A control method for flue gas heat exchanger in thermal power plants based on dynamic matrix control

Abstract

The present invention discloses a control method for a flue gas heat exchanger in a thermal power plant based on dynamic matrix control, and relates to the technical field of heat exchanger control. The present invention adopts LSTM to construct a flue gas temperature prediction model, captures the influence of boiler load fluctuations and changes in coal combustion characteristics on flue gas temperature, predicts future temperature change trends in advance, and generates temperature change adjustment instructions, so that the heat exchanger can actively adapt to flue gas temperature fluctuations and reduce adjustment lag; establishes a heat exchanger state space model based on LSTM prediction results, combines historical data and future prediction inputs, calculates the optimal heat exchange medium flow adjustment instructions, avoids overshoot or lag of PID control under load mutation conditions, and improves temperature regulation accuracy; adopts a rolling optimization strategy, so that the heat exchanger control integrates the current state and optimizes the heat exchange medium flow in multiple time steps, thereby improving the foresight and stability of control.

Description

Technical Field

[0001] The present invention relates to the technical field of heat exchanger control, in particular to a control method for a flue gas heat exchanger in a thermal power plant based on dynamic matrix control. Background Art

[0002] In thermal power plants, the high-temperature flue gas waste heat recovery system undertakes the important task of improving energy utilization and reducing fuel consumption. The function of the heat exchanger is to use the high-temperature flue gas discharged from the boiler to efficiently transfer heat to the working fluid, thereby reducing energy loss.

[0003] However, this process is greatly affected by multiple variables such as flue gas temperature, flow rate, coal characteristics and boiler operating load. In the actual operation of thermal power plants, rapid fluctuations in boiler load will cause instantaneous changes in flue gas temperature. The fluctuations are more obvious during the start-up and shutdown process or when the coal quality changes. When the boiler load increases, the combustion intensity increases and the exhaust flue gas temperature rises rapidly. The heat exchanger needs to adapt to the new heat load in a short time, otherwise the outlet temperature may rise sharply, affecting the stability of subsequent systems. On the contrary, when the load decreases, the flue gas temperature drops. If the heat exchanger cannot quickly adjust the heat exchange capacity, the outlet temperature may drop temporarily, affecting the waste heat utilization efficiency and even affecting the normal operation of downstream equipment such as the denitrification system.

[0004] Among existing control methods, PID control is the most common means of regulation. However, flue gas temperature does not change smoothly, but is affected by boiler load adjustment, coal type change, and even combustion organization mode, showing large dynamic fluctuations. Traditional PID control relies on fixed parameters, and when faced with rapidly changing operating conditions, the regulation response often lags. In addition to PID control, bypass flue gas regulation is also a common supplementary means, but this method relies on the adjustment of mechanical dampers, which has a certain inertia and responds slowly to sudden changes in boiler operating conditions. Another common regulation method is induced draft fan variable frequency control, which indirectly affects the heat exchange intensity of the heat exchanger by adjusting the flue gas flow rate, reducing the residence time of the flue gas in the heat exchanger. However, the response inertia of the induced draft fan is large, making it difficult to meet the needs of rapidly changing boiler operating conditions. When the coal quality changes or the boiler load is adjusted quickly, the adjustment of the induced draft fan often lags behind the change in flue gas temperature.

[0005] To address these issues, some traditional solutions have introduced adaptive PID control, which automatically adjusts control parameters based on changing operating conditions to improve regulation adaptability. However, adaptive PID still has certain limitations when dealing with nonlinear operating conditions, making it difficult to completely eliminate overshoot and lag. Intelligent bypass control is limited by the mechanical adjustment speed and cannot adapt to large fluctuations in a short period of time.

[0006] In summary, in the operating environment of a thermal power plant, the rapid fluctuations in boiler load, changes in coal characteristics, and nonlinear dynamic changes in flue gas temperature cause traditional control strategies to have a delayed response and insufficient adjustment accuracy when dealing with rapidly changing operating conditions. This leads to large fluctuations in the outlet temperature of the heat exchanger, affecting the waste heat recovery efficiency, causing a chain reaction on subsequent denitrification, emission control and other systems, and affecting the overall stability and economy of the thermal power plant. Therefore, there is an urgent need for a more intelligent and faster-responding thermal power plant flue gas heat exchanger control strategy based on dynamic matrix control to improve the adaptive adjustment capability of the heat exchanger. Summary of the Invention

[0007] In view of the above existing problems, the present invention is proposed.

