CN109669352B - Oily sewage treatment process optimization control method based on self-adaptive multi-target particle swarm - Google Patents

Abstract

The invention relates to an optimal control method for an oily sewage treatment process based on self-adaptive multi-target particle swarm, which mainly solves the problems that the standard particle swarm algorithm in the prior art is easy to generate premature convergence and has low search precision. Firstly, establishing an optimized set value of dissolved oxygen and nitrate nitrogen and a target function of aeration energy consumption and pumping energy consumption in the oily sewage treatment process through a fuzzy neural network; secondly, optimizing an objective function for treating the oily sewage by adopting a self-adaptive multi-objective particle swarm optimization method, and simultaneously obtaining optimized set values of dissolved oxygen and nitrate nitrogen; and finally, the controller is utilized to carry out tracking control on the optimized set values of the dissolved oxygen and the nitrate nitrogen, and the technical scheme of multi-objective optimization control in the oily sewage treatment process is completed, so that the problems are well solved, and the method can be used for optimization control in the oily sewage treatment process.

Description

Oily sewage treatment process optimization control method based on self-adaptive multi-target particle swarm

Technical Field

The invention relates to an optimal control method for an oily sewage treatment process based on self-adaptive multi-target particle swarm. The invention realizes dissolved oxygen DO and nitrate nitrogen S in the oily sewage treatment process by utilizing an optimization control method based on self-adaptive multi-target particle swarm_NOControl of concentration, dissolved oxygen DO and nitrate nitrogen S_NOThe concentration of the water is a key control parameter in the oily sewage treatment process, and has important influence on the oily sewage treatment effect, the effluent quality and the energy consumption in the oily sewage treatment process. The optimization control method based on the self-adaptive multi-target particle swarm is applied to the oily sewage treatment processIn the process, dissolved oxygen DO and nitrate nitrogen S are realized_NOThe concentration is optimally controlled, the energy consumption of the oily sewage treatment process is reduced, the investment and the operation cost are saved, the stable and efficient operation of a sewage treatment plant is ensured, and the method belongs to the field of water treatment and also belongs to the field of intelligent control.

Background

With the rapid development of economic society in China, the process of the oil industry is accelerated, the pollution of human activities to water environment is increased continuously, and the oily sewage has extremely serious influence on human beings, animals, plants and even the whole ecological system. Meanwhile, the increase of national economy and the enhancement of public environmental awareness lead the automatic technology of oily sewage treatment to meet the unprecedented development opportunity; how to prevent and treat water pollution and how to effectively treat and reuse oily sewage in time becomes an urgent problem in China; however, the oily sewage treatment process has high electric energy consumption and high operation cost, and the research on the significance of optimizing and controlling the sewage treatment process to realize energy conservation and consumption reduction is significant, and is a necessary development trend of the sewage treatment industry in the future.

The essence of the sewage biochemical treatment process is that the life activity of microorganisms in the sludge is utilized to decompose organic pollutants in the sewage, so that the sewage is purified. In the process of treating the oily sewage, the main control variables are dissolved oxygen DO and nitrate nitrogen S_NOAnd (4) concentration. Dissolved oxygen DO and nitrate nitrogen S_NOThe change of the concentration can directly influence the nitrification process and the denitrification process, thereby influencing the energy consumption in the oily sewage treatment process. The nitrification reaction is mainly carried out under the aerobic condition, when the dissolved oxygen DO concentration is increased, the concentration of the effluent ammonia nitrogen and the total nitrogen presents a descending trend, when the dissolved oxygen DO concentration is increased to a certain range, the variation amplitude of the effluent ammonia nitrogen begins to be weakened, the total nitrogen is influenced by the nitrate nitrogen, and the total nitrogen concentration is increased while the concentration of the nitrate nitrogen is increased. However, the denitrification reaction is mainly carried out in an anoxic environment in the oily sewage treatment process, and nitrate nitrogen S in an anoxic zone_NOThe concentration is an important index for measuring the denitrification effect, reflects the progress of the denitrification reaction process and converts nitrate nitrogen S_NOThe concentration is controlled to be properWithin the range, the potential of denitrification reaction can be improved. Therefore, the dissolved oxygen DO and the nitrate nitrogen S are dynamically and optimally controlled in real time_NOThe concentration is very necessary for ensuring the effluent quality and saving energy and reducing consumption. Because of the characteristics of suspended matters, high chroma content, diversified components of organic matters, nonlinearity, time-varying property, uncertain dynamics and the like of the oily sewage treatment process, the dissolved oxygen DO and the nitrate nitrogen S are increased_NOThe control difficulty of (c); in recent years, some scholars optimize the oily sewage treatment process by adopting a multi-objective genetic algorithm (MOGA) based on BSM1 so as to minimize aeration energy consumption and pumping energy consumption. However, most of the optimization schemes belong to static optimization and steady-state optimization, dynamic real-time adjustment according to the change of the quality and quantity of inlet water is difficult to carry out, and most of the optimization schemes are realized by adopting a genetic algorithm. Compared with a genetic algorithm, the particle swarm algorithm has the advantages of high convergence speed, no need of complex cross variation operation, simple algorithm, few parameters and easiness in implementation, but the standard particle swarm algorithm is easy to have the problems of premature convergence and low search precision, so that the self-adaptive multi-target particle swarm algorithm is designed, the convergence precision of the algorithm is improved, and the dissolved oxygen DO and nitrate nitrogen S can be well realized_NOThe concentration is optimally controlled, the operation cost is reduced, and the method has good practical application value.

