Steam optimization production method and equipment for steam injection boiler of thickened oil recovery coal-fired fluidized bed

Abstract

The invention relates to the technical field of boilers, and provides a steam optimization production method and equipment for a steam injection boiler of a thickened oil recovery coal-fired fluidized bed, and the production method comprises the steps: 1, constructing a coupling model of a water supply quantity feedback controller and a steam dryness feedforward predictor, 2, training and optimizing the coupling model in a data driving mode, wherein the coupling model is used for predicting the dryness of steam produced under the data conditions of specific water supply quantity, fuel characteristics and fluidization air quantity; 3, inputting current water supply quantity, fuel characteristics and fluidization air quantity data, and predicting the dryness of steam produced under the current production condition by utilizing the trained model; 4, based on the predicted steam dryness, on the basis of steam dryness prediction, under the current water supply condition, on the basis of mean square errors of the predicted steam dryness and rated steam dryness, fuel characteristics and fluidization air volume data are optimized and adjusted, and steam optimized production is achieved; a feedforward predictor is constructed through a time-varying long-short-term memory network, and accurate capture is achieved.

Description

Technical Field

[0001] This invention relates to the field of boiler technology, specifically to a method and equipment for optimizing steam production in a coal-fired fluidized bed steam injection boiler used for heavy oil extraction. Background Technology

[0002] Heavy oil is difficult to extract due to its high viscosity and poor fluidity. Currently, widely used thermal recovery technology involves injecting high-pressure wet saturated steam to reduce crude oil viscosity, enhance its fluidity, and thus improve recovery rates. This technology has been successfully applied in oilfields such as Karamay. Steam injection boilers, as the core equipment in steam thermal recovery, are among the most energy-intensive devices in oilfield production, and their operating status directly affects extraction efficiency and economic costs. Steam dryness is a key parameter in the operation of steam injection boilers, affecting not only boiler safety performance but also being a core indicator determining the effectiveness of heavy oil thermal recovery.

[0003] Early steam injection boilers primarily used natural gas as fuel, but with rising natural gas prices, steam production costs have increased significantly. To reduce costs and ensure energy security, developing coal-fired fluidized bed steam injection boilers has become an important development direction. Heavy oil extraction requires wet saturated steam, with its dryness strictly controlled between 75% and 80% to prevent scaling and achieve optimal extraction results. However, controlling the steam dryness of coal-fired fluidized bed boilers is a complex process involving multiple variables, strong coupling, nonlinearity, and time-varying characteristics, making precise control difficult using traditional methods.

[0004] Currently, the control systems of steam injection boilers mainly include technologies such as cascade PID control, fuzzy PID control, and multivariable predictive control. While these methods have improved control performance to some extent, they still have significant limitations: on the one hand, they rely too heavily on precise mathematical models, making it difficult to handle the time-varying and nonlinear characteristics of actual production; on the other hand, they lack precision in modeling and optimizing multivariable coupling relationships, making the control effect susceptible to human experience interference, leading to large fluctuations in steam dryness, low thermal recovery efficiency, and potential environmental pollution problems. Therefore, there is an urgent need to introduce more advanced intelligent control methods to build a high-precision control system capable of adapting to complex operating conditions.

[0005] Based on the aforementioned problems, we propose a method and equipment for optimizing steam production using a coal-fired fluidized bed steam injection boiler in heavy oil extraction. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and equipment for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction. This method overcomes the deficiencies of existing technologies, has significant energy-saving and consumption-reducing effects, reduces reliance on manual experience, improves automation levels, and offers good economic and environmental benefits.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] Steam optimization production methods for coal-fired fluidized bed steam injection boilers in heavy oil extraction include:

[0009] Step 1: Construct a coupled model of a feedwater feedback controller and a steam dryness feedforward predictor to predict the dryness of the steam produced under specific feedwater flow, fuel characteristics, and fluidizing air volume conditions.

[0010] Step two: Train and optimize the coupled model using a data-driven approach;

[0011] Step 3: Input the current water supply, fuel characteristics, and fluidizing air volume data, and use the trained model to predict the dryness of the steam produced under the current production conditions;

[0012] Step four: Based on the predicted steam dryness, and under the current feedwater conditions, the mean square error between the predicted steam dryness and the rated steam dryness is used as a basis to optimize steam production by adjusting fuel characteristics and fluidizing air volume data.

