Improved generalized predictive control method for coagulation dosing process

Abstract

The invention provides an improved generalized predictive control method for a coagulation dosing process, which is characterized in that for the mechanism modeling problem of the coagulation dosing process, a model is divided into three cascaded models for construction according to a scene under a real condition, and the three cascaded models comprise a mixing model, a flocculation model and a precipitation model; in the flocculation stage, a group balance model widely applied to a dispersed phase system is utilized, compared with a traditional transfer function and fitting modeling method, the method is closer to the actual situation, and the accuracy and credibility of the model are improved; for an improved generalized predictive control algorithm problem, firstly, a least square method with a forgetting factor is improved into an iterative learning least square method, so that complete estimation of fast time-varying model parameters is realized; and secondly, a step factor is introduced and a recursive relationship is formed, so that the problem that matrix inversion operation cannot be performed in a PLC (Programmable Logic Controller) in an actual coagulation dosing process is solved, and meanwhile, control increment is constrained, so that the algorithm can be applied in an actual production process.

Description

Technical Field

[0001] The invention belongs to the technical field of tap water treatment, and in particular relates to an improved generalized predictive control method for a coagulation dosing process. Background Art

[0002] In the production process of tap water, coagulation and dosing is one of the key processes and is crucial to ensuring the quality of effluent water. This process promotes the aggregation of colloids and suspended particles in water to form larger flocs by adding coagulants and coagulants, which are then removed in the sedimentation tank. However, the coagulation and dosing process has complex dynamic characteristics such as multivariable, nonlinear, strong coupling and large time lag, which brings many challenges to process control.

[0003] In recent years, as the modern water treatment industry has increasingly stringent requirements for product quality, production stability and economy, the research on the automated control of the coagulation dosing process has attracted much attention. At present, in the actual coagulation dosing control, feedforward control is the most widely used method, and there are also many other methods such as fuzzy control, dynamic matrix control, and generalized predictive control. Although the feedforward control scheme is simple and easy to use, its control accuracy is low and its energy consumption is high; while fuzzy control requires rich field experience and lacks self-correction ability. However, these methods have not fully applied the model of the coagulation dosing process to the design of the controller, and therefore have failed to reflect the characteristics of the actual process well. Although dynamic matrix control applies the step response model to the controller design, due to the time-varying nature of the coagulation dosing process, its effect tends to gradually weaken after maintaining a good control effect for a period of time.

[0004] Chinese patent document CN104865979A discloses an adaptive generalized predictive control method and system for sewage treatment process. Experiments show that this method can stably and quickly control the dissolved oxygen concentration in sewage treatment, showing significant anti-interference ability, which helps to improve the stability and efficiency of the sewage treatment process. However, this method uses a fuzzy neural network model for parameter identification, but only on the time axis, resulting in the accuracy of the identification parameters being less than ideal; at the same time, the generalized predictive control algorithm used is highly complex and difficult to implement on the PLC platform, and has limitations in the control of the coagulation dosing process. Summary of the invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to provide an improved generalized predictive control method for coagulation dosing process in view of the deficiencies of the prior art. The present invention combines iterative learning least squares method with step-wise generalized predictive control algorithm, which can not only improve the accuracy of parameter identification, but also simplify the algorithm operation complexity and improve the stability, accuracy and rapidity of the control process. The method of the present invention comprises the following steps:

[0006] Step 1: construct a mechanism model of the coagulation and dosing process. According to the actual scenario, the mechanism model of the coagulation and dosing process is divided into three cascade models for construction. The three cascade models are a mixing model, a flocculation model, and a sedimentation model.

[0007] Step 2, design an improved generalized predictive control method for the coagulation dosing process, use the least squares method based on iterative learning to obtain the controlled autoregression and moving average model (CARIMA) online identification, introduce the step factor, and use the step-by-step generalized predictive control algorithm to perform rolling optimization to generate the optimal control quantity and avoid matrix inversion operations.