Disclosure of Invention

The technical problem to be solved by the invention is that the standard particle swarm algorithm in the prior art is easy to generate premature convergence and has low search precision, and the invention provides a novel optimization control method for the oily sewage treatment process based on self-adaptive multi-target particle swarm. The method has the advantages of no premature convergence and high search precision.

In order to solve the problems, the technical scheme adopted by the invention is as follows: an optimized control method for oily sewage treatment process based on self-adaptive multi-target particle swarm comprises the steps of firstly establishing an optimized set value of dissolved oxygen and nitrate nitrogen and an objective function of aeration energy consumption and pumping energy consumption in the oily sewage treatment process through a fuzzy neural network; secondly, aiming at the defects that the standard particle swarm algorithm is easy to premature convergence and low in convergence precision, a self-adaptive multi-target particle swarm optimization method is adopted to realize the optimization of the oily sewage treatment target function and obtain the optimized set values of dissolved oxygen and nitrate nitrogen; finally, the controller is used for tracking and controlling the optimized set values of the dissolved oxygen and the nitrate nitrogen to complete the multi-objective optimization control of the oily sewage treatment process;

the method specifically comprises the following steps:

1) designing an objective function for optimal control of the oily sewage treatment process;

(2) establishing a relational expression between the optimal set values of dissolved oxygen and nitrate nitrogen and aeration energy consumption and pumping energy consumption by using a fuzzy neural network;

(3) the effluent quality is restrained;

(4) obtaining a Pareto optimal solution by utilizing a self-adaptive multi-target particle swarm optimization objective function, which specifically comprises the following steps:

initializing the velocity v of the particle swarm_i(0) Position a_i(0) Inertia weight ω_i(0) Learning factor c_1i(0) And c_2i(0) And setting the initial position of each particle as the current historical optimal position p_i(0) Simultaneously setting a population scale S, a maximum evolution algebra M and a variable dimension D;

secondly, calculating the fitness value of each particle according to the objective function, and determining the individual optimal solution p of the t iteration_i(t) defined by the formula:

the non-dominated solution set A (t) is updated by A (t-1), and the formula is as follows:

wherein A (t) ═ a₁(t),a₂(t),…,a_Q(t)]Q is the maximum capacity of the knowledge base A (t), K is the number of non-dominant solutions contained in the knowledge base,

denotes a_i(t-1) and p_i(t-1) are independent of each other;

third, determining the global optimal solution gBest (t +1) of the t +1 iteration

Wherein, gBest (t +1) is the global optimal solution of the t +1 th iteration, and dgBest (t +1) is the global optimal solution of the t +1 th iteration with better diversity, wherein, the definition formula of the dgBest (t +1) is as follows:

dgBest(t+1)＝a(t),a(t)∈Μ_tbest. (4)

cgBest (t +1) is a global optimal solution with better convergence in the t +1 th iteration, and the definition formula is as follows:

cgBest(t+1)＝argmaxCD_t(a_i(t)), (5)

CD_t(a_i(t)) non-dominant solution a_i(t) convergence, i ═ 1,2,3 … K, e (t) is the distribution entropy of the non-dominated solution set for the t th iteration, which is defined by the formula:

wherein u is_nIs the cell of the non-dominant solution in the knowledge base, N is 1,2,3 …, N is the total number of cells, p_t(u_n) Is the t-th iteration cell u_nThe expression of the probability distribution function of (1) is:

m_t(u_n) Is the t-th iteration cell u_nNumber of non-dominant solutions, M_t(u_n) Is the t-th iteration cell u_nOf all non-dominant solutions contained in (c), wherein the smallest m is_t(u_n) Is marked as m_tbestAccordingly, M_tbestIs optimal for the t-th iterationThe non-dominant solution set of (2), and the selection of cgBest (t +1) is determined by the degree of convergence that reflects the dominant relationship, thereby determining the degree of convergence CD_t(a_i(t)) is defined as:

wherein,

is the jth solution a that can be non-dominated_i(t) solution, DS_t(a_i(t)) is the t-th iteration non-dominated solution a_i(t) a dominant intensity defined by the formula:

DS_t(a_i(t)) is the non-dominant solution a for the t-th iteration_i(t) the total number of solutions that can be dominated, A (t) is the set of non-dominated solutions for the t iteration, A (t-1) is the set of non-dominated solutions for the t-1 iteration;

updating the speed and position of each particle;

judging whether the maximum evolution times M is reached, if so, ending, otherwise, returning to the step II;

(5) finding a group of satisfied optimal solutions in the current state from a group of Pareto optimal solutions obtained by a self-adaptive multi-target particle swarm algorithm to serve as an optimal set value of a bottom controller;

(6) executing a bottom layer control strategy, and respectively passing the concentrations of dissolved oxygen and nitrate nitrogen through an oxygen conversion coefficient K of the aeration tank_La5And internal reflux quantity Q_aAnd (6) carrying out adjustment.

In the above technical solution, preferably, the controller is a PID.