[0013] Preferably, in the coupled model of the water supply feedback controller and the steam dryness feedforward predictor, the input of the feedback controller includes the current steam pressure value measured by the steam pressure sensor, the rated steam pressure value, and the time-delayed steam pressure value processed by the time delayer, and its output is the current water supply.

[0014] The inputs to the feedforward predictor include current feedwater flow, fuel characteristics, and fluidizing air volume data, as well as time-delayed feedwater flow, time-delayed fuel characteristics, and time-delayed fluidizing air volume data processed by the time delayer. Its output is the predicted result of the steam dryness produced under the current operating conditions.

[0015] Preferably, under the current water supply conditions, the optimal control of steam dryness is achieved by optimizing fuel characteristics and fluidizing air volume data. The objective function expression is as follows:

[0016]

[0017] in, , The optimal values ​​for fuel characteristics and fluidization air volume at time t are given. This represents the optimal value for steam dryness.

[0018] Preferably, the feedback controller measures the steam pressure value at time t using deployed steam pressure sensors. ,

[0019] The time-varying relationship between the steam pressure value at time t and the steam pressure value after a time delay is expressed as follows: ,in For time-varying delay variables,

[0020] The deviation between the steam pressure at time t and the rated steam pressure is calculated using the following expression: ,in, This is the rated steam pressure value.

[0021] Preferably, the feedback controller is a parameter-adaptive fuzzy PID controller; and The dataset is divided into seven fuzzy subsets: NB, NM, NS, ZO, PS, PM, and PB, where N represents negative, P represents positive, ZO represents 0, S represents small, M represents medium, and B represents large. Each subset uses a corresponding membership function. The fuzzification factor is used to further define the subsets. and Fuzzification is performed using preset fuzzy rules for inference. The output of the fuzzy rule inference is used to adjust the values ​​of the proportional weight coefficient k and the integral weight coefficient k' in real time. The weights k and k' are divided into fuzzy subsets S and B, and their membership functions are as follows:

[0022]

[0023]

[0024] in For k or , and, These represent the membership degrees of S and B, respectively.

[0025] Fuzzy rules are determined based on fuzzy subsets, in order to and Using fuzzy inference as input, the optimal proportional coefficient k and integral coefficient k' values ​​for the current moment are obtained, and then substituted into the control equations to calculate and output a unique optimal water supply control command. The expression is:

[0026] ,

[0027] in, The integral coefficients are small constants, based on the governing equations, and the input is... and Output the optimal water supply.

[0028] Preferably, a time-varying long short-term memory network is constructed as a steam dryness feedforward predictor:

[0029] Step 1: Collect the feedwater volume during the operation of the steam injection boiler. Fuel characteristics Fluidized air volume data Based on historical data and combined with a time delay device, the time-varying changes of each data point are calculated, and the expression is:

[0030]

[0031]

[0032]

[0033] in, For time-varying delay variables;

[0034] Step 2: Build a time-varying long short-term memory network model and train and optimize the model using historical data;

[0035] Step 3: Transfer real-time data , , and its corresponding time variables , The data is input into a time-varying long short-term memory network model after training to predict the steam dryness y(t) at the current moment.

[0036] Preferably, the time-varying long short-term memory network adopts a time-varying fine-grained neuron model, the structure of which includes a time-varying input gate, a time-varying forget gate, and an output gate; the expression for the time-varying input gate at time t is:

[0037]

[0038] in, For the input gate function, , For the input gate weights and gain parameters, For real-time water supply data or real-time fuel characteristic data Or real-time fluidization volume data , Time variable for water supply data or fuel characteristic data time-varying Or fluidized air volume data variables , The neuron output.