[0008] Step 1 includes:

[0009] Step 1-1, determine the input, output and interference of the coagulation dosing process mechanism model: the input is the dosage and the raw water turbidity, the output is the effluent turbidity, and the interference is the influent flow rate;

[0010] Step 1-2, establish a mixed model;

[0011] Steps 1-3, establishing a flocculation model;

[0012] Steps 1-4, establishing a precipitation model;

[0013] Steps 1-5: connect the three sub-models of mixing model, flocculation model and sedimentation model in series one by one to obtain the mechanism model of the dynamic behavior of the entire coagulation dosing system, and derive the effluent turbidity y that describes the entire coagulation dosing system. Turout expression.

[0014] Step 1-2 includes: the mixing process of the mixing model simulates the process of adding coagulant to raw water through a metering pump, the mixed liquid after addition enters the mixing tank for rapid stirring and mixing, and the mixing tank is regarded as a continuous flow complete mixing reactor to ensure that the reactants continue to enter evenly and are fully mixed in the tank, and the reaction products are continuously discharged from the reactor; the concentration and composition of the output material are consistent with the concentration and composition of the reactants in the reactor; the micelle concentration of the substance inside the mixing tank is set to be equal to the micelle concentration of the output; the lumped parameter model is used to describe the rapid mixing process, and the concentration change formula before and after the mixing tank is obtained:

[0015]

[0016]

[0017] Where V m is the volume of the mixing pool, x Dosin (t) and x Turin(t) are the coagulant concentration and raw water turbidity at time t; x Dosout (t) and x Turout (t) are the coagulant concentration and turbidity of the effluent from the mixing tank at time t, q represents the raw water flow rate; d is the integral symbol;

[0018] When the surface charge of the colloid is close to neutral, the colloid particles will be in a destabilized state and the coagulation process will start. At this time, the following formula is used to calculate the concentration of unstable colloids in the effluent of the mixing tank:

[0019]

[0020] Where N0 is the concentration of destabilized colloidal particles, x Dosout and x Turout are the coagulant concentration and turbidity of the effluent from the mixing tank, respectively, Dosbest is the optimal dosing concentration, m Tur and m Dos are the unit mass of impurity particles and the unit mass of coagulant particles respectively;

[0021] Set the number concentration N of flocs in the effluent of the mixing tank i Satisfies the following distribution:

[0022]

[0023] Where V0 represents the average volume of flocs, V i is the volume of flocs of size class i; exp is the natural exponential function.

[0024] Steps 1-3 include: the flocculation model regards the flocculation process in the flocculation tank as two parallel phenomena of aggregation and fragmentation, and derives the following description of the change rate of floc number concentration r of size class i: i The expression is:

[0025]

[0026] The first two items on the right represent the number of flocs of size class i generated by the collision of flocs of size class smaller than i; the third and fourth items represent the number of flocs of size class i reduced by agglomeration with other flocs; the fifth item represents the number of flocs of size class i reduced by breakage; the last item represents the number of flocs of size class i generated by the breakage of flocs of size class greater than i; where N i and N j are the number concentration of flocs of size class i and size class j, respectively, α i,j is the collision efficiency of flocs of size classes i and j, β i,j is the collision frequency of flocs of size classes i and j, S iis the breakage rate of flocs of size class i, Γ i,j is the fragmentation distribution function of flocs of size classes i and j.

[0027] Steps 1-4 include: the sedimentation model regards the sedimentation process in the sedimentation tank as free sedimentation, that is, any parameters of the flocs do not change after entering the sedimentation tank, and once they sink to the bottom, they will not return to the upper clear water. The suspended particle removal rate η is calculated using the following formula:

[0028]

[0029]

[0030] Where u is the sedimentation velocity, ρ and ρ0 are the density of impurity particles and water respectively, d is the particle size, u0 is the cut-off sedimentation velocity of the sedimentation tank, P0 is the percentage of sedimentation velocity less than u0, P is the residual particle concentration ratio, g is the gravitational acceleration, and μ is the dynamic viscosity coefficient of water.

