used to Mapped to attenuation coefficient;

[0075] When providing regulatory feedback on the implementation effect of correcting and optimizing the objective function weights, obtain and analyze the regulatory evaluation array after correcting and optimizing the objective function weights;

[0076] If all elements in the regulatory assessment array are 0, then the existing correction and optimization scheme will be maintained;

[0077] If there are non-zero elements in the regulatory assessment array, prompts will be given to add, delete, or modify the existing correction and optimization scheme rules;

[0078] The regulatory assessment array contains a first element and a second element. The steps for obtaining the first and second elements include:

[0079] After the system reaches steady state, specifically 30 seconds after the disturbance ends, calculate the measured value of the furnace outlet temperature. With design value The difference between the values ​​is analyzed. If the calculated difference is less than or equal to the temperature difference threshold, the value of the first element is set to 0. The design value and the temperature difference threshold can be determined according to the furnace's operating design parameters or actual operating parameter requirements. The specific values ​​are not limited.

[0080] If the calculated difference is greater than the temperature difference threshold, then the value of the first element is set to 1;

[0081] And, calculate the root mean square (RMSE) value of the temperature deviation:

[0082] ;in, This represents the total number of samples. The time step index indicates the number of steps. Each sampling time; For the first The actual measured value of the furnace outlet temperature in the step;

[0083] If the root mean square value is less than or equal to the standard error value, then the value of the second element is set to 0;

[0084] Conversely, the value of the second element is set to 1; the standard error value can also be determined according to the furnace's operating design parameters or actual operating parameter requirements, and the specific value is not limited.

[0085] It is worth noting that by combining multi-dimensional data supervision and processing to obtain a regulatory assessment array, and then conducting data analysis on the regulatory assessment array, the reliability and accuracy of the regulatory analysis prompts for the implementation effect of correcting and optimizing the objective function weights can be effectively improved.

[0086] It should be noted that by using multi-step prediction over several seconds to address the large time delay problem in boiler systems, the impact of traditional control lag can be effectively reduced, allowing control actions to adapt to disturbances in advance and achieving proactive control. Real-time capture of coal type switching through convolutional neural networks and rapid adaptive correction of control quantities through weight matrix adjustments improve dynamic adaptability. Rolling time-domain optimization combined with operating condition classification weights effectively enhances the control accuracy of furnace outlet temperature and reduces overshoot. Real-time limitation of bed temperature and control quantity range through constraint conditions avoids the risk of coking or flameout due to coal type fluctuations, enhancing safety.

[0087] Example 2, as Figure 2 As shown, a circulating fluidized bed boiler furnace outlet temperature control system includes:

[0088] The hybrid model construction data processing module constructs a mechanism-data fusion dynamic model: Based on the laws of mass conservation and energy conservation, a furnace combustion heat transfer mechanism model is established, and the state vector is defined to include furnace outlet temperature, bed temperature, coal feed rate, primary air volume and secondary air volume; the error feedback type data-driven module compensates for the mechanism model deviation in real time, using the mechanism model prediction error and real-time operating parameters as input, and outputs dynamic correction coefficients;

[0089] The adaptive predictive regulatory analysis and control module designs an adaptive predictive controller based on a fusion model: it uses a constructed mechanism data fusion model to predict the furnace outlet temperature change trend in the next few seconds, uses a rolling time-domain optimization algorithm to generate an initial control sequence, uses a convolutional neural network operating condition feature extractor to perform feature dimensionality reduction on real-time operating data, dynamically adjusts the weight matrix of the predictive controller, and provides regulatory prompts on the implementation effect of correcting the weights of the objective function based on the calculated regulatory evaluation array.

