Intelligent dosing method and system for boiler feed water based on RBF-PID

Abstract

The invention discloses an RBF-PID (Radial Basis Function-Proportion Integration Differentiation)-based intelligent dosing method and system for boiler feed water. The method comprises the following steps: collecting information; an RBF neural network model is established, the RBF neural network model is optimized based on a gradient descent algorithm to set PID controller parameters online, an RBF-PID control signal is output, a feedforward model is established, collected information is substituted into the feedforward model, and a feedforward control signal is output; the RBF-PID control signal and the feedforward control signal are superposed to obtain a master control signal; the dosing control unit receives a master control signal to adjust the frequency of an ammonia adding pump to complete dosing; according to the invention, the online learning capability of the RBF neural network is fully utilized to self-tune the PID parameters, the PID control response speed is improved, and in combination with the feedforward model based on feed water flow change, the dosing requirement of rapid load change can be met.

Description

Technical Field

[0001] The invention relates to the technical field of automatic chemical dosing in thermal power plants, and in particular to an intelligent dosing method and system for boiler feed water based on RBF-PID. Background Art

[0002] In order to prevent acid corrosion in the furnace system and reduce the corrosion rate of the four boiler tubes, the boiler feed water must be treated with ammonia to increase the pH value. The traditional way of manually adjusting the dosage is that the chemical operator monitors the online instrument data of the water vapor analysis system in real time, judges and adjusts the status of the dosing pump according to the feed water conductivity and pH value supervision standards, and manually starts and stops the dosing pump through the computer. The dosage and the operation time of the dosing pump are basically determined by the operator's operation and experience, and it is often easy to have high or low water quality indicators, which are difficult to control. At present, the new feed water ammonia addition system mostly uses PID control to achieve automatic dosing, but the pH adjustment process of feed water ammonia addition has large time lag and nonlinear characteristics. When the unit load fluctuates and the ammonia concentration is inaccurate, the PID control performance is poor, and the controller overshoots often cause water quality to exceed the standard. As the deep peak-shaving operation of thermal power units becomes the norm, it is necessary to improve the traditional PID control method and establish a new feed water intelligent ammonia addition system, which can reduce the operating burden of personnel on the one hand and improve the rapid response of the dosing system on the other hand. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a method and system for intelligent dosing of boiler feed water based on RBF-PID, which aims to solve the problems in the background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a boiler feed water intelligent dosing method based on RBF-PID, comprising the following steps:

[0005] Step S1: Collect information, including water supply pH, hydrogen conductivity CC, specific conductivity SC, water supply flow Q, ammonia agent conductivity, ammonia agent concentration C a , water supply ammonia pump frequency f;

[0006] Step S2: Establish an RBF neural network model, optimize the RBF neural network model based on the gradient descent algorithm to adjust the PID controller parameters online, and output the RBF-PID control signal based on the water flow rate Q and the ammonia concentration C. a A feedforward model is established based on the mathematical relationship between the frequency f of the water supply and ammonia pump, the collected information is substituted into the feedforward model, and a feedforward control signal is output; the RBF-PID control signal and the feedforward control signal are superimposed to obtain a total control signal;

[0007] Step S3: The dosing control unit receives the total control signal to adjust the frequency of the ammonia dosing pump to complete the dosing.

[0008] Furthermore, according to the characteristics of large hysteresis and nonlinearity of regulating pH by adding ammonia to water, the transfer function of the dosing control unit is approximately represented by a first-order inertia link plus a pure delay link:

[0009]

[0010] In the formula, K is the gain coefficient; T is the time constant; t is the delay time; G (s) represents the transfer function of the dosing control unit; s represents the Laplace operator; e represents the base of the natural logarithm.

[0011] Furthermore, the specific process of optimizing the parameters of the PID controller online based on the gradient descent algorithm for the RBF neural network model and outputting the RBF-PID control signal is as follows:

[0012] Determine the parameters of the RBF neural network model;

[0013] Update the parameters of the RBF neural network model using the gradient descent method;

[0014] The PID controller adopts an incremental PID controller, and calculates the parameters of the incremental PID controller through the updated RBF neural network model;

[0015] Optimize the parameters of the PID controller using the gradient descent method;

[0016] Substitute the optimized parameters of the PID controller into the PID controller to obtain the RBF-PID control signal.