[0031] In steps 1-5, the whole coagulation dosing system is described to calculate the effluent turbidity y Turout The expression is:

[0032]

[0033] where x Turin is the turbidity of raw water, and m0 is the average mass of flocs whose sedimentation rate can reach the cut-off sedimentation rate of the sedimentation tank.

[0034] Step 2 includes:

[0035] Step 2-1, select dosage as the input of the prediction model, effluent turbidity as the output of the prediction model, and construct a single-input single-output prediction model;

[0036] Step 2-2, select the basic CARIMA model as the prediction model, as shown below;

[0037] A(z -1 )y(k)=B(z -1 )u(k-1)+ξ(k) / Δ

[0038] Among them, y(k), u(k), ξ(k) represent the output, input and interference signal of the prediction model at time k respectively, A(z -1 ), B(z -1 ) are n a Order z -1 Polynomials and n b Order z -1 polynomial, Δ is the difference operator;

[0039] Step 2-3, for the mechanism model of coagulation dosing process established in step 1, build it in MATLAB, where the input is the inlet turbidity and the dosage, and the output is the outlet turbidity. The inlet turbidity is fixed, the dosage is changed, and the historical data is obtained;

[0040] Step 2-4, using iterative learning least squares method, combined with the historical data obtained in step 2-3 to identify the parameters to be identified in step 2-2;

[0041] Step 2-5, in order to solve the prediction model, introduce the Diophantine equation:

[0042] 1=E j (z -1 )A(z -1 )Δ+z -j F j (z -1 )

[0043] In the formula, E j and F j is a polynomial uniquely determined by the dimension of the output polynomial A and the prediction length j, and has:

[0044] E j (z -1 )=e0+e1z -1 +…+e j-1 z -(j-1)

[0045]

[0046] where e j-1 and represents the j-1th and nth terms of the above polynomial a The coefficient of the term;

[0047] By solving the Diophantine equation, let y0(k+j)=H j Δu(k-1)+F j y(k), represents the influence of the current output and the past control increment, H j and F j Indicates z -1 Polynomial, thus obtaining the optimal prediction output after j steps at the current time k for:

[0048]

[0049] Among them G j represents a polynomial uniquely determined by the dimension of the input polynomial B and the prediction length j, and Δu(k+j-1) is the control amount increment at time k+j-1.

[0050] Rewriting it into matrix form gives:

[0051]

[0052] in represents the predicted output in the future, Y0(k) represents the actual output vector of the present and the past, and G represents the vector generated by G j The polynomial matrix constructed, ΔU(k) ​​represents the current and future control increment vectors;

[0053] in,

[0054]

[0055] in are the coefficients of the polynomial, Δu(k+N u -1) is k+N u -1 moment control increment;

[0056] Step 2-6, the calculation of the optimal control law is related to the selected index, considering the ability to quickly track the reference trajectory function and the small tracking error, and then considering the influence of the control increment; define the rolling optimization objective function J, as shown in the following formula:

[0057]

[0058] where y d (k) is the expected output at time k, λ>0 is the weighted control factor;

[0059] Step 2-7, online identification is performed by iterative learning of the least squares method, and the self-correction algorithm is implemented based on the data and parameters of the previous k time periods.

[0060] Steps 2-4 include:

[0061] Step 2-4-1, collect a set of repeated excitation data;

[0062] Step 2-4-2, for t = 0, 1, ..., N, give the initial parameter estimates And the initial covariance matrix P required by the iterative formula -1 (t) value, and set k = 0;

[0063] Step 2-4-3, when repeating the kth time, record the input data u of the kth iteration k (t) and output data y k (t);

[0064] Step 2-4-4, calculate the parameter estimate for the kth iteration according to the following formula: and the covariance matrix and P k (t);

[0065]

[0066]

[0067] φ k (t) = [-y k (t-1),…,-y k (tn),-u k (t-1),...-u k (tm)] T ,θ(t)=[a1(t),…,a n (t),b1(t),…,b m (t)] T ;