[0039] Preferably, the time-varying long short-term memory network obtains the predicted steam dryness value through the following forward computation process, where at time t, the network neurons receive real-time input data from the sensor network. , , and time variables , And, combined with the neuron's steady-state from the previous moment, the following gating and states are calculated sequentially:

[0040] The forget gate filters historical information. At time t, the output expression of the time-varying forget gate is:

[0041]

[0042] in, Forget gate function, , These are the forget gate weights and gain parameters;

[0043] The output gate focuses on modeling the variable at the current time. At time t, the output expression of the output gate is:

[0044]

[0045] in, For output gate function, , These are the output gate weights and gain parameters;

[0046] Subsequently, the state update of the neuron is calculated; based on the gating output and input information, the neuron state is divided into transient and stable states:

[0047] At time t, the transient state calculation expression is:

[0048]

[0049] in, Let be the transient state of the neuron at time t, and tanh be the hyperbolic tangent function. , These are the transient response weights and gain parameters of the neuron;

[0050] At time t, the steady-state calculation expression is:

[0051]

[0052] in, This represents the steady state of the neuron at time t. For delay The steady state of neurons at any given moment;

[0053] By integrating the historical state filtered by the forget gate, the transient state, and the time-varying features extracted by the output gate, dynamic state updates are achieved. Finally, steam dryness prediction is completed. Through the above-mentioned gating mechanism and progressive calculation of state updates, the time-varying long short-term memory network transforms real-time operating data into the dynamic state of neurons, and finally outputs the predicted steam dryness value y(t) under the current operating conditions.

[0054] A steam optimization production equipment for a coal-fired fluidized bed steam injection boiler used to realize the steam optimization production method of the steam injection boiler, comprising a steel frame, on which a coiled furnace is installed. The bottom and top of the coiled furnace are respectively provided with a coiled water-cooled air chamber and a steam-water separation device with a built-in separator; an air distribution plate is provided at the connection between the coiled water-cooled air chamber and the coiled furnace; an ignition burner is connected to the rear wall flange of the coiled water-cooled air chamber; a coal feeding device is provided on the side wall of the coiled furnace and communicates with it; and a slag discharge pipe is provided at the bottom of the ignition burner.

[0055] The outlet of the coil furnace is connected to the inlet of the cyclone separator. The cyclone separator is located on the side wall of the furnace and close to the steam-water separation device. The bottom of the cyclone separator is equipped with a return feeder connected to the coil furnace. The top outlet of the cyclone separator is connected to the flue. Inside the flue, along the flue gas flow direction, an evaporator convection tube bundle, a superheater, an economizer, a secondary air preheater, and a primary air preheater are arranged in sequence. An external energy-saving heat exchanger is provided between the economizer and the preheater. An ash hopper is provided at the bottom of the flue, and a flue gas outlet is provided on the side of the ash hopper. An expansion joint is provided between the coil furnace and the cyclone separator.

[0056] The outlet of the secondary air preheater is connected to the secondary air nozzle in the middle of the coil furnace; the outlet of the primary air preheater is connected to the primary air inlet of the coil water-cooled air chamber.

[0057] This invention provides a method and equipment for optimizing steam production in a coal-fired fluidized bed steam injection boiler used for heavy oil extraction. It offers the following advantages: by constructing a feedforward predictor using a time-varying long short-time memory network (TV-LSTM), the dynamic characteristics of the boiler operation can be accurately captured, enabling high-precision prediction of steam dryness and providing a reliable basis for optimized control.

[0058] By adopting a parameter-adaptive fuzzy PID control strategy, the control parameters are dynamically adjusted according to real-time operating conditions, which significantly enhances the system's adaptability to time-varying operating conditions and improves the accuracy and robustness of control.

[0059] By optimizing fuel characteristics and fluidized air volume data under current water supply conditions, closed-loop optimization control of steam dryness was achieved, effectively stabilizing the dryness within the ideal range (75%-80%), improving the thermal recovery effect of heavy oil while avoiding the risk of scaling. Attached Figure Description

[0060] Figure 1 This is a structural diagram of the coal-fired fluidized bed steam injection boiler of the present invention.

[0061] Figure 2 This invention presents a coupled model of a feedwater feedback controller and a steam dryness feedforward predictor for predicting steam dryness in steam production.