In the above technical solution, preferably, the objective function for optimal control of the oily wastewater treatment process is:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the optimization variable at time k, x₁(k) The dissolved oxygen DO concentration setpoint at time k, x₂(k) Is the nitrate nitrogen concentration set value at the time k, f_AE(x) And f_PE(x) Respectively are relational expressions g of aeration energy consumption, pumping energy consumption and optimization variables₁(x) And g₂(x) Respectively are relational expressions of effluent ammonia nitrogen, total nitrogen concentration and optimization variable, C₁And C₂Respectively the constraint values of the effluent ammonia nitrogen and the total nitrogen, C₁∈[0,4],C₂∈[0,18]；

And

respectively representing the lower limit and the upper limit, x, of the optimum set value of dissolved oxygen₁∈[0.4,3],

And

respectively represents the lower limit and the upper limit, x, of the nitrate nitrogen concentration optimization set value₂∈[0.5,2]The optimization period is 2 h.

In the above technical solution, preferably, the fuzzy neural network is calculated as follows:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the fuzzy neural network input at time k, c_j＝[c_1j，c_2j]，σ_j＝[σ_1j，σ_2j]Respectively the central vector and the width vector of the jth neuron of the RBF layer,

the output of the RBF layer of the j-th neuron at the k-th moment, wherein P is the number of neurons of the RBF layer and the rule layer; v. of_l(k) V (k) is [ v ] for the l-th regular layer output corresponding to the k-th time₁(k)，v₂(k)…v_P(k)]^TOutputting a vector for the k-th time rule layer; w is a¹＝[w₁ ¹，w₂ ¹…w_P ¹]And w²＝[w₁ ²，w₂ ²…w_P ²]Is the weight vector between the output neuron and the rule layer, y (k) is the output of the neural network,

the actual physical quantity output of the sewage treatment system is obtained based on BSM1 model data;

setting the target function of network adjustment at the moment k as follows:

by adopting a gradient descent algorithm, the weight updating formula is as follows:

in the formula, alpha_q(k)＝[θ_q(k)^T c_q(k)^T σ_q(k)^T]^TThe learning parameter vector of the network is represented, and the network learning rate eta is 0.01;

in the above technical solution, preferably, the constraint in the optimization model is processed by a penalty function method, where a constraint penalty term is defined as:

f_penalty(x)＝max{g₁(x)-4,0}+max{g₂(x)-18,0}, (16)

the aeration energy consumption and pumping energy consumption objective function added with penalty items is as follows:

and converting the established constraint optimization problem in the oily sewage treatment process into an unconstrained multi-objective optimization problem, wherein epsilon is a penalty factor and is set as a larger positive real number.

In the above technical solution, preferably, the speed and the position of each particle are updated:

v_i(t+1)＝ω_i(t)v_i(t)+c₁r₁(p_i(t)-a_i(t))+c₂r₂(gBest_d(t)-a_i(t))； (18)

a_i(t+1)＝a_i(t-1)+v_i(t+1)；

wherein r is₁And r₂Respectively representing the best previous position coefficient and the global optimum position coefficient, r₁And r₂Take [0,1]Any number of (a).

In the above technical solution, preferably, in the step (6), a bottom layer control strategy is executed, and the concentrations of dissolved oxygen and nitrate nitrogen respectively pass through the 5 th zone oxygen conversion coefficient K of the aeration tank_La5And internal reflux quantity Q_aAnd (6) carrying out adjustment.

In the above-described aspect, preferably, in step (r), the velocity v of the particle group is initialized_i(0) Position a_i(0) Inertia weight ω_i(0) Learning factor c_1i(0) And c_2i(0) And setting the initial position of each particle as the current historical optimal position p_i(0)，Meanwhile, the population size S is set to be 40, the maximum evolution algebra M is set to be 30, and the variable dimension D is set to be 2.

The invention provides an optimal control method for an oily sewage treatment process based on self-adaptive multi-target particle swarm, which comprises the following steps:

(1) designing an objective function for optimal control of the oily sewage treatment process:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the optimization variable at time k, x₁(k) The dissolved oxygen DO concentration setpoint at time k, x₂(k) Is the nitrate nitrogen concentration set value at the time k, f_AE(x) And f_PE(x) Respectively are relational expressions g of aeration energy consumption, pumping energy consumption and optimization variables₁(x) And g₂(x) Respectively are relational expressions of effluent ammonia nitrogen, total nitrogen concentration and optimization variable, C₁And C₂Respectively the constraint values of the effluent ammonia nitrogen and the total nitrogen, C₁∈[0,4],C₂∈[0,18]；

And

respectively representing the lower limit and the upper limit, x, of the optimum set value of dissolved oxygen₁∈[0.4,3],

And

respectively represents the lower limit and the upper limit, x, of the nitrate nitrogen concentration optimization set value₂∈[0.5,2]The optimization period is 2 h.

(2) Establishing a relational expression between the optimal set values of the dissolved oxygen and the nitrate nitrogen and the aeration energy consumption and the pumping energy consumption by using a fuzzy neural network, wherein the calculation mode of the fuzzy neural network is as follows:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the fuzzy neural network input at time k, c_j＝[c_1j，c_2j]，σ_j＝[σ_1j，σ_2j]Respectively the central vector and the width vector of the jth neuron of the RBF layer,

the output of the RBF layer of the j-th neuron at the k-th moment, wherein P is the number of neurons of the RBF layer and the rule layer; v. of_l(k) V (k) is [ v ] for the l-th regular layer output corresponding to the k-th time₁(k)，v₂(k)…v_P(k)]^TOutputting a vector for the k-th time rule layer; w is a¹＝[w₁ ¹，w₂ ¹…w_P ¹]And w²＝[w₁ ²，w₂ ²…w_P ²]Is the weight vector between the output neuron and the rule layer, y (k) is the output of the neural network,

the actual physical quantity output of the sewage treatment system is obtained based on BSM1 model data.