[0068] where φ k (t) represents the input vector of the kth iteration, and P k-1 (t) represent the parameter estimates and covariance matrix of the k-1th iteration, e(k) represents the deviation of the parameter estimate, T represents the matrix transpose, and y k (t-1)…y k (tn) represents the output from time t-1 to time tn of the kth iteration, a1(t)…a n (t) represents the coefficient of the output from time t-1 to time tn of the kth iteration, u k (t-1)…u k (tm) represents the input from the kth iteration t-1 to tm, b1(t)…b m (t) represents the input coefficient from time t-1 to time tm of the kth iteration;

[0069] Step 2-4-5, check whether k is greater than 101 or whether the deviation between the parameter estimate and the true value is less than 1%. If the stopping condition is met, stop the operation and obtain the parameter estimate of the prediction model. Otherwise, set k=k+1 and go to step 2-3-3.

[0070] Steps 2-6 also include:

[0071] Step 2-6-1, adopt step-by-step control for the objective function J, by setting the target increment value Δu(k) = δ, And bring it into the objective function J:

[0072]

[0073] where Y d (k) is the expected value vector, β is the step factor, and δ is the control increment;

[0074] Step 2-6-2, let in For the sign of the partial derivative, we get the control law:

[0075]

[0076] in:

[0077]

[0078] The actual output control quantity u(k) at time k is obtained as:

[0079] u(k)=u(k-1)+δ.

[0080] The present invention also provides a storage medium storing a computer program or instruction. When the computer program or instruction is run on a computer, the steps of the improved generalized predictive control method for coagulation dosing process are executed.

[0081] Compared with the prior art, the present invention has the following significant advantages:

[0082] (1) The present invention provides a mechanism model of the tap water coagulation and dosing process, which can provide a simulation environment for algorithm verification. Compared with the traditional transfer function and fitting modeling methods, this method is closer to the actual situation and improves the accuracy and credibility of the model.

[0083] (2) The present invention has made improvements in two aspects of the generalized predictive control algorithm: first, the least squares method with forgetting factor is improved to the iterative learning least squares method, thereby achieving complete estimation of the fast time-varying model parameters; second, a step factor is introduced and a recursive relationship is formed, which solves the problem that the matrix inversion operation cannot be performed in the PLC during the actual coagulation and dosing process, and at the same time, the control increment is constrained, so that the algorithm can be applied in the actual production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0085] Figure 1 It is a structural diagram of the coagulation dosing mechanism process model provided by the present invention.

[0086] Figure 2 It is a generalized predictive control structure diagram of the improved coagulation dosing process of the present invention.

[0087] Figure 3a , Figure 3b , Figure 3c , Figure 3dThis is a comparison chart of the least squares identification results with the forgetting factor.

[0088] Figure 3e , Figure 3f , Figure 3g , Figure 3h It is a comparison chart of parameter predictions using iterative learning least squares method.

[0089] Figure 4 This is a comparison chart of the control effects of the improved GPC algorithm and the PID algorithm. DETAILED DESCRIPTION

[0090] This embodiment is as follows Figure 1 Taking the mechanism model of the coagulation dosing process shown in FIG. 1 as an example, an improved generalized predictive control method for the coagulation dosing process of the present invention is used to control it.

[0091] Step 1: construct a mechanism model of the coagulation and dosing process. According to the actual scenario, the entire model is divided into three cascade models, including a mixing model, a flocculation model, and a sedimentation model.

[0092] Step 1-1, determine the input, output and interference of the coagulation dosing process mechanism model. The input is the dosage (g / m 3 ) and raw water turbidity (NTU), the output is the outlet turbidity (NTU), the interference is the inlet flow rate (m 3 / s).

[0093] Step 1-2, establish a mixing model, and regard the mixing tank as a continuous flow complete mixing reactor (CSTR), which can ensure that the reactants continue to enter evenly and are fully mixed in the tank, and the reaction products are continuously discharged from the reactor. Therefore, the concentration and composition of the output material are basically consistent with the concentration and composition of the reactants in the reactor. In addition, it is assumed that the micelle concentration of the material inside the mixing tank is equal to the micelle concentration of the output. Based on the above assumptions, the lumped parameter model can be used to describe the rapid mixing process. From the law of conservation of mass, the concentration change formula before and after the mixing tank can be obtained:

[0094]

[0095]

[0096] Where V m is the volume of the mixing pool, x Dosout (t) and x Turout (t) are the coagulant concentration and turbidity of the effluent from the mixing tank at time t, respectively.