[0062] The attached diagram is labeled as follows: 1. Ignition burner; 2. Coal feeding device; 3. Coiled water-cooled air chamber; 4. Coiled furnace; 5. Steam-water separator; 6. Expansion joint; 7. Return feeder; 8. Cyclone separator; 9. Evaporator convection tube bundle; 10. Superheater; 11. Economizer; 12. Secondary air preheater; 13. Primary air preheater; 14. Flue gas outlet; 15. Ash hopper; 16. Steel frame; 17. External heat exchanger; 18. Air distribution plate; 19. Primary air inlet; 20. Secondary air nozzle; 21. Ash discharge pipe. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0064] See attached document Figure 2 As shown, the method for optimizing steam production from a coal-fired fluidized bed steam injection boiler in heavy oil extraction includes:

[0065] Step 1: Construct a coupled model of a feedwater feedback controller and a steam dryness feedforward predictor to predict the dryness of the steam produced under specific feedwater flow, fuel characteristics, and fluidizing air volume conditions.

[0066] Step two: Train and optimize the coupled model using a data-driven approach;

[0067] Step 3: Input the current water supply, fuel characteristics, and fluidizing air volume data, and use the trained model to predict the dryness of the steam produced under the current production conditions;

[0068] Step four: Based on the predicted steam dryness, and under the current feedwater conditions, using the mean square error between the predicted steam dryness and the rated steam dryness as a basis, steam production optimization is achieved by optimizing and adjusting fuel characteristics and fluidized air volume data. Finally, by combining the optimization control method with the coal-fired fluidized bed steam injection boiler body, a highly efficient steam production optimization solution is constructed.

[0069] In the coupled model of the water supply feedback controller and the steam dryness feedforward predictor, the input of the feedback controller includes the current steam pressure value measured by the steam pressure sensor, the rated steam pressure value, and the time-delayed steam pressure value processed by the time delayer, and its output is the current water supply.

[0070] The inputs to the feedforward predictor include current feedwater flow, fuel characteristics, and fluidizing air volume data, as well as time-delayed feedwater flow, time-delayed fuel characteristics, and time-delayed fluidizing air volume data processed by the time delayer. Its output is the predicted result of the steam dryness produced under the current operating conditions.

[0071] Under the current water supply conditions, the optimal control of steam dryness is achieved by optimizing fuel characteristics and fluidizing air volume data. The objective function expression is as follows:

[0072]

[0073] in, , The optimal values ​​for fuel characteristics and fluidization air volume at time t are given. The optimal value for steam dryness was determined; steam production optimization for coal-fired fluidized bed steam injection boilers in heavy oil extraction was completed;

[0074] The feedback controller measures the steam pressure value at time t using deployed steam pressure sensors. ,

[0075] The time-varying relationship between the steam pressure value at time t and the steam pressure value after a time delay is expressed as follows: ,in For time-varying delay variables,

[0076] The deviation between the steam pressure at time t and the rated steam pressure is calculated using the following expression: ,in, This is the rated steam pressure value.

[0077] The feedback controller is a parameter-adaptive fuzzy PID controller; and The dataset is divided into seven fuzzy subsets: NB, NM, NS, ZO, PS, PM, and PB, where N represents negative, P represents positive, ZO represents 0, S represents small, M represents medium, and B represents large. Each subset uses a corresponding membership function. The fuzzification factor is used to further define the subsets. and Fuzzification is performed using preset fuzzy rules for inference. The output of the fuzzy rule inference is used to adjust the values ​​of the proportional weight coefficient k and the integral weight coefficient k' in real time. The weights k and k' are divided into fuzzy subsets S and B, and their membership functions are as follows:

[0078]

[0079]

[0080] in For k or , and, These represent the membership degrees of S and B, respectively.

[0081] Fuzzy rules are determined based on fuzzy subsets, in order to and Using fuzzy inference as input, the optimal proportional coefficient k and integral coefficient k' values ​​for the current moment are obtained, and then substituted into the control equations to calculate and output a unique optimal water supply control command. The expression is:

[0082] ,

[0083] in, The integral coefficients are small constants, based on the governing equations, and the input is... and The system outputs the optimal water supply. It achieves feedback control of the optimal water supply H(t) by adaptively adjusting internal parameters k and k'.