[0008] The present invention provides a control method for a flue gas heat exchanger in a thermal power plant based on dynamic matrix control to solve the problems of traditional PID control hysteresis, insufficient bypass flue gas regulation accuracy, slow response of the induced draft fan, difficulty in coping with rapid fluctuations in flue gas temperature, and impact on the stability of the heat exchanger outlet temperature.

[0009] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0010] The embodiment of the present invention provides a control method for a flue gas heat exchanger in a thermal power plant based on dynamic matrix control, which includes:

[0011] Step S1: Install a temperature sensor, a flow meter, and a pressure sensor at the inlet and outlet of the flue gas heat exchanger, respectively, to collect flue gas sensor data and boiler operation data in real time and generate a current operating condition data set;

[0012] Step S2: Based on the current working condition data set of step S1, a flue gas temperature prediction model is constructed using a long short-term memory network (LSTM) to predict the flue gas temperature change trend in the future time period and generate a temperature change adjustment instruction;

[0013] Step S3, receiving the temperature change adjustment instruction generated in step S2, using dynamic matrix control (DMC) to predictively control the outlet temperature of the heat exchanger, and generating a heat exchange medium flow adjustment instruction;

[0014] Step S4, using the heat exchange medium flow adjustment instruction of step S3 to control the multi-channel heat exchange tube bundle structure inside the heat exchanger, and dynamically distribute the flue gas flow entering different heat exchange areas;

[0015] Step S5: Based on the temperature adjustment feedback data at the heat exchanger outlet, the DMC control parameters of step S3 are adaptively optimized using a reinforcement learning algorithm.

[0016] As a preferred solution of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to the present invention, the flue gas sensor data includes flue gas temperature, flue gas flow, flue gas pressure and heat exchange medium temperature;

[0017] The boiler operation data includes boiler load changes, coal combustion characteristics, and air flow.

[0018] As a preferred solution of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control described in the present invention, the step of constructing a flue gas temperature prediction model using a long short-term memory network LSTM is as follows:

[0019] Construct a flue gas temperature prediction model and collect the current operating condition data set D t , used for LSTM network training, the dataset is represented as D t :

[0020] D t ={T g ,Q g ,P g ,T m ,L,C,Q a},

[0021] Among them, T g Indicates the flue gas temperature, Q g Indicates the flue gas flow rate, P g Indicates the flue gas pressure, T m represents the temperature of the heat exchange medium, L represents the boiler load, C represents the coal characteristics, Q a Indicates air flow;

[0022] The state update process of LSTM cells includes:

[0023] i t =σ(W i [h t-1 ,x t ]+b i ),

[0024] f t =σ(W f [h t-1 ,x t ]+b f ),

[0025] o t =σ(W o [h t-1 ,x t ]+b o ),

[0026]

[0027] h t =o t ⊙tanh(C t ),

[0028] Among them, i t represents the input gate activation value, f t Represents the activation value of the forget gate, o t represents the output gate activation value, represents the candidate state value, C t Indicates the unit state value, C t1 Indicates the unit state value of the previous time step, h t represents the hidden state value, h t1 Represents the hidden state value of the previous time step, W i ,W f ,W o ,W c is the weight matrix, b i ,b f ,b o ,b c is the bias term, x t Represents input data, σ represents the Sigmoid activation function, tanh represents the hyperbolic tangent activation function, and ⊙ represents element-by-element multiplication operation.

[0029] As a preferred solution of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to the present invention, the step of predicting the flue gas temperature change trend in a future time period and generating a temperature change adjustment instruction is as follows:

[0030] The hidden state h calculated based on LSTM t , predict the flue gas temperature in the future time step, the prediction formula is:

[0031]

[0032] in, represents the predicted flue gas temperature at time t+τ, f(·) is the LSTM prediction model, and θ is the parameter set obtained by LSTM training.