[0097] When the surface charge of the colloid is close to neutral, the colloid particles will be in a destabilized state and the coagulation process will start. Considering the interaction area ratio between the coagulant and the colloid surface in the water, the concentration of the unstable colloid can be approximately calculated. However, when the coagulant is added in excess, the unstable floccules will further adsorb too many aluminum hydroxide particles, causing them to regain charge and stabilize. Therefore, the expression for calculating the concentration of unstable colloids in the effluent of the mixing tank can be obtained:

[0098]

[0099] Where N0 is the concentration of destabilized colloidal particles, x Dosbest is the optimal dosing concentration, m Tur and m Dos are the unit masses of impurity particles and coagulant particles respectively.

[0100] Since the colloidal particles have been initially flocculated during the mixing stage, the present invention assumes that the number concentration of flocs in the effluent of the mixing tank is N i The following distribution is satisfied:

[0101]

[0102] Where V i is the volume of flocs of size class i.

[0103] Step 1-3, establish a flocculation model, and regard the flocculation process in the flocculation tank as two parallel phenomena of aggregation and fragmentation. In order to quantitatively describe it, this part can use the population equilibrium model widely used in discrete phase systems, and the change rate r of the number concentration of flocs with size class i can be derived. i The expression is:

[0104]

[0105] The first two terms on the right side of the above formula represent the number of flocs of size class i generated by the collision of flocs of size class smaller than i; the third and fourth terms represent the number of flocs of size class i reduced by agglomeration with other flocs; the fifth term represents the number of flocs of size class i reduced by breakage; the last term represents the number of flocs of size class i generated by breakage of flocs of size class greater than i. i,j is the collision efficiency of flocs of size classes i and j, β i,j is the collision frequency of flocs, N i is the number concentration of flocs, S i is the floc breakage rate, Γ i,j is the broken distribution function.

[0106] Step 1-4, establish a sedimentation model, and regard the sedimentation process in the sedimentation tank as free sedimentation, that is, no parameters of the flocs will change after entering the sedimentation tank, and once they sink to the bottom, they will not return to the upper clear water. The following formula can be used to estimate the suspended particle removal rate:

[0107]

[0108]

[0109] Steps 1-5, organically connect the sub-models to obtain the mechanism model of the dynamic behavior of the entire coagulation dosing system, and derive the turbidity y of the effluent of the entire coagulation dosing system. Turout The expression is:

[0110]

[0111] In step 1 of the present invention, each sub-model is designed independently. A mixed model is established based on the reactor theory and its model and adsorption-electric neutralization effect; a flocculation model that can describe the two phenomena of aggregation and fragmentation in the flocculation process is established based on the population equilibrium model; and a precipitation model is established based on the free sedimentation process of particles.

[0112] Step 2, according to Figure 2 The control block diagram shown designs an improved generalized predictive control method for the coagulation dosing process. The CARIMA parameter model is obtained by online identification using the least squares method based on iterative learning. A step factor is also introduced, and a step-by-step generalized predictive control algorithm is used for rolling optimization to generate the optimal control quantity, avoiding matrix inversion operations.

[0113] Step 2-1, select the dosage as the input of the prediction model and the effluent turbidity as the output of the prediction model to build a single-input single-output prediction model.

[0114] Step 2-2, establish a forecasting model. Here we choose the basic CARIMA model, as shown in the following formula;

[0115] A(z -1 )y(k)=B(z -1 )u(k-1)+ξ(k) / Δ

[0116] Among them, y(k), u(k), ξ(k) are the output, input and interference signals of the system, and A(z -1 ), B(z -1 ) are n a and n b Order z -1 polynomial, Δ is the difference operator.