[0084] Considering the time-varying characteristics of boiler operation, a time-varying long short-time memory network (TV-LSTM) is constructed as a steam dryness feedforward predictor:

[0085] Step 1: Collect the feedwater volume during the operation of the steam injection boiler. Fuel characteristics Fluidized air volume data Based on historical data and combined with a time delay device, the time-varying changes of each data point are calculated, and the expression is:

[0086]

[0087]

[0088]

[0089] in, For time-varying delay variables;

[0090] Step 2: Build a time-varying long short-term memory network model and train and optimize the model using historical data;

[0091] Step 3: Transfer real-time data , , and its corresponding time variables , The data is input into the trained time-varying long short-term memory (TV-LSTM) network model to predict the steam dryness y(t) at the current moment.

[0092] The time-varying long short-term memory network (TV-LSTM) employs a time-varying fine-grained neuron model, whose structure includes a time-varying input gate, a time-varying forget gate, and an output gate. The time-varying input gate at time t introduces a time variable to account for the significant time-varying characteristics during boiler operation, and its output expression is:

[0093]

[0094] in, For the input gate function, , For the input gate weights and gain parameters, For real-time water supply data or real-time fuel characteristic data Or real-time fluidization volume data , Time variable for water supply data or fuel characteristic data time-varying Or fluidized air volume data variables , The neuron output.

[0095] The time-varying long short-term memory network (TV-LSTM) obtains the predicted steam dryness value through the following forward computation process. At time t, the network neurons receive real-time input data from the sensor network. , , (i.e., water supply, fuel characteristics, fluidization air volume data) and their time variables , This, combined with the neuron's stable state at the previous moment, forms the basis of computation.

[0096] Secondly, a gating mechanism for time-varying characteristics is constructed; to capture the time-varying characteristics during boiler operation, a targeted gating unit is designed for the network:

[0097] Calculate the following gates and states in sequence:

[0098] The forget gate filters historical information at time t. The time-varying forget gate introduces a time variable to address the significant time-varying characteristics during boiler operation. The output expression is:

[0099]

[0100] in, Forget gate function, , These are the forget gate weights and gain parameters;

[0101] The output gate focuses on modeling the current time variable. At time t, the output gate models the time variable based on the more obvious time-varying characteristics during boiler operation, and the output expression is:

[0102]

[0103] in, For output gate function, , These are the output gate weights and gain parameters;

[0104] Subsequently, the state update of the neuron is calculated; based on the gating output and input information, the neuron state is divided into transient and stable states:

[0105] At time t, the transient state (currently assumed) is calculated as follows:

[0106]

[0107] in, Let be the transient state of the neuron at time t, and tanh be the hyperbolic tangent function. , These are the transient response weights and gain parameters of the neuron;

[0108] At time t, the steady state (current cell state) is calculated as follows:

[0109]

[0110] in, This represents the steady state of the neuron at time t. For delay The steady state of neurons at any given moment;

[0111] By integrating the historical state, transient state, and time-varying features extracted by the output gate, the dynamic update of the state is achieved.

[0112] Finally, the steam dryness prediction is completed. Through the above-mentioned gating mechanism and progressive calculation of state update, the time-varying long short-term memory network (TV-LSTM) transforms the real-time operating data into the dynamic state of the neurons, and finally outputs the steam dryness prediction value y(t) under the current operating conditions, thus completing the optimized prediction process of steam dryness.

[0113] See attached document Figure 1As shown, the steam optimization production equipment for a coal-fired fluidized bed steam injection boiler in heavy oil extraction is used to realize the steam optimization production method of the steam injection boiler. It includes a steel frame 16, on which a coiled furnace 4 is installed. The bottom and top of the coiled furnace 4 are respectively provided with a coiled water-cooled air chamber 3 and a steam-water separation device 5 with a built-in separator. An air distribution plate 18 is provided at the connection between the coiled water-cooled air chamber 3 and the coiled furnace 4. An ignition burner 1 is connected to the rear wall flange of the coiled water-cooled air chamber 3. A coal feeding device 2 communicating with the coiled furnace 4 is provided on the side wall of the furnace 4. A slag discharge pipe 21 is provided at the bottom of the ignition burner 1.