[0017] Furthermore, the specific process of updating the parameters of the RBF neural network model using the gradient descent method is as follows:

[0018] Using the gradient descent algorithm, calculate the update formulas for the weight vector W, the internal center point C of the hidden layer neurons, and the base width vector B of the RBF neural network model respectively:

[0019] Update formula for the weight vector W:

[0020] ω j (k) = ω j (k - 1) + η[y(k) - y m (k)]h j + α[ω j (k - 1) - ω j (k - 2)];

[0021] In the formula, η represents the learning rate; α represents the momentum factor; ω j (k) represents the weight coefficient of the jth hidden layer neuron at the kth moment; ω j(k - 1) and ω j (k - 2) represent the weight coefficients of the j-th hidden layer neuron at the (k - 1)-th and (k - 2)-th moments respectively;

[0022] Base width vector B update formula:

[0023]

[0024] b j (k) = b j (k - 1) + Δb j (k) + α[b j (k - 1) - b j (k - 2)];

[0025] In the formula, b j (k) represents the base width parameter of the j-th hidden layer neuron at the k-th moment;

[0026] Center point C update formula:

[0027]

[0028] c ji (k) = c ji (k - 1) + Δc ji (k) + α[c ji (k - 1) - c ji (k - 2)];

[0029] In the formula, c ji (k) represents the center point corresponding to the i-th input in the j-th hidden layer neuron at the k-th moment;

[0030] The Jacobian information of the RBF neural network model is:

[0031]

[0032] In the formula, u(k) is the output of the PID controller at the k-th moment, and x1 = Δu(k).

[0033] Furthermore, an incremental PID controller is adopted, and the control error e(k) = y d (k) - y(k), where y d (k) is the preset target ammonia pump frequency at the k-th moment; the input of the PID controller is:

[0034]

[0035] In the formula, x c (1), x c (2), x c(3) represent the first input parameter, the second input parameter, and the third input parameter of the PID controller respectively; e(k - 1) and e(k - 2) represent the frequency deviation of the ammonia addition pump at times k - 1 and k - 2 respectively;

[0036] The output of the PID controller is:

[0037]

[0038] In the formula, K p , K i , K d represent the proportional coefficient, integral coefficient, and differential coefficient of the PID controller respectively;

[0039] The tuning index of the RBF neural network model is:

[0040] Furthermore, the specific process of optimizing the parameters of the PID controller by the gradient descent method is:

[0041] Adjust K p , K i and K d at time k using the gradient descent method, which is expressed as:

[0042]

[0043] In the formula, η p , η i and η d are the learning rates of proportional, integral, and differential respectively;

[0044] The parameters of the PID controller at time k + 1 are:

[0045]

[0046] Furthermore, a feedforward model is established based on the mathematical relationship between the feed water flow Q, ammonia water reagent concentration C a and the frequency f of the feed water ammonia addition pump, which is expressed as:

[0047]

[0048] In the formula, k1 and k2 are the parameters to be identified in the feedforward model, which are obtained by fitting according to the on-site operation historical data; A is the ammonia ion concentration in the feed water;

[0049] The relationship between A and the feed water specific conductivity SC and the feed water pH is:

[0050] A = (13.2 * SC 2 + 62.7 * SC);

[0051] pH = 8.57 + log SC.

[0052] Further, the specific process of establishing the RBF neural network model is as follows:

[0053] Establish the RBF neural network model; determine the number of neurons in the input layer, the number of neurons in the hidden layer, and the number of neurons in the output layer of the RBF neural network model;

[0054] The input vector of the RBF neural network model is x = [x1, x2,..., x n , where x n is the nth element in the input vector of the RBF neural network model; the radial basis vector of the RBF neural network model is h = [h1, h2,..., h j ,..., h m , j = 1, 2,..., m, and h j is the jth element in the radial basis vector;

[0055] Among them, h j is expressed as:

[0056]

[0057] In the formula, c j is the center vector inside the jth hidden layer neuron of the RBF neural network model; C j = [c j1 , c j2 ,..., c ji ,..., c jn T represents the center point inside the jth hidden layer neuron, c ji represents the center point of the jth hidden layer neuron corresponding to the ith input, i = 1, 2,..., n, b j is the base width parameter of the jth hidden layer neuron of the RBF neural network model, and T represents the transpose;

[0058] The base width vector of the RBF neural network model is B = [b1, b2,..., b j ,..., b m T , j = 1, 2,..., m;

[0059] The weight vector of the RBF neural network model is W = [ω1, ω2,..., ω j ,..., ω m T , j = 1, 2,..., m, and ω m represents the weight coefficient of the mth hidden layer neuron;

[0060] ​​​The output of the RBF neural network model at time k is: y m (k) = ω1h1 + ω2h2 +,..., + ω m h m ;

[0061] The performance index function of the RBF neural network model is: y(k) is the output value of the dosing control unit at time k, that is, the frequency of the ammonia addition pump.