[0033] Generate temperature change adjustment instructions, the generation process includes:

[0034] Temperature change trend, calculated as:

[0035]

[0036] Where, ΔT g is the flue gas temperature change, To predict the flue gas temperature, T g,tis the flue gas temperature at the current time step t,

[0037] If |ΔT g |>δ T , then generate temperature change adjustment instructions:

[0038] U T =g(ΔT g ),

[0039] Among them, δ T is the temperature change threshold, U T represents the temperature change adjustment instruction, and g(·) is the adjustment function.

[0040] As a preferred solution of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control described in the present invention, in step S3, a state space model of the heat exchanger is established, historical temperature change data, current measurement data and future predicted temperature trends are integrated to calculate a heat exchange medium flow adjustment plan and generate a heat exchange medium flow adjustment instruction.

[0041] As a preferred embodiment of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to the present invention, the steps of predictively controlling the outlet temperature of the heat exchanger by using dynamic matrix control (DMC) and generating a heat exchange medium flow adjustment instruction are as follows:

[0042] The heat exchanger state space model is established, and the heat exchanger state space equation is defined as:

[0043] X t+1 =AX t +BU t ,

[0044] Y t =CX t +DU t ,

[0045] Among them, X t+1 represents the state variable vector for the next time step, X t Represents the state variable vector of the current time step, U t Represents the input variable vector of the current time step, Y t represents the output variable vector of the current time step, A represents the heat exchanger state transfer matrix, which describes the evolution of the heat exchanger state over time, B represents the input influence matrix, which describes the influence of the heat exchange medium flow on the heat exchanger state, C represents the measurement matrix, which maps the state variables to the output variables, and D represents the input direct influence matrix, which describes the direct influence of the input on the output.

[0046] Temperature changes are predicted based on historical data and future inputs. The prediction formula is:

[0047]

[0048] in, is the heat transfer medium temperature predicted at the future time step t+τ, X t+τ is the state variable at time step t+τ, U t+τ is the input variable at time step t+τ,

[0049] DMC uses a rolling optimization strategy to calculate the optimal heat exchange medium flow regulation instruction. The calculation formula is:

[0050]

[0051] in, is the optimal heat exchange medium flow regulation instruction, is the expected heat transfer medium temperature, N is the prediction step length, w τ is the weight coefficient of the time step τ, λ is the regularization coefficient, ΔU m The input change of heat exchange medium flow control, U m Represents the control input for the heat transfer medium flow rate.

[0052] As a preferred solution of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control described in the present invention, wherein:

[0053] In step S4, the heat exchange area is divided into a high temperature area, a medium temperature area, and a low temperature area;

[0054] In step S4, according to the heat exchange medium flow adjustment instruction, the adjustable baffle or rotary valve is used to adjust the flue gas flow ratio entering each zone, and heat exchanger outlet temperature adjustment feedback data is generated.

[0055] As a preferred embodiment of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to the present invention, the steps of using the heat exchange medium flow rate adjustment instruction of step S3 to control the multi-channel heat exchange tube bundle structure inside the heat exchanger and dynamically distribute the flue gas flow entering different heat exchange areas are as follows:

[0056] The internal area of the heat exchanger is divided into:

[0057] High temperature zone, used to initially reduce the flue gas temperature,

[0058] Medium temperature zone, used for further heat exchange,

[0059] and a low-temperature zone for final smoke exhaust;

[0060] The flue gas flow ratio is adjusted by adjusting the damper or rotary valve. The adjustment process is expressed as:

[0061] Q g,i=α i Q g ,

[0062] Among them, Q g,i represents the flue gas flow rate entering the i-th heat exchange area, Q g represents the total flue gas flow, α i is the distribution coefficient, satisfying ∑ i α i =1.

[0063] As a preferred solution of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control described in the present invention, wherein:

[0064] In step S5, the deviation between the predicted temperature and the actual outlet temperature is compared to determine the error range of the DMC control strategy;

[0065] In step S5, the DMC control weight is updated based on the error range, and the zone heat exchange adjustment parameters are adjusted.