Setting the target function of network adjustment at the moment k as follows:

by adopting a gradient descent algorithm, the weight updating formula is as follows:

in the formula, alpha_q(k)＝[θ_q(k)^T c_q(k)^T σ_q(k)^T]^TThe network learning rate η is 0.01 for the learning parameter vector of the network.

(3) Effluent quality constraint treatment

And processing the constraint in the optimization model by adopting a penalty function method, wherein a constraint penalty term is defined as:

f_penalty(x)＝max{g₁(x)-4,0}+max{g₂(x)-18,0}, (7)

the aeration energy consumption and pumping energy consumption objective function added with penalty items is as follows:

namely, the established constraint optimization problem in the oily sewage treatment process is converted into an unconstrained multi-objective optimization problem. Wherein epsilon is a penalty factor and is set as a large positive real number.

(4) Obtaining a Pareto optimal solution by utilizing a self-adaptive multi-target particle swarm optimization objective function, which specifically comprises the following steps:

initializing the velocity v of the particle swarm_i(0) Position a_i(0) Inertia weight ω_i(0) Learning factor c_1i(0) And c_2i(0) And setting the initial position of each particle as the current historical optimal position p_i(0). Meanwhile, the population scale S is set to be 40, the maximum evolution algebra M is set to be 30, and the variable dimension D is set to be 2;

and secondly, calculating the fitness value of each particle according to the objective function. Determining the t-th iterationIndividual optimal solution p_i(t) defined by the formula:

the non-dominated solution set A (t) is updated by A (t-1), and the formula is as follows:

wherein A (t) ═ a₁(t),a₂(t),…,a_Q(t)]Q is the maximum capacity of the knowledge base A (t), K is the number of non-dominant solutions contained in the knowledge base,

denotes a_i(t-1) and p_i(t-1) are not mutually exclusive.

And thirdly, determining the global optimal solution gBest (t +1) of the t +1 th iteration.

Wherein, gBest (t +1) is the global optimal solution of the t +1 th iteration, and dgBest (t +1) is the global optimal solution of the t +1 th iteration with better diversity, wherein, the definition formula of the dgBest (t +1) is as follows:

dgBest(t+1)＝a(t),a(t)∈Μ_tbest. (12)

cgBest (t +1) is a global optimal solution with better convergence in the t +1 th iteration, and the definition formula is as follows:

cgBest(t+1)＝argmaxCD_t(a_i(t)), (13)

CD_t(a_i(t)) non-dominant solution a_i(t) convergence, i ═ 1,2,3 … K, e (t) is the distribution entropy of the non-dominated solution set for the t th iteration, which is defined by the formula:

wherein u is_nIs the cell of the non-dominant solution in the knowledge base, N is 1,2,3 …, N is the total number of cells, p_t(u_n) Is the t-th iteration cell u_nThe expression of the probability distribution function of (1) is:

m_t(u_n) Is the t-th iteration cell u_nNumber of non-dominant solutions, M_t(u_n) Is the t-th iteration cell u_nA solution set of all non-dominant solutions contained in (a). Wherein the smallest m_t(u_n) Is marked as m_tbestAccordingly, M_tbestIs the optimal non-dominated solution set for the t-th iteration. Further, the selection of cgBest (t +1) is determined by the convergence degree that can reflect the dominant relationship, and the convergence degree CD_t(a_i(t)) is defined as:

wherein,

is the jth solution a that can be non-dominated_i(t) solution, DS_t(a_i(t)) is the t-th iteration non-dominated solution a_i(t) a dominant intensity defined by the formula:

DS_t(a_i(t)) is the non-dominant solution a for the t-th iteration_i(t) the total number of solutions that can be dominated, A (t) is the set of non-dominated solutions for the t iteration, A (t-1) is the set of non-dominated solutions for the t-1 iteration;

updating the speed and the position of each particle:

v_i(t+1)＝ω_i(t)v_i(t)+c₁r₁(p_i(t)-a_i(t))+c₂r₂(gBest_d(t)-a_i(t))； (18)

a_i(t+1)＝a_i(t-1)+v_i(t+1)； (19)

wherein r is₁And r₂Respectively representing the best previous position coefficient and the global optimum position coefficient, r₁And r₂Take [0,1]Any number of (a);

judging whether the maximum evolution times M is reached, if so, ending, otherwise, returning to the step II.

(5) And finding a group of satisfactory optimal solutions in the current state from a group of Pareto optimal solutions obtained by the self-adaptive multi-target particle swarm algorithm to serve as the optimal set values of the bottom PID controller.

(6) Executing a bottom layer PID control strategy, and respectively passing the concentrations of dissolved oxygen and nitrate nitrogen through the oxygen conversion coefficient K of the 5 th subarea of the aeration tank_La5And internal reflux quantity Q_aAnd (6) carrying out adjustment.