[0062] An intelligent dosing system for boiler feed water based on RBF-PID includes:

[0063] A collection module for collecting information, including feed water pH, hydrogen conductivity CC, specific conductivity SC, feed water flow Q, conductivity of ammonia water reagent, concentration C of ammonia water reagent a and the frequency f of the feed water ammonia addition pump;

[0064] A signal tuning module for establishing an RBF neural network model, optimizing the parameters of the PID controller online based on the gradient descent algorithm, outputting an RBF-PID control signal, establishing a feedforward model based on the mathematical relationship between the feed water flow Q, the concentration C of ammonia water reagent a and the frequency f of the feed water ammonia addition pump, substituting the collected information into the feedforward model, and outputting a feedforward control signal; superimposing the RBF-PID control signal and the feedforward control signal to obtain a total control signal;

[0065] A dosing module for receiving the total control signal through the dosing control unit to adjust the frequency of the ammonia addition pump to complete dosing.

[0066] Compared with the existing technology, the present invention has the following beneficial effects: The present invention makes full use of the online learning ability of the RBF neural network to self-tune the PID parameters, improves the response speed of the PID control, and combines the feedforward model based on the change of feed water flow, which can meet the dosing requirements for rapid load changes; the present invention can automatically control the ammonia concentration according to the target conductivity value of ammonia water, reducing the burden of manual dosing; by integrating the dosing and dosing control units, the chemical dosing of the unit's water vapor system is centralized and systematic. Brief Description of the Drawings

[0067] Figure 1 It is a schematic diagram of the intelligent dosing system for boiler feed water of the present invention. Detailed Embodiment

[0068] As Figure 1 shown, the present invention provides a technical solution: An intelligent dosing method for boiler feed water based on RBF-PID includes the following steps:

[0069] Step S1: Collect information, including feed water pH, hydrogen conductivity CC, specific conductivity SC, feed water flow rate Q, conductivity of ammonia water reagent, and concentration C of ammonia water reagent a , and the frequency f of the feed water ammonia addition pump.

[0070] Step S2: Optimize the parameters of the PID controller online based on the gradient descent algorithm for the RBF neural network model, output the RBF-PID control signal, establish a feedforward model based on the mathematical relationship among the feed water flow rate Q, the concentration C of the ammonia water reagent a and the frequency f of the feed water ammonia addition pump, substitute the collected information into the feedforward model, and output the feedforward control signal; superimpose the RBF-PID control signal and the feedforward control signal to obtain the total control signal.

[0071] Step S3: Receive the total control signal through the dosing control unit to adjust the frequency of the ammonia addition pump to complete dosing.

[0072] Among them, the dosing control unit includes a 1m 3 medicine preparation tank that can automatically adjust the concentration of ammonia water. The concentrated reagent for dosing is concentrated ammonia water or liquid ammonia. The medicine preparation tank adjusts the makeup water volume by controlling the specific conductivity of ammonia water at 1150 - 1250 μs / cm.

[0073] According to the characteristics of large hysteresis and nonlinearity of adjusting pH by adding ammonia to feed water, the transfer function of the dosing control unit is approximately represented by a first-order inertia link plus a pure delay link:

[0074]

[0075] In the formula, K is the gain coefficient; T is the time constant; t is the delay time; G (s) represents the transfer function of the dosing control unit; s represents the Laplace operator; e represents the base of the natural logarithm.

[0076] The specific process of online tuning the parameters of the PID controller based on the RBF neural network model is as follows:

[0077] Establish an RBF neural network model; determine that the number of neurons in the input layer of the RBF neural network model is n, the number of neurons in the hidden layer of the RBF neural network model is m, and the number of neurons in the output layer of the RBF neural network model is 1.

[0078] Among them, the input vector of the RBF neural network model is x = [x1, x2,..., x m , x n is the nth element in the input vector of the RBF neural network model; the radial basis vector of the RBF neural network model is h = [h1, h2,..., h j ,..., h m , j = 1, 2,..., m, hj is the j-th element in the radial basis vector.