[0114] The outlet of the coil furnace 4 is connected to the inlet of the cyclone separator 8 (high-temperature insulation). The cyclone separator 8 is located on the side wall of the furnace and close to the steam-water separation device 5. The bottom of the cyclone separator 8 is provided with a return feeder 7 connected to the coil furnace 4. The top outlet of the cyclone separator 8 is connected to the flue. The flue contains, in sequence along the flue gas flow direction, an evaporator convection tube bundle 9, a superheater 10, an economizer 11, a secondary air preheater 12, and a primary air preheater 13. An (energy-saving) external heat exchanger 17 is provided between the economizer 11 and the preheater. The bottom of the flue contains an ash hopper 15, and the side of the ash hopper 15 contains a flue gas outlet 14. An expansion joint 6 is provided between the coil furnace 4 and the cyclone separator 8.

[0115] The outlet of the secondary air preheater 12 is connected to the secondary air nozzle 20 in the middle of the coil furnace 4; the outlet of the primary air preheater 13 is connected to the primary air inlet 19 of the coil water-cooled air chamber 3.

[0116] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing steam production in a coal-fired fluidized bed steam injection boiler used for heavy oil extraction, characterized in that: include: Step 1: Construct a coupled model of a feedwater feedback controller and a steam dryness feedforward predictor to predict the dryness of the steam produced under specific feedwater flow, fuel characteristics, and fluidizing air volume conditions. Step two: Train and optimize the coupled model using a data-driven approach; Step 3: Input the current water supply, fuel characteristics, and fluidizing air volume data, and use the trained model to predict the dryness of the steam produced under the current production conditions; Step four: Based on the predicted steam dryness, and under the current feedwater conditions, the mean square error between the predicted steam dryness and the rated steam dryness is used as a basis to optimize steam production by adjusting fuel characteristics and fluidizing air volume data.

2. The method for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction according to claim 1, characterized in that, In the coupled model of the water supply feedback controller and the steam dryness feedforward predictor, the input of the feedback controller includes the current steam pressure value measured by the steam pressure sensor, the rated steam pressure value, and the time-delayed steam pressure value processed by the time delayer, and its output is the current water supply. The inputs to the feedforward predictor include current feedwater flow, fuel characteristics, and fluidizing air volume data, as well as time-delayed feedwater flow, time-delayed fuel characteristics, and time-delayed fluidizing air volume data processed by the time delayer. Its output is the predicted result of the steam dryness produced under the current operating conditions.

3. The method for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction according to claim 2, characterized in that, Under the current water supply conditions, the optimal control of steam dryness is achieved by optimizing fuel characteristics and fluidizing air volume data. The objective function expression is as follows: in, , The optimal values ​​for fuel characteristics and fluidization air volume at time t are given. This represents the optimal value for steam dryness.

4. The method for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction according to claim 2, characterized in that, The feedback controller measures the steam pressure value at time t using deployed steam pressure sensors. , The time-varying relationship between the steam pressure value at time t and the steam pressure value after a time delay is expressed as follows: ,in For time-varying time delay variables, The deviation between the steam pressure at time t and the rated steam pressure is calculated using the following expression: ,in, This is the rated steam pressure value.

5. The method for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction according to claim 2, characterized in that, The feedback controller is a parameter-adaptive fuzzy PID controller; and The dataset is divided into seven fuzzy subsets: NB, NM, NS, ZO, PS, PM, and PB, where N represents negative, P represents positive, ZO represents 0, S represents small, M represents medium, and B represents large. Each subset uses a corresponding membership function. The fuzzification factor is used to further define the subsets. and Fuzzification is performed using preset fuzzy rules for inference. The output of the fuzzy rule inference is used to adjust the values ​​of the proportional weight coefficient k and the integral weight coefficient k' in real time. The weights k and k' are divided into fuzzy subsets S and B, and their membership functions are as follows: in For k or , and, These represent the membership degrees of S and B, respectively. Fuzzy rules are determined based on fuzzy subsets, in order to and Using fuzzy inference as input, the optimal proportional coefficient k and integral coefficient k' values ​​for the current moment are obtained, and then substituted into the control equations to calculate and output a unique optimal water supply control command. The expression is: , in, The integral coefficients are small constants, based on the governing equations, and the input is... and Output the optimal water supply.