The invention aims at the complex, dynamic and unstable biochemical reaction process of the current oily sewage treatment process by an activated sludge method and the characteristics of nonlinearity, time-varying property and hysteresis; simultaneously dissolving oxygen DO and nitrate nitrogen S_NOThe concentration has strong coupling relation, so as to meet the requirement of reducing the operation energy consumption when the effluent quality reaches the standard and realize the dissolved oxygen DO and the nitrate nitrogen S_NOThe multi-target control of the concentration adopts an oily sewage treatment model predictive control method based on multi-target particle swarm to realize dissolved oxygen DO and nitrate nitrogen S_NOThe concentration control has the characteristics of high control precision, good stability and the like; the invention adopts an oily sewage treatment model based on self-adaptive multi-target particle swarm to dissolve oxygen DO and nitrate nitrogen S in the sewage treatment process_NOThe concentration is optimally controlled, and the optimal control method solves the problem of optimal solution of a plurality of objective functions, so that the controller better meets the change of the current environment, and the dissolved oxygen DO and the nitrate nitrogen S are realized_NOThe concentration is accurately controlled in a real-time closed loop manner, and the condition that a plurality of controllers are required to be designed for the current sewage treatment plant is avoidedThe complex process of line control has the characteristics of strong real-time performance, simple structure and the like, and obtains better technical effect.

Drawings

FIG. 1 is an overall structure diagram of the adaptive multi-objective particle swarm optimization control system of the invention;

FIG. 2 is a graph showing the results of the dissolved oxygen DO concentration in the control system of the present invention;

FIG. 3 is a DO concentration error plot of the dissolved oxygen of the control system of the present invention;

FIG. 4 shows nitrate nitrogen S in the control system of the present invention_NOA concentration result graph;

FIG. 5 shows nitrate nitrogen S in the control system of the present invention_NOConcentration result error plot.

The present invention will be further illustrated by the following examples, but is not limited to these examples.

Detailed Description

[ example 1 ]

An optimized control method for oily sewage treatment process based on self-adaptive multi-target particle swarm is shown in figure 1, and realizes dissolved oxygen DO and nitrate nitrogen S in the oily sewage treatment process_NOMulti-objective optimization control of concentration; the control method obtains dissolved oxygen DO and nitrate nitrogen S through online modeling of a fuzzy neural network_NOThe function relation between the optimized set value of the concentration and the aeration energy consumption, the pumping energy consumption and the effluent quality is realized by applying the optimization control method based on the self-adaptive multi-target particle swarm to the oily sewage treatment process_NOThe concentration is optimally controlled, the energy consumption of the oily sewage treatment process is reduced, the investment and the operation cost are saved, the stable and efficient operation of a sewage treatment plant is ensured, and the method belongs to the field of water treatment and also belongs to the field of intelligent control.

The invention adopts the following technical scheme and implementation steps:

1. the design of the optimal control method for the oily sewage treatment process based on the self-adaptive multi-target particle swarm comprises the following steps: aiming at the dissolved oxygen DO concentration and nitrate nitrogen S in a sequencing batch intermittent activated sludge system_NOControl is carried out by taking aeration energy consumption and pumping energy consumption as control quantitiesDissolved oxygen DO and nitrate nitrogen S_NOThe concentration is controlled quantity, and the whole framework of the self-adaptive multi-target particle swarm optimization control system is as shown in figure 1;

(1) designing an objective function for optimal control of the oily sewage treatment process:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the optimization variable at time k, x₁(k) Is a dissolved oxygen concentration set value at the time k, x₂(k) Is the nitrate nitrogen concentration set value at the time k, f_AE(x) And f_PE(x) Respectively are relational expressions g of aeration energy consumption, pumping energy consumption and optimization variables₁(x) And g₂(x) Respectively are relational expressions of effluent ammonia nitrogen, total nitrogen concentration and optimization variable, C₁And C₂Respectively the constraint values of the effluent ammonia nitrogen and the total nitrogen, C₁∈[0,4],C₂∈[0,18]；

And

respectively representing the lower limit and the upper limit, x, of the optimum set value of dissolved oxygen₁∈[0.4,3],

And

respectively represents the lower limit and the upper limit, x, of the nitrate nitrogen concentration optimization set value₂∈[0.5,2]The optimization period is 2 h.

(2) Establishing a relational expression between the optimal set values of the dissolved oxygen and the nitrate nitrogen and the aeration energy consumption and the pumping energy consumption by using a fuzzy neural network, wherein the calculation mode of the fuzzy neural network is as follows:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the fuzzy neural network input at time k, c_j＝[c_1j，c_2j]，σ_j＝[σ_1j，σ_2j]Respectively the central vector and the width vector of the jth neuron of the RBF layer,

the output of the RBF layer of the j-th neuron at the k-th moment, wherein P is the number of neurons of the RBF layer and the rule layer; v. of_l(k) V (k) is [ v ] for the l-th regular layer output corresponding to the k-th time₁(k)，v₂(k)…v_P(k)]^TOutputting a vector for the k-th time rule layer; w is a¹＝[w₁ ¹，w₂ ¹…w_P ¹]And w²＝[w₁ ²，w₂ ²…w_P ²]Is the weight vector between the output neuron and the rule layer, y (k) is the output of the neural network,

the actual physical quantity output of the sewage treatment system is obtained based on BSM1 model data.