[0079] Among them, h j is expressed as:

[0080]

[0081] In the formula, c j is the center vector inside the j-th hidden layer neuron of the RBF neural network model;

[0082] C j = [c j1 , c j2 ,..., c ji , …, c jn T represents the center point inside the j-th hidden layer neuron, c ji represents the center point of the j-th hidden layer neuron corresponding to the i-th input, i = 1, 2, …, n, b j is the basis width parameter of the j-th hidden layer neuron of the RBF neural network model, and T represents the transpose.

[0083] The basis width vector of the RBF neural network model is B = [b1, b2,..., b j ,..., b m T , j = 1, 2,..., m.

[0084] The weight vector of the RBF neural network model is W = [ω1, ω2,..., ω j ,..., ω m T , j = 1, 2,..., m, ω m represents the weight coefficient of the m-th hidden layer neuron.

[0085] The output of the RBF neural network model at time k is: y m (k) = ω1h1 + ω2h2 +,..., + ω m h m .

[0086] The performance index function of the RBF neural network model is: y(k) is the output value of the dosing control unit at time k, that is, the frequency of the ammonia addition pump.

[0087] Using the gradient descent algorithm, calculate the update formulas of the weight vector W, the center point C inside the hidden layer neuron, and the basis width vector B of the RBF neural network model respectively:

[0088] Update formula of the weight vector W:

[0089] ωj ω(k) = ω j (k - 1) + η[y(k) - y m (k)]h j + α[ω j (k - 1) - ω j (k - 2)];

[0090] where η represents the learning rate; α represents the momentum factor; ω j (k) represents the weight coefficient of the j-th hidden layer neuron at time k; ω j (k - 1) and ω j (k - 2) represent the weight coefficients of the j-th hidden layer neuron at times k - 1 and k - 2 respectively.

[0091] Base width vector B update equation:

[0092]

[0093] b j (k) = b j (k - 1) + Δb j (k) + α[b j (k - 1) - b j (k - 2)];

[0094] where b j (k) represents the base width parameter of the j-th hidden layer neuron at time k.

[0095] Center point C update equation:

[0096]

[0097] c ji (k) = c ji (k - 1) + Δc ji (k) + α[c ji (k - 1) - c ji (k - 2)];

[0098] where c ji (k) represents the center point corresponding to the i-th input in the j-th hidden layer neuron at time k.

[0099] The Jacobian information of the RBF neural network model (i.e., the sensitivity information of the object's output to changes in the control input) is:

[0100]

[0101] where u(k) is the output of the PID controller at time k, and x1 = Δu(k).

[0102] The Jacobian information reflects the variation characteristics of the output of the RBF neural network with respect to the input. By identifying the Jacobian information, real-time estimation and dynamic adjustment of the system state can be achieved.

[0103] An incremental PID controller is adopted, and the control error e(k) = y d (k) - y(k), where y d (k) is the preset target ammonia addition pump frequency at time k; the inputs to the PID controller are:

[0104]

[0105] In the formula, x c (1), x c (2), x c (3) represent the first input parameter, the second input parameter, and the third input parameter of the PID controller respectively; e(k - 1) and e(k - 2) represent the ammonia addition pump frequency deviations at times k - 1 and k - 2 respectively.

[0106] The output of the PID controller is:

[0107]

[0108] In the formula, K p , K i , K d represent the proportional coefficient, integral coefficient, and differential coefficient of the PID controller respectively.

[0109] The tuning index of the RBF neural network model is:

[0110] The gradient descent method is used to adjust K at time k p , K i and K d , which is expressed as:

[0111]

[0112] In the formula, η p , η i and η d are the learning rates of proportion, integral, and differential respectively.

[0113] The PID controller parameters at time k + 1 are:

[0114]

[0115] Substitute the PID controller parameters at time k + 1 into the PID controller to obtain the RBF - PID control signal.

[0116] Among them, a feedforward model is established based on the mathematical relationship of the feed water flow rate Q, the ammonia water agent concentration C a and the feed water ammonia addition pump frequency f. The specific process of substituting the collected information into the feedforward model and outputting the feedforward control signal is as follows:

[0117] The feedforward model is expressed as:

[0118]

[0119] In the formula, k1 and k2 are the parameters to be identified in the feedforward model, which are obtained by fitting according to the on-site operation historical data; A is the ammonia ion concentration in the feed water.