[0066] As a preferred embodiment of the method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to the present invention, the step of adaptively optimizing the DMC control parameters of step S3 based on the temperature adjustment feedback data at the heat exchanger outlet and the reinforcement learning algorithm is as follows:

[0067] The prediction error of the heat exchanger outlet temperature is calculated at each time step t, and the calculation formula is:

[0068]

[0069] Among them, e t represents the temperature prediction error, represents the predicted temperature of the heat transfer medium at time step t, T m,t represents the actual measured temperature of the heat transfer medium at time step t,

[0070] Set the reinforcement learning goal, construct the objective function J, and measure the optimization degree of the control strategy:

[0071]

[0072] Where J represents the control optimization objective function, T is the number of time steps in the optimization cycle, λ is the regularization coefficient, and Δw τ represents the change in the weight of the control parameter,

[0073] The gradient descent method is used to update the DMC control weight, and the update formula is:

[0074]

[0075] in, represents the weight of the DMC control parameter after the update at time step t+1, represents the current DMC control parameter weight at time step t, η is the learning rate, Denotes the loss function J with respect to w τ The gradient,

[0076] Based on the optimized DMC control weight, the flue gas flow distribution in the heat exchange area is adjusted. The adjustment formula is:

[0077]

[0078] in, represents the flue gas flow distribution coefficient of the ith heat exchange area after the update at time step t+1, It represents the flue gas flow distribution coefficient of the i-th heat exchange area at time step t, and γ is the adjustment step size.

[0079] The beneficial effects of the present invention are as follows: the present invention adopts LSTM to construct a flue gas temperature prediction model, captures the influence of boiler load fluctuations and changes in coal combustion characteristics on flue gas temperature, predicts future temperature change trends in advance, and generates temperature change adjustment instructions, so that the heat exchanger can actively adapt to flue gas temperature fluctuations and reduce adjustment lag; based on the LSTM prediction results, a heat exchanger state space model is established, and the optimal heat exchange medium flow adjustment instructions are calculated based on historical data and future prediction inputs, avoiding overshoot or lag of PID control under load mutation conditions, thereby improving temperature regulation accuracy.

[0080] The present invention adopts a rolling optimization strategy to make the heat exchanger control comprehensively consider the current state and optimize the heat exchange medium flow within multiple time steps, thereby improving the foresight and stability of the control; at the heat exchange execution level, combined with the zoned dynamic heat exchange strategy, the interior of the heat exchanger is divided into high-temperature zone, medium-temperature zone and low-temperature zone, and an adjustable baffle or rotary valve is used to dynamically adjust the flue gas flow distribution of each zone according to the heat exchange medium flow adjustment instruction calculated by DMC, thereby optimizing the heat load distribution, improving the heat exchange efficiency, and reducing local overheating or uneven temperature distribution.

[0081] The present invention adopts a reinforcement learning algorithm to optimize the DMC control weights, and through a feedback adjustment strategy, dynamically adjusts the control parameters according to historical errors, thereby improving the adaptability of the heat exchanger under different boiler load conditions. Specifically, the temperature prediction error is calculated by the reinforcement learning objective function, and the DMC weights are optimized in combination with the gradient descent method. The adjustment strategy is continuously optimized during long-term operation to improve the stability of temperature control. Compared with traditional PID and bypass flue gas adjustment schemes, it can respond more quickly, reduce large fluctuations in outlet temperature, and improve the efficiency of waste heat utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0083] Figure 1 The figure is a flow chart of a method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to the present invention. DETAILED DESCRIPTION

[0084] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0085] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0086] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0087] Example 1, reference Figure 1 This embodiment provides a method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control, comprising the following steps:

[0088] Step S1: Install a temperature sensor, a flow meter, and a pressure sensor at the inlet and outlet of the flue gas heat exchanger, respectively, to collect flue gas sensor data and boiler operation data in real time and generate a current operating condition data set;

[0089] Flue gas sensing data includes flue gas temperature, flue gas flow, flue gas pressure and heat exchange medium temperature;

[0090] Boiler operating data includes boiler load changes, coal characteristics, and air flow;

[0091] Step S2: Based on the current working condition data set of step S1, a flue gas temperature prediction model is constructed using a long short-term memory network (LSTM) to predict the flue gas temperature change trend in the future time period and generate a temperature change adjustment instruction;

[0092] The flue gas temperature prediction model is constructed using the long short-term memory network LSTM to predict the flue gas temperature change trend in the future time period. The steps for generating temperature change adjustment instructions are as follows:

[0093] Construct a flue gas temperature prediction model and collect the current operating condition data set D t , used for LSTM network training, the dataset is represented as D t :