Setting the target function of network adjustment at the moment k as follows:

by adopting a gradient descent algorithm, the weight updating formula is as follows:

in the formula, alpha_q(k)＝[θ_q(k)^T c_q(k)^T σ_q(k)^T]^TThe network learning rate η is 0.01 for the learning parameter vector of the network.

(3) Effluent quality constraint treatment

And processing the constraint in the optimization model by adopting a penalty function method, wherein a constraint penalty term is defined as:

f_penalty(x)＝max{g₁(x)-4,0}+max{g₂(x)-18,0}, (7)

the aeration energy consumption and pumping energy consumption objective function added with penalty items is as follows:

namely, the established constraint optimization problem in the oily sewage treatment process is converted into an unconstrained multi-objective optimization problem. Wherein epsilon is a penalty factor and is set as a large positive real number.

(4) Obtaining a Pareto optimal solution by utilizing a self-adaptive multi-target particle swarm optimization objective function, which specifically comprises the following steps:

initializing the velocity v of the particle swarm_i(0) Position a_i(0) Inertia weight ω_i(0) Learning factor c_1i(0) And c_2i(0) And setting the initial position of each particle as the current historical optimal position p_i(0). Meanwhile, the population scale S is set to be 40, the maximum evolution algebra M is set to be 30, and the variable dimension D is set to be 2;

and secondly, calculating the fitness value of each particle according to the objective function. Determining individual optimal solution p of t-th iteration_i(t) defined by the formula:

the non-dominated solution set A (t) is updated by A (t-1), and the formula is as follows:

wherein A (t) ═ a₁(t),a₂(t),…,a_Q(t)]Q is the maximum capacity of the knowledge base A (t), K is the number of non-dominant solutions contained in the knowledge base,

denotes a_i(t-1) and p_i(t-1) are not mutually exclusive.

And thirdly, determining the global optimal solution gBest (t +1) of the t +1 th iteration.

Wherein, gBest (t +1) is the global optimal solution of the t +1 th iteration, and dgBest (t +1) is the global optimal solution of the t +1 th iteration with better diversity, wherein, the definition formula of the dgBest (t +1) is as follows:

dgBest(t+1)＝a(t),a(t)∈Μ_tbest. (12)

cgBest (t +1) is a global optimal solution with better convergence in the t +1 th iteration, and the definition formula is as follows:

cgBest(t+1)＝argmaxCD_t(a_i(t)), (13)

CD_t(a_i(t)) non-dominant solution a_i(t) convergence, i ═ 1,2,3 … K, e (t) is the distribution entropy of the non-dominated solution set for the t th iteration, which is defined by the formula:

wherein u is_nIs a cell of the non-dominant solution in the knowledge base,n is 1,2,3 …, N is the total number of cells, p_t(u_n) Is the t-th iteration cell u_nThe expression of the probability distribution function of (1) is:

m_t(u_n) Is the t-th iteration cell u_nNumber of non-dominant solutions, M_t(u_n) Is the t-th iteration cell u_nA solution set of all non-dominant solutions contained in (a). Wherein the smallest m_t(u_n) Is marked as m_tbestAccordingly, M_tbestIs the optimal non-dominated solution set for the t-th iteration. Further, the selection of cgBest (t +1) is determined by the convergence degree that can reflect the dominant relationship, and the convergence degree CD_t(a_i(t)) is defined as:

wherein,

is the jth solution a that can be non-dominated_i(t) solution, DS_t(a_i(t)) is the t-th iteration non-dominated solution a_i(t) a dominant intensity defined by the formula:

DS_t(a_i(t)) is the non-dominant solution a for the t-th iteration_i(t) the total number of solutions that can be dominated, A (t) is the set of non-dominated solutions for the t iteration, A (t-1) is the set of non-dominated solutions for the t-1 iteration;

updating the speed and the position of each particle:

v_i(t+1)＝ω_i(t)v_i(t)+c₁r₁(p_i(t)-a_i(t))+c₂r₂(gBest_d(t)-a_i(t))； (18)

a_i(t+1)＝a_i(t-1)+v_i(t+1)； (19)

wherein r is₁And r₂Respectively representing the best previous position coefficient and the global optimum position coefficient, r₁And r₂Take [0,1]Any number of (a); judging whether the maximum evolution times M is reached, if so, ending, otherwise, returning to the step II.

(5) And finding a group of satisfactory optimal solutions in the current state from a group of Pareto optimal solutions obtained by the self-adaptive multi-target particle swarm algorithm to serve as the optimal set values of the bottom PID controller.

(6) And executing a bottom layer PID control strategy, wherein the parameters of the PID controller are set as follows: k_P,1＝200，K_I,1＝15，K _D,12 and K_P,2＝20000，K_I,2＝5000，K_D,2400. The concentrations of dissolved oxygen and nitrate nitrogen respectively pass through the 5 th subarea oxygen conversion coefficient K of the aeration tank_La5And internal reflux quantity Q_aAnd (6) carrying out adjustment.