[0117] Step 2-3, using iterative learning least squares method, combined with historical data of the coagulation dosing process mechanism model to identify the parameters to be identified in step 2-2, the steps include:

[0118] Step 2-3-1, collect data of n system repeated operations, where t = 0, 1, ..., N;

[0119] Step 2-3-2, for t = 0, 1, ..., N, give the initial estimate of this parameter and P required for the iterative formula -1 (t) value, and set k = 0;

[0120] Step 2-3-3, when repeating the kth time, record the data u k (t),y k (t);

[0121] Step 2-3-4, calculate according to the following formula and P k (t);

[0122]

[0123]

[0124] where φ k (t) = [-y k (t-1),…,-y k (tn),-u k (t-1),...-u k (tm)] T ,θ(t)=[a1(t),…,a n (t),b1(t),…,b m (t)] T

[0125] Step 2-3-5, check the stopping condition, if the condition is met, stop running and get the parameter estimation value of the prediction model Otherwise, set k=k+1 and go to step 2-3-3.

[0126] from Figure 3a , Figure 3b , Figure 3c , Figure 3d , Figure 3e , Figure 3f , Figure 3g , Figure 3h It can be seen that the identification effect of the iterative learning least squares method for fast time-varying parameters is significantly better than the least squares method based on the forgetting factor. Figure 3a , Figure 3b , Figure 3c , Figure 3d , Figure 3e , Figure 3f , Figure 3g , Figure 3h The horizontal axis represents time, and the vertical axis represents parameter values.

[0127] Step 2-4, in order to solve the prediction model, introduce the Diophantine equation:

[0128] 1=E j (z -1 )A(z -1 )Δ+z -j F j (z -1 )

[0129] Where:

[0130] E j (z -1 )=e0+e1z -1 +…+e j-1 z -(j-1)

[0131]

[0132] By solving the Diophantine equation, the optimal prediction output equation after j steps at the current time k is obtained as:

[0133]

[0134] Rewriting it into matrix form gives:

[0135]

[0136] in,

[0137]

[0138] Step 2-5, select the objective function of the control system, as shown below:

[0139]

[0140] where y d (k) is the expected output, and λ>0 is the weighted control factor.

[0141] Step 2-5-1, adopt step-by-step control for the above objective function. This idea is to explicitly plan the future control quantity increment by setting the target increment value Δu(k) = δ. And bring it into the objective function:

[0142]

[0143] Step 2-5-2, since it is an unconstrained optimization problem, there is an analytical solution, so let The control law is obtained:

[0144]

[0145] in:

[0146]

[0147] Therefore, the actual output control quantity is:

[0148] u(k)=u(k-1)+δ;

[0149] Step 2-6, finally, the iterative learning least squares method introduced above is used for online identification, and the self-correction algorithm is implemented based on the data and parameters of the previous k time periods. The control effect comparison diagram of the self-correction algorithm with PID control is shown in the figure below: Figure 4 As shown, the horizontal axis represents time and the vertical axis represents the turbidity of the effluent water.

[0150] The present invention provides an improved generalized predictive control method for coagulation dosing process. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. An improved generalized predictive control method for coagulation dosing process, characterized in that: The steps include: Step 1: construct a mechanism model of the coagulation and dosing process. According to the actual scenario, the mechanism model of the coagulation and dosing process is divided into three cascade models for construction. The three cascade models are a mixing model, a flocculation model, and a sedimentation model. Step 2, design an improved generalized predictive control method for the coagulation dosing process, use the least squares method based on iterative learning to obtain the controlled autoregressive moving average model online, introduce the step factor, and use the step-by-step generalized predictive control algorithm to perform rolling optimization to generate the optimal control quantity, avoiding matrix inversion operations.