[0094] D t ={T g ,Q g ,P g ,T m ,L,C,Q a},

[0095] Among them, T g Indicates the flue gas temperature, Q g Indicates the flue gas flow rate, P g Indicates the flue gas pressure, T m represents the temperature of the heat exchange medium, L represents the boiler load, C represents the coal characteristics, Q a Indicates air flow;

[0096] The state update process of LSTM cells includes:

[0097] i t =σ(W i [h t-1 ,x t ]+b i ),

[0098] f t =σ(W f [h t-1 ,x t ]+b f ),

[0099] o t =σ(W o [h t-1 ,x t ]+b o ),

[0100]

[0101] h t =o t ⊙tanh(C t ),

[0102] Among them, i t represents the input gate activation value, f t Represents the activation value of the forget gate, o t represents the output gate activation value, represents the candidate state value, C t Indicates the unit state value, C t1 Indicates the unit state value of the previous time step, h t represents the hidden state value, h t1 Represents the hidden state value of the previous time step, W i ,W f ,W o ,W c is the weight matrix, b i ,b f ,b o ,b c is the bias term, x t Represents input data, σ represents Sigmoid activation function, tanh represents hyperbolic tangent activation function, and ⊙ represents element-by-element multiplication operation;

[0103] The hidden state h calculated based on LSTM t , predict the flue gas temperature in the future time step, the prediction formula is:

[0104]

[0105] in, represents the predicted flue gas temperature at time t+τ, f(·) is the LSTM prediction model, and θ is the parameter set obtained by LSTM training.

[0106] Generate temperature change adjustment instructions, the generation process includes:

[0107] Temperature change trend, calculated as:

[0108]

[0109] Where, ΔT g is the flue gas temperature change, To predict the flue gas temperature, T g,t is the flue gas temperature at the current time step t,

[0110] If |ΔT g |>δ T , then generate temperature change adjustment instructions:

[0111] U T =g(ΔT g ),

[0112] Among them, δ T is the temperature change threshold, U T represents the temperature change adjustment instruction, g(·) is the adjustment function;

[0113] Specifically, this step uses a long short-term memory network to model the flue gas temperature variation trend. A time series dataset is constructed based on boiler operation data and flue gas sensor data. LSTM captures the long-term dependency of temperature changes through a gating mechanism and predicts the flue gas temperature in future time steps based on hidden states. When the temperature prediction error is large, the system generates a temperature change adjustment instruction to dynamically adjust the flue gas temperature.

[0114] Step S3, receiving the temperature change adjustment instruction generated in step S2, using dynamic matrix control (DMC) to predictively control the outlet temperature of the heat exchanger, and generating a heat exchange medium flow adjustment instruction;

[0115] In step S3, a state space model of the heat exchanger is established, and the historical temperature change data, current measurement data, and future predicted temperature trends are integrated to calculate the heat exchange medium flow adjustment plan and generate a heat exchange medium flow adjustment instruction;

[0116] Dynamic matrix control (DMC) is used to predictively control the outlet temperature of the heat exchanger. The steps to generate the heat exchange medium flow regulation instruction are as follows:

[0117] The heat exchanger state space model is established, and the heat exchanger state space equation is defined as:

[0118] X t+1 =AX t +BU t ,

[0119] Y t =CX t +DU t ,

[0120] Among them, X t+1 represents the state variable vector for the next time step, X t Represents the state variable vector of the current time step, U t Represents the input variable vector of the current time step, Y t represents the output variable vector of the current time step, A represents the heat exchanger state transfer matrix, which describes the evolution of the heat exchanger state over time, B represents the input influence matrix, which describes the influence of the heat exchange medium flow on the heat exchanger state, C represents the measurement matrix, which maps the state variables to the output variables, and D represents the input direct influence matrix, which describes the direct influence of the input on the output.