(7) Using PID controller to control dissolved oxygen DO and nitrate nitrogen S_NOThe concentration optimization set value is tracked and controlled, and the output of the whole control system is dissolved oxygen DO and nitrate nitrogen S_NOOptimizing the set value and tracking control value of the concentration value; fig. 2 shows the dissolved oxygen DO concentration optimization setpoint and tracking control values for the system, X-axis: time, in days, Y-axis: the unit of the optimized set value and the tracking control value of the dissolved oxygen DO is milligram/liter, the solid line is the optimized set value of the dissolved oxygen DO concentration, and the dotted line is the actual tracking control value of the dissolved oxygen DO; the error between the optimized set dissolved oxygen DO concentration value and the actual dissolved oxygen DO tracking control concentration value is shown in FIG. 3, X axis: time, in days, Y-axis: dissolved oxygen DO concentration error in milligrams per liter; FIG. 4 shows nitrate nitrogen S of the system_NOOptimization set value and tracking control value, X-axis: time, in days, Y-axis: nitrate nitrogen S_NOIn mg/l, and the solid line is nitrate nitrogen S_NOConcentration optimization set point, dotted line is nitrate nitrogen S_NOTracking control of concentrationA value of the metric; optimized setting of nitrate nitrogen S_NOConcentration value and actual nitrate nitrogen S_NOError in tracking control concentration values as shown in fig. 5, X-axis: time, in days, Y-axis: nitrate nitrogen S_NOThe error in concentration, in mg/l, proved the effectiveness of the method.

Particular attention is paid to: the invention is described for convenience only, and the dissolved oxygen DO and nitrate nitrogen S are adopted_NOThe concentration control and the ammonia nitrogen control in the sewage treatment process can also be applied to the control of the ammonia nitrogen and the like, and the control by adopting the principle of the invention is within the scope of the invention.

Claims (8)

1. An optimized control method for oily sewage treatment process based on self-adaptive multi-target particle swarm comprises the steps of firstly establishing an optimized set value of dissolved oxygen and nitrate nitrogen and an objective function of aeration energy consumption and pumping energy consumption in the oily sewage treatment process through a fuzzy neural network; secondly, aiming at the defects of easy premature convergence and low convergence precision of a standard particle swarm algorithm, a self-adaptive multi-target particle swarm optimization method is adopted to realize the optimization of an oily sewage treatment target function, and simultaneously, optimized set values of dissolved oxygen and nitrate nitrogen are obtained, so that premature convergence is avoided, and the search precision is high; finally, the controller is used for tracking and controlling the optimized set values of the dissolved oxygen and the nitrate nitrogen, so that the complex process that a plurality of controllers are required to be designed for control in the current sewage treatment plant is avoided, and the multi-objective optimized control of the oily sewage treatment process is completed;

the method specifically comprises the following steps:

(1) designing an objective function for optimal control of the oily sewage treatment process;

(2) establishing a relational expression between the optimal set values of dissolved oxygen and nitrate nitrogen and aeration energy consumption and pumping energy consumption by using a fuzzy neural network;

(3) the effluent quality is restrained;

(4) obtaining a Pareto optimal solution by utilizing a self-adaptive multi-target particle swarm optimization objective function, which specifically comprises the following steps:

initializing the velocity v of the particle swarm_i(0) Position a_i(0) The inertia weightHeavy omega_i(0) Learning factor c_1i(0) And c_2i(0) And setting the initial position of each particle as the current historical optimal position p_i(0) Simultaneously setting a population scale S, a maximum evolution algebra M and a variable dimension D;

secondly, calculating the fitness value of each particle according to the objective function, and determining the individual optimal solution p of the t iteration_i(t) defined by the formula:

the non-dominated solution set A (t) is updated by A (t-1), and the formula is as follows:

wherein A (t) ═ a₁(t),a₂(t),…,a_Q(t)]Q is the maximum capacity of the knowledge base A (t), K is the number of non-dominant solutions contained in the knowledge base,

denotes a_i(t-1) and p_i(t-1) are independent of each other;

third, determining global optimum solution gBest (t +1) of the t-th iteration

Wherein, gBest (t +1) is the global optimal solution of the t +1 th iteration, and dgBest (t +1) is the global optimal solution of the t +1 th iteration with better diversity, wherein, the definition formula of the dgBest (t +1) is as follows:

dgBest(t+1)＝a(t),a(t)∈Μ_tbest.

cgBest (t +1) is a global optimal solution with better convergence in the t +1 th iteration, and the definition formula is as follows:

cgBest(t+1)＝arg maxCD_t(a_i(t)),

CD_t(a_i(t)) non-dominant solution a_i(t) convergence, i ═ 1,2,3 … K, e (t) is the distribution entropy of the non-dominated solution set for the t th iteration, which is defined by the formula:

wherein u is_nIs the cell of the non-dominant solution in the knowledge base, N is 1,2,3 …, N is the total number of cells, p_t(u_n) Is the t-th iteration cell u_nThe expression of the probability distribution function of (1) is:

m_t(u_n) Is the t-th iteration cell u_nNumber of non-dominant solutions, M_t(u_n) Is the t-th iteration cell u_nOf all non-dominant solutions contained in (c), wherein the smallest m is_t(u_n) Is marked as m_tbestAccordingly, M_tbestIs a non-dominated solution set optimal for the t-th iteration, and the selection of cgBest (t +1) is determined by the degree of convergence that reflects the dominance relationship, and the degree of convergence CD_t(a_i(t)) is defined as:

wherein,

is the jth solution a that can be non-dominated_i(t) solution, DS_t(a_i(t)) is the t-th iteration non-dominated solution a_i(t) a dominant intensity defined by the formula:

DS_t(a_i(t)) is the non-dominant solution a for the t-th iteration_i(t) the total number of solutions that can be dominated, A (t) is the set of non-dominated solutions for the t iteration, B (t-1) is the set of non-dominated solutions for the t-1 iteration;

updating the speed and position of each particle;

judging whether the maximum evolution times M is reached, if so, ending, otherwise, returning to the step II;

(5) finding a group of satisfied optimal solutions in the current state from a group of Pareto optimal solutions obtained by a self-adaptive multi-target particle swarm algorithm to serve as an optimal set value of a bottom controller;

(6) executing a bottom layer control strategy, and respectively passing the concentrations of dissolved oxygen and nitrate nitrogen through an oxygen conversion coefficient K of the aeration tank_La5And internal reflux quantity Q_aAnd (6) carrying out adjustment.

2. The oily sewage treatment process optimization control method based on the adaptive multi-target particle swarm of claim 1, wherein the controller is PID.

3. The oily sewage treatment process optimization control method based on the adaptive multi-objective particle swarm as claimed in claim 1, wherein the objective function for the oily sewage treatment process optimization control is as follows:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the optimization variable at time k, x₁(k) The dissolved oxygen DO concentration setpoint at time k, x₂(k) Nitre at time kNitrogen concentration set value, f_AE(x) And f_PE(x) Respectively are relational expressions g of aeration energy consumption, pumping energy consumption and optimization variables₁(x) And g₂(x) Respectively are relational expressions of effluent ammonia nitrogen, total nitrogen concentration and optimization variable, C₁And C₂Respectively the constraint values of the effluent ammonia nitrogen and the total nitrogen, C₁∈[0,4],C₂∈[0,18]；

And

respectively representing the lower limit and the upper limit, x, of the optimum set value of dissolved oxygen₁∈[0.4,3],

And

respectively represents the lower limit and the upper limit, x, of the nitrate nitrogen concentration optimization set value₂∈[0.5,2]The optimization period is 2 h.

4. The oily sewage treatment process optimization control method based on the adaptive multi-target particle swarm as claimed in claim 1, wherein the fuzzy neural network is calculated in the following way:

wherein x (k) ═ x₁(k)，x₂(k)]^TFor the fuzzy neural network input at time k, c_j＝[c_1j，c_2j]，σ_j＝[σ_1j，σ_2j]Respectively the central vector and the width vector of the jth neuron of the RBF layer,

the output of the RBF layer of the j-th neuron at the k-th moment, wherein P is the number of neurons of the RBF layer and the rule layer; v. of_l(k) V (k) is [ v ] for the l-th regular layer output corresponding to the k-th time₁(k)，v₂(k)…v_P(k)]^TOutputting a vector for the k-th time rule layer;

and

is the weight vector between the output neuron and the rule layer, y (k) is the output of the neural network,

the actual physical quantity output of the sewage treatment system is obtained based on BSM1 model data;

setting the target function of network adjustment at the moment k as follows:

by adopting a gradient descent algorithm, the weight updating formula is as follows:

in the formula, alpha_q(k)＝[θ_q(k)^T c_q(k)^T σ_q(k)^T]^TAs a networkThe network learning rate η of (1) is 0.01.

5. The oily sewage treatment process optimization control method based on the adaptive multi-target particle swarm is characterized in that constraints in an optimization model are processed by adopting a penalty function method, wherein a constraint penalty term is defined as:

f_penalty(x)＝max{g₁(x)-4,0}+max{g₂(x)-18,0},

the aeration energy consumption and pumping energy consumption objective function added with penalty items is as follows:

and converting the established constraint optimization problem in the oily sewage treatment process into an unconstrained multi-objective optimization problem, wherein epsilon is a penalty factor and is set as a larger positive real number.

6. The oily sewage treatment process optimization control method based on the adaptive multi-target particle swarm of claim 1, which is characterized in that the speed and the position of each particle are updated:

v_i(t+1)＝ω_i(t)v_i(t)+c₁r₁(p_i(t)-a_i(t))+c₂r₂(gBest_d(t)-a_i(t))；

a_i(t+1)＝a_i(t-1)+v_i(t+1)；

wherein r is₁And r₂Respectively representing the best previous position coefficient and the global optimum position coefficient, r₁And r₂Take [0,1]Any number of (a).

7. The method for optimally controlling the oily sewage treatment process based on the adaptive multi-target particle swarm in the claim 1, wherein in the step (6), a bottom layer control strategy is implemented, and the dissolved oxygen and nitrate nitrogen concentrations respectively pass through the 5 th subarea oxygen conversion coefficient K of the aeration tank_La5And internal reflux quantity Q_aAnd (6) carrying out adjustment.

8. The method for optimally controlling the oily sewage treatment process based on the self-adaptive multi-target particle swarm in the claim 1, wherein in the step (r), the speed v of the particle swarm is initialized_i(0) Position a_i(0) Inertia weight ω_i(0) Learning factor c_1i(0) And c_2i(0) And setting the initial position of each particle as the current historical optimal position p_i(0) Meanwhile, the population size S is set to 40, the maximum evolution algebra M is set to 30, and the variable dimension D is set to 2.

Images