[0090] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0091] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0093] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A circulating fluidized bed boiler furnace outlet temperature control method, characterized by, include: Step 1, Constructing a Mechanism-Data Fusion Dynamic Model: Based on the laws of mass and energy conservation, establish a furnace combustion heat transfer mechanism model, defining a state vector including furnace outlet temperature, bed temperature, coal feed rate, primary air volume, and secondary air volume; use an error feedback-type data-driven module to compensate for mechanism model deviations in real time, taking the mechanism model prediction error and real-time operating parameters as input, and outputting dynamic correction coefficients; specific steps include: Based on the laws of mass and energy conservation, when constructing a dynamic mechanism model of the furnace of a circulating fluidized bed boiler, a state vector is set. ;in, These are the furnace outlet temperature, average bed temperature, coal feed rate, primary air volume, and secondary air volume, respectively. The Arrhenius equation describes the combustion reaction rate of coal particles, and the relevant expression is: Where A is the pre-exponential factor, E is the activation energy, and R is the gas constant; Based on the one-dimensional heat transfer equation of a fluidized bed, an energy conservation relationship is established, involving the following expressions: ;in, For bed heat capacity, Combustion releases heat. For heat loss in the furnace, This refers to the heat transfer between the bed and the heated surface; By relating the ratio of primary and secondary air to the bed material circulation rate using the mass conservation equation, the following derivation is made. and The system of third-order nonlinear differential equations involves the following expressions: ;in, This refers to the bed material circulation rate; The temperature of the bed; k1 represents the furnace outlet temperature; k1, k2, and k3 are all material circulation coefficients; k4 is the reaction rate coefficient. hA is the lower heating value of coal; hA is the overall heat transfer coefficient of the bed. For ambient reference temperature; The return material temperature; The specific heat capacity of the flue gas; The density of the flue gas; This refers to the residence time of the flue gas. The sensible heat carried away by the circulating material; When constructing an error feedback type data-driven module, an improved long short-term memory network is designed based on the output of the mechanism model to achieve real-time deviation compensation; The input layer uses a mechanistic model to predict errors. Real-time operating parameters are used as input; among them, To measure the outlet temperature, These are predicted values ​​from the mechanistic model; real-time operating parameters include the calorific value Q of the coal type and particle size distribution. Bed pressure ; Introducing error gradient gating into the hidden layers of Long Short-Term Memory (LSTM) networks controls the rate of change of error. As a gating signal , The sampling time interval, When the error value from the previous time step is mapped to the historical error weight adjustment factor g(t) through a nonlinear transformation, if |e|>5℃, or When the temperature is greater than 2℃ / s, g(t) increases, which strengthens the weight of recent error. When |e| < 1℃ and When <0.5℃ / s, g(t) approaches 0.5, balancing the weights of historical and recent errors; The expression for the gating signal is: ;in, This is the sensitivity coefficient, with a value range of [0.8, 1.2]. This is a threshold parameter, with a value range of [1, 1.5]. The output layer uses the hyperbolic tangent activation function, and the range of the correction coefficient k is limited to [-0.2, 0.2] to avoid overcompensation; The mechanism model is fused with the output of the data-driven module to obtain the mechanism-data fusion model, which involves the following expressions: ;in, The furnace outlet temperature after fusion; This is the adaptive compensation term based on operating conditions, obtained by weighting the coal type correction coefficient and the load fluctuation coefficient; the coal type correction coefficient is based on air-dried basis moisture content from industrial analysis data. Air-dried ash content The load fluctuation coefficient is calculated by taking the ratio of the current load to the rated load; k is the dynamic correction coefficient output by the data-driven module. The fusion result is analyzed using a Kalman filter. State estimation is performed to suppress sensor noise; among which, the filter gain The gain is dynamically adjusted based on the variance of e, and reduced when the variance of e is less than 0.5℃². Step 2: Design an adaptive predictive controller based on a fusion model: Use the constructed mechanism data fusion model to predict the furnace outlet temperature change trend in the next few seconds, use a rolling time-domain optimization algorithm to generate an initial control sequence, use a convolutional neural network operating condition feature extractor to perform feature dimensionality reduction on real-time operating data, dynamically adjust the weight matrix of the predictive controller, and provide monitoring prompts on the implementation effect of correcting the weights of the objective function based on the calculated monitoring evaluation array.

2. The method for controlling the furnace outlet temperature of a circulating fluidized bed boiler according to claim 1, characterized in that, Using the constructed mechanism data fusion model, when predicting the furnace outlet temperature change trend in the next few seconds, the current state vector and real-time operating parameters are input, and the furnace outlet temperature prediction sequence for the next N times is output. The furnace outlet temperature prediction sequence is then used to generate the initial control sequence for secondary air volume and return valve opening through rolling time domain optimization.

3. The method for controlling the furnace outlet temperature of a circulating fluidized bed boiler according to claim 2, characterized in that, A convolutional neural network operating condition feature extractor is introduced, and the weight matrix of the objective function is dynamically adjusted and optimized based on real-time operating data, so that the control sequence can adapt to coal type switching or load changes.

4. The method for controlling the furnace outlet temperature of a circulating fluidized bed boiler according to claim 3, characterized in that, When providing regulatory feedback on the implementation effect of correcting and optimizing the objective function weights, obtain and analyze the regulatory evaluation array after correcting and optimizing the objective function weights; If all elements in the regulatory assessment array are 0, then the existing correction and optimization scheme will be maintained; If there are non-zero elements in the regulatory assessment array, prompts will be given to add, delete, or modify the existing correction and optimization scheme rules.

5. The method for controlling the furnace outlet temperature of a circulating fluidized bed boiler according to claim 4, characterized in that, The regulatory assessment array contains a first element and a second element.

6. A circulating fluidized bed boiler furnace outlet temperature control system, used to execute the circulating fluidized bed boiler furnace outlet temperature control method as described in any one of claims 1-5, characterized in that, include: The hybrid model construction data processing module constructs a mechanism-data fusion dynamic model: Based on the laws of mass conservation and energy conservation, a furnace combustion heat transfer mechanism model is established, and the state vector is defined to include furnace outlet temperature, bed temperature, coal feed rate, primary air volume and secondary air volume; the error feedback type data-driven module compensates for the mechanism model deviation in real time, using the mechanism model prediction error and real-time operating parameters as input, and outputs dynamic correction coefficients; The adaptive predictive regulatory analysis and control module designs an adaptive predictive controller based on a fusion model: it uses a constructed mechanism data fusion model to predict the furnace outlet temperature change trend in the next few seconds, uses a rolling time-domain optimization algorithm to generate an initial control sequence, uses a convolutional neural network operating condition feature extractor to perform feature dimensionality reduction on real-time operating data, dynamically adjusts the weight matrix of the predictive controller, and provides regulatory prompts on the implementation effect of correcting the weights of the objective function based on the calculated regulatory evaluation array.

Images