[0121] Temperature changes are predicted based on historical data and future inputs. The prediction formula is:

[0122]

[0123] in, is the heat transfer medium temperature predicted at the future time step t+τ, X t+τis the state variable at time step t+τ, U t+τ is the input variable at time step t+τ,

[0124] DMC uses a rolling optimization strategy to calculate the optimal heat exchange medium flow regulation instruction. The calculation formula is:

[0125]

[0126] in, is the optimal heat exchange medium flow regulation instruction, is the expected heat transfer medium temperature, N is the prediction step length, w τ is the weight coefficient of the time step τ, λ is the regularization coefficient, ΔU m The input change of heat exchange medium flow control, U m The control input representing the flow rate of the heat transfer medium;

[0127] Specifically, based on the state-space model of the heat exchanger, dynamic matrix control is used for predictive control of the outlet temperature. DMC predicts the future outlet temperature of the heat exchanger through rolling optimization and calculates the optimal heat exchange medium flow adjustment command based on the prediction results. Compared with traditional PID control, this method can effectively consider future temperature change trends, reduce temperature fluctuations, and improve the stability of the heat exchanger outlet temperature. The establishment of the state-space model provides dynamic input for DMC, enabling the control system to adapt to complex operating conditions, thereby improving the energy efficiency and responsiveness of the heat exchanger.

[0128] Step S4, using the heat exchange medium flow adjustment instruction of step S3 to control the multi-channel heat exchange tube bundle structure inside the heat exchanger, and dynamically distribute the flue gas flow entering different heat exchange areas;

[0129] In step S4, the heat exchange area is divided into a high temperature area, a medium temperature area, and a low temperature area;

[0130] In step S4, according to the heat exchange medium flow adjustment instruction, the adjustable baffle or rotary valve is used to adjust the flue gas flow ratio entering each zone, and the heat exchanger outlet temperature adjustment feedback data is generated;

[0131] The steps of using the heat exchange medium flow adjustment instruction of step S3 to control the multi-channel heat exchange tube bundle structure inside the heat exchanger and dynamically distribute the flue gas flow entering different heat exchange areas are as follows:

[0132] The internal area of the heat exchanger is divided into:

[0133] High temperature zone, used to initially reduce the flue gas temperature,

[0134] Medium temperature zone, used for further heat exchange,

[0135] and a low-temperature zone for final smoke exhaust;

[0136] The flue gas flow ratio is adjusted by adjusting the damper or rotary valve. The adjustment process is expressed as:

[0137] Q g,i =α i Q g ,

[0138] Among them, Q g,i represents the flue gas flow rate entering the i-th heat exchange area, Q g represents the total flue gas flow, α i is the distribution coefficient, satisfying ∑ i α i =1;

[0139] Specifically, based on the heat exchange medium flow regulation instruction calculated in step S3, the distribution of the flue gas flow inside the heat exchange tube bundle structure is dynamically controlled to divide the heat exchange tube bundle into high temperature, medium temperature and low temperature zones. The flue gas flow entering different zones is dynamically adjusted using adjustable baffles or rotary valves to optimize the heat exchange efficiency.

[0140] Step S5, based on the temperature adjustment feedback data at the heat exchanger outlet, adaptively optimize the DMC control parameters of step S3 using a reinforcement learning algorithm;

[0141] In step S5, the deviation between the predicted temperature and the actual outlet temperature is compared to determine the error range of the DMC control strategy;

[0142] In step S5, the DMC control weight is updated based on the error range, and the zone heat exchange adjustment parameters are adjusted;

[0143] Based on the temperature adjustment feedback data of the heat exchanger outlet, the steps of adaptively optimizing the DMC control parameters of step S3 based on the reinforcement learning algorithm are as follows:

[0144] The prediction error of the heat exchanger outlet temperature is calculated at each time step t, and the calculation formula is:

[0145]

[0146] Among them, e t represents the temperature prediction error, represents the predicted temperature of the heat transfer medium at time step t, T m,t represents the actual measured temperature of the heat transfer medium at time step t,

[0147] Set the reinforcement learning goal, construct the objective function J, and measure the optimization degree of the control strategy:

[0148]

[0149] Where J represents the control optimization objective function, T is the number of time steps in the optimization cycle, λ is the regularization coefficient, and Δw τ represents the change in the weight of the control parameter,

[0150] The gradient descent method is used to update the DMC control weight, and the update formula is:

[0151]

[0152] in, represents the weight of the DMC control parameter after the update at time step t+1, represents the current DMC control parameter weight at time step t, η is the learning rate, Denotes the loss function J with respect to w τ The gradient,

[0153] Based on the optimized DMC control weight, the flue gas flow distribution in the heat exchange area is adjusted. The adjustment formula is:

[0154]

[0155] in, represents the flue gas flow distribution coefficient of the ith heat exchange area after the update at time step t+1, represents the flue gas flow distribution coefficient of the i-th heat exchange area at time step t, and γ is the adjustment step length;

[0156] Specifically, the reinforcement learning algorithm is used to optimize the control parameters of the dynamic matrix control, and the DMC control strategy is adaptively adjusted based on the feedback data of the heat exchanger outlet temperature. By calculating the temperature prediction error and using the gradient descent method to update the DMC control weight, the control strategy can be optimized and adjusted according to the historical error, and the control parameters can be adaptively adjusted to improve the adaptability of the system.