[0120] The relationship between A and the feed water specific conductivity SC and the feed water pH is:

[0121] A = (13.2 * SC 2 + 62.7 * SC);

[0122] pH = 8.57 + logSC;

[0123] Substitute the collected information into the feedforward model to obtain the feedforward control signal.

[0124] An intelligent dosing system for boiler feed water based on RBF-PID includes:

[0125] A collection module for collecting information, including feed water pH, hydrogen conductivity CC, specific conductivity SC, feed water flow rate Q, ammonia water agent conductivity, ammonia water agent concentration C a and the feed water ammonia addition pump frequency f.

[0126] A signal tuning module for establishing an RBF neural network model, optimizing the parameters of the RBF neural network model online based on the gradient descent algorithm to tune the PID controller, outputting an RBF-PID control signal, establishing a feedforward model based on the mathematical relationship of the feed water flow rate Q, the ammonia water agent concentration C a and the feed water ammonia addition pump frequency f, substituting the collected information into the feedforward model, and outputting a feedforward control signal; superimposing the RBF-PID control signal and the feedforward control signal to obtain a total control signal.

[0127] A dosing module for receiving the total control signal through a dosing control unit to adjust the ammonia addition pump frequency to complete dosing.

[0128] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent dosing of boiler feed water based on RBF-PID, characterized in that: The steps include: Step S1: Collect information, including water supply pH, hydrogen conductivity CC, specific conductivity SC, water supply flow Q, ammonia agent conductivity, ammonia agent concentration C a , water supply ammonia pump frequency f; Step S2: Establish an RBF neural network model, optimize the RBF neural network model based on the gradient descent algorithm to adjust the PID controller parameters online, and output the RBF-PID control signal based on the water flow rate Q and the ammonia concentration C. a A feedforward model is established based on the mathematical relationship between the frequency f of the water supply and ammonia pump, the collected information is substituted into the feedforward model, and a feedforward control signal is output; the RBF-PID control signal and the feedforward control signal are superimposed to obtain a total control signal; Step S3: The dosing control unit receives the total control signal to adjust the frequency of the ammonia dosing pump to complete the dosing.

2. The intelligent dosing method for boiler feed water based on RBF-PID according to claim 1 is characterized in that: According to the large hysteresis and nonlinear characteristics of pH adjustment by adding ammonia to feed water, the transfer function of the dosing control unit is approximately expressed by a first-order inertia link plus a pure delay link: In the formula, K is the gain coefficient; T is the time constant; t is the delay time; G (s) represents the transfer function of the dosing control unit; s represents the Laplace operator; and e represents the base of the natural logarithm.

3. The intelligent dosing method for boiler feed water based on RBF-PID according to claim 2 is characterized in that: The specific process of optimizing the RBF neural network model online to tune the PID controller parameters based on the gradient descent algorithm and outputting the RBF-PID control signal is as follows: Determine the parameters of the RBF neural network model; The gradient descent method is used to update the parameters of the RBF neural network model; The PID controller adopts an incremental PID controller, and the parameters of the incremental PID controller are calculated by the RBF neural network model after updating the parameters; Optimize the parameters of the PID controller by gradient descent method; Substitute the optimized PID controller parameters into the PID controller to obtain the RBF-PID control signal.

4. The intelligent dosing method for boiler feed water based on RBF-PID according to claim 3 is characterized by: The specific process of updating the parameters of the RBF neural network model using the gradient descent method is as follows: The gradient descent algorithm is used to calculate the update formulas of the weight vector W, the internal center point C of the hidden layer neuron and the base width vector B of the RBF neural network model: The weight vector W is updated as follows: oh j (k)=ω j (k-1)+η[y(k)-y m (k)]h j +a[ω j (k-1)-ω j (k-2)]; In the formula, η represents the learning rate; α represents the momentum factor; ω j (k) represents the weight coefficient of the jth hidden layer neuron at time k; ω j (k-1) and ω j (k-2) represents the weight coefficient of the jth hidden layer neuron at time k-1 and k-2 respectively; Update formula of base width vector B: b j (k)=b j (k-1)+Δb j (k)+α[b j (k-1)-b j (k-2)]; Where b j (k) represents the base width parameter of the jth hidden layer neuron at time k; Update formula of center point C: c ji (k)=c ji (k-1)+Δc ji (k)+α[c ji (k-1)-c ji (k-2)]; In the formula, c ji (k) represents the center point of the jth hidden layer neuron corresponding to the i-th input at time k; The Jacobian information of the RBF neural network model is: Where u(k) is the output of the PID controller at time k, and x1 = Δu(k).