2. The method according to claim 1, characterized in that Step 1 includes: Step 1-1, determine the input, output and interference of the coagulation dosing process mechanism model: the input is the dosage and the raw water turbidity, the output is the effluent turbidity, and the interference is the influent flow rate; Step 1-2, establish a mixed model; Steps 1-3, establishing a flocculation model; Steps 1-4, establishing a precipitation model; Steps 1-5: connect the three sub-models of mixing model, flocculation model and sedimentation model in series one by one to obtain the mechanism model of the dynamic behavior of the entire coagulation dosing system, and derive the effluent turbidity y that describes the entire coagulation dosing system. Turout expression.

3. The method according to claim 2, characterized in that Step 1-2 includes: the mixing process of the mixing model simulates the process of adding coagulant to raw water through a metering pump, the mixed liquid after addition enters the mixing tank for rapid stirring and mixing, and the mixing tank is regarded as a continuous flow complete mixing reactor to ensure that the reactants continue to enter evenly and are fully mixed in the tank, and the reaction products are continuously discharged from the reactor; the concentration and composition of the output material are consistent with the concentration and composition of the reactants in the reactor; the micelle concentration of the substance inside the mixing tank is set to be equal to the micelle concentration of the output; the lumped parameter model is used to describe the rapid mixing process, and the concentration change formula before and after the mixing tank is obtained: Where V m is the volume of the mixing pool, x Dosin (t) and x Turin (t) are the coagulant concentration and raw water turbidity at time t; x Dosout (t) and x Turout (t) are the coagulant concentration and turbidity of the effluent from the mixing tank at time t, q represents the raw water flow rate; d is the integral symbol; When the surface charge of the colloid is close to neutral, the colloid particles will be in a destabilized state and the coagulation process will start. At this time, the following formula is used to calculate the concentration of unstable colloids in the effluent of the mixing tank: Where N0 is the concentration of destabilized colloidal particles, x Dosout and x Turout are the coagulant concentration and turbidity of the effluent from the mixing tank, respectively, Dosbest is the optimal dosing concentration, m Tur and m Dos are the unit mass of impurity particles and the unit mass of coagulant particles respectively; Set the number concentration N of flocs in the effluent of the mixing tank i The following distribution is satisfied: Where V0 represents the average volume of flocs, V i is the volume of flocs of size class i; exp is the natural exponential function.

4. The method according to claim 3, characterized in that Steps 1-3 include: the flocculation model regards the flocculation process in the flocculation tank as two parallel phenomena of aggregation and fragmentation, and derives the following description of the change rate of floc number concentration r of size class i: i The expression is: Among them, N i and N j are the number concentration of flocs of size class i and size class j, respectively, α i,j is the collision efficiency of flocs of size classes i and j, β i,j is the collision frequency of flocs of size classes i and j, S i is the breakage rate of flocs of size class i, Γ i,j is the fragmentation distribution function of flocs of size classes i and j.

5. The method according to claim 4, characterized in that Steps 1-4 include: the sedimentation model regards the sedimentation process in the sedimentation tank as free sedimentation, and calculates the suspended particle removal rate η using the following formula: Where u is the sedimentation velocity, ρ and ρ0 are the density of impurity particles and water respectively, d is the particle size, u0 is the cut-off sedimentation velocity of the sedimentation tank, P0 is the percentage of sedimentation velocity less than u0, P is the residual particle concentration ratio, g is the gravitational acceleration, and μ is the dynamic viscosity coefficient of water.

6. The method according to claim 5, characterized in that In steps 1-5, the whole coagulation dosing system is described to calculate the effluent turbidity y Turout The expression is: where x Tur in is the turbidity of raw water, and m0 is the average mass of flocs whose sedimentation rate can reach the cut-off sedimentation rate of the sedimentation tank.