[0157] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A control method for a flue gas heat exchanger in a thermal power plant based on dynamic matrix control, characterized in that: The method includes: Step S1: Install a temperature sensor, a flow meter, and a pressure sensor at the inlet and outlet of the flue gas heat exchanger, respectively, to collect flue gas sensor data and boiler operation data in real time and generate a current operating condition data set; Step S2: Based on the current working condition data set of step S1, a flue gas temperature prediction model is constructed using a long short-term memory network (LSTM) to predict the flue gas temperature change trend in the future time period and generate a temperature change adjustment instruction; Step S3, receiving the temperature change adjustment instruction generated in step S2, using dynamic matrix control (DMC) to predictively control the outlet temperature of the heat exchanger, and generating a heat exchange medium flow adjustment instruction; Step S4, using the heat exchange medium flow adjustment instruction of step S3 to control the multi-channel heat exchange tube bundle structure inside the heat exchanger, and dynamically distribute the flue gas flow entering different heat exchange areas; Step S5, based on the temperature adjustment feedback data at the heat exchanger outlet, adaptively optimize the DMC control parameters of step S3 using a reinforcement learning algorithm; The steps of constructing a flue gas temperature prediction model using a long short-term memory network (LSTM) are as follows: Construct a flue gas temperature prediction model and collect the current operating condition data set D t , used for LSTM network training, the dataset is represented as D t : D t ={T g ,Q g ,P g ,T m ,L,C,Q a }, Among them, T g Indicates the flue gas temperature, Q g Indicates the flue gas flow rate, P g Indicates the flue gas pressure, T m represents the temperature of the heat exchange medium, L represents the boiler load, C represents the coal characteristics, Q a Indicates air flow; The state update process of LSTM cells includes: i t =σ(W i [h t-1 ,x t ]+b i ), f t =σ(W f [h t-1 ,x t ]+b f ), the t =σ(W o [h t-1 ,x t ]+b o ), h t =o t ⊙tanh(C t ), Among them, i t represents the input gate activation value, f t Represents the activation value of the forget gate, o t represents the output gate activation value, represents the candidate state value, C t Indicates the unit state value, C t-1 Indicates the unit state value of the previous time step, h t represents the hidden state value, h t-1 Represents the hidden state value of the previous time step, W i ,W f ,W o ,W c is the weight matrix, b i ,b f ,b o ,b c is the bias term, x t Represents input data, σ represents Sigmoid activation function, tanh represents hyperbolic tangent activation function, and ⊙ represents element-by-element multiplication operation; The steps of adaptively optimizing the DMC control parameters of step S3 based on the temperature adjustment feedback data of the heat exchanger outlet and the reinforcement learning algorithm are as follows: The prediction error of the heat exchanger outlet temperature is calculated at each time step t, and the calculation formula is: Among them, e t represents the temperature prediction error, represents the predicted temperature of the heat transfer medium at time step t, T m,t represents the actual measured temperature of the heat transfer medium at time step t, Set the reinforcement learning goal, construct the objective function J, and measure the optimization degree of the control strategy: Where J represents the control optimization objective function, T is the number of time steps in the optimization cycle, λ is the regularization coefficient, and Δw τ represents the change in the weight of the control parameter, The gradient descent method is used to update the DMC control weight, and the update formula is: in, represents the weight of the DMC control parameter after the update at time step t+1, represents the current DMC control parameter weight at time step t, η is the learning rate, Denotes the loss function J with respect to w τ The gradient, Based on the optimized DMC control weight, the flue gas flow distribution in the heat exchange area is adjusted. The adjustment formula is: in, represents the flue gas flow distribution coefficient of the ith heat exchange area after the update at time step t+1, It represents the flue gas flow distribution coefficient of the i-th heat exchange area at time step t, and γ is the adjustment step size.