5. The intelligent dosing method for boiler feed water based on RBF-PID according to claim 4 is characterized in that: Using incremental PID controller, control error e(k) = y d (k)-y(k),y d (k) is the target ammonia pump frequency preset at time k; the input of the PID controller is: In the formula, x c (1), x c (2) x c (3) respectively represent the first input parameter, the second input parameter and the third input parameter of the PID controller; e(k-1) and e(k-2) respectively represent the frequency deviation of the ammonia pump at time k-1 and time k-2; The output of the PID controller is: In the formula, K p , K i , K d They represent the proportional coefficient, integral coefficient, and differential coefficient of the PID controller respectively; The tuning index of the RBF neural network model is:

6. The intelligent dosing method for boiler feed water based on RBF-PID according to claim 5 is characterized by: The specific process of optimizing the parameters of the PID controller by the gradient descent method is: Use gradient descent method to adjust K at time k p , K i and K d , expressed as: Where η p , η i and η d are the learning rates for proportion, integration and differentiation respectively; The PID controller parameters at time k+1 are:

7. The intelligent dosing method for boiler feed water based on RBF-PID according to claim 6 is characterized by: Based on the water flow rate Q, ammonia concentration C a The mathematical relationship between the frequency f of the feed water ammonia pump and the feed forward model is established, which is expressed as: Where, k1 and k2 are the parameters to be identified in the feedforward model, which are obtained by fitting based on the historical data of field operation; A is the ammonia ion concentration in the feed water; The relationship between A and the feed water specific conductivity SC and feed water pH is: A=(13.2*SC 2 +62.7*SC); pH = 8.57 + log SC.

8. The intelligent dosing method for boiler feed water based on RBF-PID according to claim 7 is characterized by: The specific process of establishing the RBF neural network model is: Establishing an RBF neural network model; determining the number of neurons in the input layer of the RBF neural network model, the number of neurons in the hidden layer of the RBF neural network model, and the number of neurons in the output layer of the RBF neural network model; The input vector of the RBF neural network model is x = [x1, x2, ..., x n ], x n is the nth element in the input vector of the RBF neural network model; the radial basis vector of the RBF neural network model is h = [h1, h2, ..., h j , ..., h m ], j = 1, 2, ..., m, h j is the jth element in the radial basis vector; Among them, h j It is expressed as: In the formula, c j is the center vector inside the jth hidden layer neuron of the RBF neural network model; C j =[c j1 , c j2 , ..., c ji , ..., c jn ] T represents the internal center point of the jth hidden layer neuron, c ji Indicates that the jth hidden layer neuron corresponds to the center point of the ith input, i = 1, 2, ..., n, b j is the base width parameter of the jth hidden layer neuron in the RBF neural network model, and T represents transposition; The base width vector of the RBF neural network model is B = [b1, b2, ..., b j ,...,b m ] T , j = 1, 2, ..., m; The weight vector of the RBF neural network model is W = [ω1, ω2, ..., ω j ,...,ω m ] T , j = 1, 2, ..., m, ω m Represents the weight coefficient of the mth hidden layer neuron; The output of the RBF neural network model at time k is: y m (k)=ω1h1+ω2h2+,...,+ω m h m ; The performance index function of the RBF neural network model is: y(k) is the output value of the dosing control unit at time k, that is, the frequency of the ammonia dosing pump.

9. An intelligent dosing system for boiler feed water based on RBF-PID, characterized in that: include: The acquisition module is used to collect information, including water supply pH, hydrogen conductivity CC, specific conductivity SC, water supply flow Q, ammonia agent conductivity, ammonia agent concentration C a , water supply ammonia pump frequency f; The signal tuning module is used to establish the RBF neural network model, optimize the RBF neural network model based on the gradient descent algorithm, tune the PID controller parameters online, and output the RBF-PID control signal based on the water flow rate Q and the ammonia concentration C. a A feedforward model is established based on the mathematical relationship between the frequency f of the water supply and ammonia pump, the collected information is substituted into the feedforward model, and a feedforward control signal is output; the RBF-PID control signal and the feedforward control signal are superimposed to obtain a total control signal; The dosing module is used to receive the total control signal through the dosing control unit to adjust the frequency of the ammonia dosing pump to complete the dosing.

Images