6. The method for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction according to claim 1, characterized in that, Constructing a time-varying long short-term memory network as a steam dryness degree feedforward predictor: Step 1: Collect the feedwater volume during the operation of the steam injection boiler. Fuel characteristics Fluidized air volume data Based on historical data and combined with a time delay device, the time-varying changes of each data point are calculated, and the expression is: in, For time-varying delay variables; Step 2: Build a time-varying long short-term memory network model and train and optimize the model using historical data; Step 3: Transfer real-time data , , and its corresponding time variables , The data is input into a time-varying long short-term memory network model after training to predict the steam dryness y(t) at the current moment.

7. The method for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction according to claim 6, characterized in that, The time-varying long short-term memory network employs a time-varying fine-grained neuron model, whose structure includes a time-varying input gate, a time-varying forget gate, and an output gate; the expression for the time-varying input gate at time t is: in, For the input gate function, , For the input gate weights and gain parameters, For real-time water supply data or real-time fuel characteristic data Or real-time fluidization volume data , Time variable for water supply data or fuel characteristic data time-varying Or fluidized air volume data variables , The neuron output.

8. The method for optimizing steam production in a coal-fired fluidized bed steam injection boiler for heavy oil extraction according to claim 7, characterized in that, The time-varying long short-term memory network obtains the predicted steam dryness value through the following forward computation process. At time t, the network neurons receive real-time input data from the sensor network. , , and time variables , And, combined with the neuron's steady-state from the previous moment, the following gating and states are calculated sequentially: The forget gate filters historical information. At time t, the output expression of the time-varying forget gate is: in, Forget gate function, , These are the forget gate weights and gain parameters; The output gate focuses on modeling the variable at the current time. At time t, the output expression of the output gate is: in, For output gate function, , These are the output gate weights and gain parameters; Subsequently, the state update of the neuron is calculated; based on the gating output and input information, the neuron state is divided into transient and stable states: At time t, the transient state calculation expression is: in, Let be the transient state of the neuron at time t, and tanh be the hyperbolic tangent function. , These are the transient response weights and gain parameters of the neuron; At time t, the steady-state calculation expression is: in, This represents the steady state of the neuron at time t. For delay The steady state of neurons at any given moment; By integrating the historical state filtered by the forget gate, the transient state, and the time-varying features extracted by the output gate, dynamic state updates are achieved. Finally, steam dryness prediction is completed. Through the above-mentioned gating mechanism and progressive calculation of state updates, the time-varying long short-term memory network transforms real-time operating data into the dynamic state of neurons, and finally outputs the predicted steam dryness value y(t) under the current operating conditions.

9. Steam optimization production equipment for a coal-fired fluidized bed steam injection boiler used to implement the steam optimization production method of the steam injection boiler as described in any one of claims 1 to 8, characterized in that, The furnace includes a steel frame on which a coiled furnace is mounted. The bottom and top of the coiled furnace are respectively equipped with a coiled water-cooled air chamber and a steam-water separation device with a built-in separator. An air distribution plate is provided at the connection between the coiled water-cooled air chamber and the coiled furnace. An ignition burner is connected to the rear wall flange of the coiled water-cooled air chamber. A coal feeding device communicating with the side wall of the coiled furnace is provided. A slag discharge pipe is provided at the bottom of the ignition burner. The outlet of the coil furnace is connected to the inlet of the cyclone separator. The cyclone separator is located on the side wall of the furnace and close to the steam-water separation device. The bottom of the cyclone separator is equipped with a return feeder connected to the coil furnace. The top outlet of the cyclone separator is connected to the flue. Inside the flue, along the flue gas flow direction, an evaporator convection tube bundle, a superheater, an economizer, a secondary air preheater, and a primary air preheater are arranged in sequence. An external energy-saving heat exchanger is provided between the economizer and the preheater. An ash hopper is provided at the bottom of the flue, and a flue gas outlet is provided on the side of the ash hopper. An expansion joint is provided between the coil furnace and the cyclone separator. The outlet of the secondary air preheater is connected to the secondary air nozzle in the middle of the coil furnace; the outlet of the primary air preheater is connected to the primary air inlet of the coil water-cooled air chamber.

Images