2. The method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to claim 1, characterized in that: The steps of predicting the flue gas temperature change trend in the future time period and generating the temperature change adjustment instruction are as follows: The hidden state h calculated based on LSTM t , predict the flue gas temperature in the future time step, the prediction formula is: in, represents the predicted flue gas temperature at time t+τ, f(·) is the LSTM prediction model, and θ is the parameter set obtained by LSTM training. Generate temperature change adjustment instructions, the generation process includes: Temperature change trend, calculated as: Where, ΔT g is the flue gas temperature change, To predict the flue gas temperature, T g,t is the flue gas temperature at the current time step t, If |ΔT g |>δ T , then generate temperature change adjustment instructions: U T =g(ΔT g ), Among them, δ T is the temperature change threshold, U T represents the temperature change adjustment instruction, and g(·) is the adjustment function.

3. The method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to claim 2, characterized in that: In step S3, a state space model of the heat exchanger is established, and the historical temperature change data, current measurement data and future predicted temperature trends are integrated to calculate the heat exchange medium flow adjustment plan and generate a heat exchange medium flow adjustment instruction.

4. The method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to claim 3, characterized in that: The steps of using the dynamic matrix control (DMC) to predictively control the outlet temperature of the heat exchanger and generate a heat exchange medium flow adjustment instruction are as follows: The heat exchanger state space model is established, and the heat exchanger state space equation is defined as: X t+1 =AX t +BU t , Y t =CX t +YOU t , Among them, X t+1 represents the state variable vector for the next time step, X t Represents the state variable vector of the current time step, U t Represents the input variable vector of the current time step, Y t represents the output variable vector of the current time step, A represents the heat exchanger state transfer matrix, which describes the evolution of the heat exchanger state over time, B represents the input influence matrix, which describes the influence of the heat exchange medium flow on the heat exchanger state, C represents the measurement matrix, which maps the state variables to the output variables, and D represents the input direct influence matrix, which describes the direct influence of the input on the output. Temperature changes are predicted based on historical data and future inputs. The prediction formula is: in, is the heat transfer medium temperature predicted at the future time step t+τ, X t+τ is the state variable at time step t+τ, U t+τ is the input variable at time step t+τ, DMC uses a rolling optimization strategy to calculate the optimal heat exchange medium flow regulation instruction. The calculation formula is: in, is the optimal heat exchange medium flow regulation instruction, is the expected heat transfer medium temperature, N is the prediction step length, w τ is the weight coefficient of the time step τ, λ is the regularization coefficient, ΔU m The input change of heat exchange medium flow control, U m Represents the control input for the heat transfer medium flow rate.

5. The method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to claim 4, characterized in that: In step S4, the heat exchange area is divided into a high temperature area, a medium temperature area, and a low temperature area; In step S4, according to the heat exchange medium flow adjustment instruction, the adjustable baffle or rotary valve is used to adjust the flue gas flow ratio entering each zone, and heat exchanger outlet temperature adjustment feedback data is generated.

6. The method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to claim 5, characterized in that: The steps of using the heat exchange medium flow rate adjustment instruction of step S3 to control the multi-channel heat exchange tube bundle structure inside the heat exchanger and dynamically distribute the flue gas flow entering different heat exchange areas are as follows: The internal area of the heat exchanger is divided into: High temperature zone, used to initially reduce the flue gas temperature, Medium temperature zone, used for further heat exchange, and a low-temperature zone for final smoke exhaust; The flue gas flow ratio is adjusted by adjusting the damper or rotary valve. The adjustment process is expressed as: Q g,i =α i Q g , Among them, Q g,i represents the flue gas flow rate entering the i-th heat exchange area, Q g represents the total flue gas flow, α i is the distribution coefficient, satisfying ∑ i α i =1.

7. The method for controlling a flue gas heat exchanger in a thermal power plant based on dynamic matrix control according to claim 6, characterized in that: In step S5, the deviation between the predicted temperature and the actual outlet temperature is compared to determine the error range of the DMC control strategy; In step S5, the DMC control weight is updated based on the error range, and the zone heat exchange adjustment parameters are adjusted.

Images