7. The method according to claim 6, characterized in that Step 2 includes: Step 2-1, select dosage as the input of the prediction model, effluent turbidity as the output of the prediction model, and construct a single-input single-output prediction model; Step 2-2, select the basic CARIMA model as the prediction model, as shown below; A(z -1 )y(k)=B(z -1 )u(k-1)+ξ(k) / Δ Among them, y(k), u(k), ξ(k) represent the output, input and interference signal of the prediction model at time k respectively, A(z -1 ), B(z -1 ) are n a Order z -1 Polynomials and n b Order z -1 polynomial, Δ is the difference operator; Step 2-3, for the mechanism model of coagulation dosing process established in step 1, the input is the inlet turbidity and the dosage, the output is the outlet turbidity, the inlet turbidity is fixed, the dosage is changed, and the historical data is obtained; Step 2-4, using iterative learning least squares method, combined with the historical data obtained in step 2-3 to identify the parameters to be identified in step 2-2; Step 2-5, in order to solve the prediction model, introduce the Diophantine equation: 1=E j (With -1 )Until -1 )Δ+z -j F j (With -1 ) In the formula, E j and F j is a polynomial uniquely determined by the dimension of the output polynomial A and the prediction length j, and has: E j (With -1 )=e0+e1z -1 +…+e j-1 With -(j-1) where e j-1 and represents the j-1th and nth terms of the above polynomial a The coefficient of the term; By solving the Diophantine equation, let y0(k+j)=H j Δu(k-1)+F j y(k), represents the influence of the current output and the past control increment, H j and F j Indicates z -1 Polynomial, thus obtaining the optimal prediction output after j steps at the current time k for: Among them G j represents a polynomial uniquely determined by the dimension of the input polynomial B and the prediction length j, Δu(k+j-1) is the control increment at time k+j-1; Rewriting it into matrix form gives: in represents the predicted output in the future, Y0(k) represents the actual output vector of the present and the past, and G represents the vector generated by G j The polynomial matrix constructed, ΔU(k) ​​represents the current and future control increment vectors; in, in are the coefficients of the polynomial, Δu(k+N u -1) is k+N u -1 moment control increment; Step 2-6, define the rolling optimization objective function J, as shown below: where y d (k) is the expected output at time k, λ>0 is the weighted control factor; Step 2-7, online identification is performed by iterative learning of the least squares method, and the self-correction algorithm is implemented based on the data and parameters of the previous k time periods.

8. The method according to claim 7, characterized in that Steps 2-4 include: Step 2-4-1, collect a set of repeated excitation data; Step 2-4-2, for t = 0, 1, ..., N, give the initial parameter estimates And the initial covariance matrix P required by the iterative formula -1 (t) value, and set k = 0; Step 2-4-3, when repeating the kth time, record the input data u of the kth iteration k (t) and output data y k (t); Step 2-4-4, calculate the parameter estimate for the kth iteration according to the following formula: and the covariance matrix and P k (t); φ k (t)=[-y k (t-1),…,-y k (t-n),-u k (t-1),…-u k (t-m)] T , θ(t)=[a1(t),…,a n (t),b1(t),…,b m (t)] T ; where φk(t) represents the input vector of the kth iteration, and P k-1 (t) represent the parameter estimates and covariance matrix of the k-1th iteration, e(k) represents the deviation of the parameter estimate, T represents the matrix transpose, and y k (t-1)…y k (tn) represents the output from time t-1 to time tn of the kth iteration, a1(t)…a n (t) represents the coefficient of the output from time t-1 to time tn of the kth iteration, u k (t-1)…u k (tm) represents the input from the kth iteration t-1 to tm, b1(t)…b m (t) represents the input coefficient from time t-1 to time tm of the kth iteration; Step 2-4-5, check whether k is greater than 101 or whether the deviation between the parameter estimate and the true value is less than 1%. If the stopping condition is met, stop the operation and obtain the parameter estimate of the prediction model. Otherwise, set k=k+1 and go to step 2-3-3.

9. The method according to claim 8, characterized in that Steps 2-6 also include: Step 2-6-1, adopt step-by-step control for the objective function J, by setting the target increment value Δu(k) = δ, And bring it into the objective function J: where Y d (k) is the expected value vector, β is the step factor, and δ is the control increment; Step 2-6-2, let in For the sign of the partial derivative, we get the control law: in: The actual output control quantity u(k) at time k is obtained as: u(k)=u(k-1)+δ.

10. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is run on a computer, the steps of the method according to any one of claims 1 to 9 are executed.

Images