CN115327890B - Method for optimizing main steam pressure of PID control thermal power depth peak shaving unit by improved crowd searching algorithm - Google Patents

Abstract

The invention provides a method for optimizing the main steam pressure of a PID control thermal power depth peak shaver set by an improved crowd searching algorithm, which is characterized in that a fuzzy step length is used as a guiding position updating mode to easily induce the algorithm to sink into local optimum, crowd lack of information exchange is easy to reduce crowd diversity and the like, and an original crowd searching algorithm is jointly improved by adopting a reverse differential evolution mechanism and a self-adaptive t distribution strategy, so that the capability of deviating from the local optimum is further enhanced. The improved SOA algorithm adopted by the invention has definite meaning and simple parameter setting in the design process of adaptability function, optimizing step length and direction, iterative updating of individual position and the like in the process of setting PID parameters, and the invention can directly model by adopting on-site actual data, so that the designed PID controller has stronger pertinence to the production process and engineering practical value.

Description

A method for optimizing the main steam pressure of thermal power deep peak-shaving units by using an improved crowd search algorithm to control PID

Technical Field

The invention belongs to the technical field of thermal control of thermal power units, and specifically relates to a method for optimizing PID control of main steam pressure of a thermal power deep peak regulation unit using an improved crowd search algorithm.

Background technique

In order to improve the consumption rate of new energy and the flexibility of system regulation, thermal power units will take on more tasks of peak load and frequency regulation of the power grid. The large hysteresis, large inertia and nonlinear control characteristics of the main steam pressure of thermal power units are particularly obvious under deep peak load conditions. The resulting problems such as slow response of main steam pressure control and large dynamic deviation seriously restrict the peak load ramp rate of thermal power units and are not conducive to the safe and stable operation of the units.

At present, thermal power units generally adopt a coordinated control system based on boiler following, that is, the boiler adjusts the main steam pressure and the turbine adjusts the load. The boiler master control adopts a PID regulator, and designs the differential feedforward of the main steam pressure deviation and the unit load command feedforward. For subcritical thermal power units, the boiler master control PID generates the fuel master control adjustment command and sends it to the fuel master control. The fuel master control PID is responsible for generating fuel instructions and serves as a coal feeder (direct-blowing pulverizing system) or a pulverizer (intermediate storage pulverizing system). The fuel master control PID is mostly a fast follow-up regulator, with simple parameter setting, and the regulation quality can generally meet the requirements of fuel quantity regulation accuracy; while the regulated quantity of the boiler master control PID is the boiler main steam pressure, and the setting of its regulation parameters plays a decisive role in the regulation quality of the main steam pressure. Field tests have found that under deep peak-shaving conditions, due to changes in boiler combustion characteristics and turbine flow characteristics, the parameter setting of the main steam pressure control PID regulator in the conventional load section has poor applicability and cannot meet the regulation quality requirements. It is necessary to use advanced optimization algorithms to optimize the main steam pressure control system.

Summary of the invention

The technical problem to be solved by the present invention is to provide a method for optimizing PID control of main steam pressure under deep peak regulation conditions of thermal power units using an improved crowd optimization algorithm, and to control the main steam pressure of thermal power units using a method for optimizing PID control parameters based on an improved crowd search algorithm (Seeker Optimization Algorithm, SOA), so that the main steam pressure has good performance in tracking the set value and has good robustness.

The technical solution provided by the present invention is a method for optimizing PID control of main steam pressure of a thermal power deep peak regulation unit by using an improved crowd search algorithm, comprising the following steps:

(1) Establish a main steam pressure identification transfer function model based on field test data;

(2) PID parameter optimization based on crowd search algorithm;

(3) The reverse differential evolution mechanism and the adaptive t-distribution strategy are used to jointly improve the original population search algorithm, optimize the parameters of the main steam pressure control PID, insert the optimized control parameters into the on-site DCS configuration, and verify the control effect through a constant value disturbance test.

Furthermore, in step (1), the main steam pressure identification modeling based on the field data is carried out. The steady-state operation condition of a deep peak-shaving thermal power unit is experimentally modeled, the amount of fuel entering the furnace is changed stepwise, the main steam pressure response curve is obtained, and the main steam pressure transfer function model is obtained by least square method identification. The main steam pressure is a typical thermal process with self-balancing ability. The model structure can be selected as formula (1), and the transfer function model as formula (2) is obtained.

In the formula: k is the static gain of the main steam pressure transfer function; n is the order of the transfer function; τ is the delay time; s is the Laplace operator.

Furthermore, in step (2), the main steam pressure PID control parameters are optimized based on a crowd search algorithm.

It can be seen from formula (2) that the main steam pressure model of the thermal power unit under deep peak regulation conditions is a second-order inertia plus pure delay object. The on-site PID adjustment mostly adopts the trial and error method based on experience. The debugging process is time-consuming and the dynamic indicators of the adjustment process are not easy to meet the accuracy requirements.

At present, the built-in PID modules of mainstream DCS control systems at home and abroad all adopt parallel PID, which forms the control law as shown in formula (3) according to the deviation e(t) between the input value rin(t) and the output value yout(t). Formula (5) is the transfer function form of PID, where _Kp is the proportional coefficient, _Ti is the integral time constant, and _Td is the differential time constant. The three are the parameters to be optimized.

e(t)＝rin(t)-yout(t) (3)

In the formula, e(t) is the deviation between the system output and the expected output; u(t) is the control output signal of PID.

Select a suitable sampling period and discretize the PID controller and the controlled object. When using MATLAB for algorithm programming simulation, you can use its built-in function c2d for discretization.

Parameter encoding: The three parameters of PID are taken as a search individual, and the dimension of the position vector of each individual is D=3. It is defined that there are S individuals in the population P, so the population P can be expressed by formula (6):

Selection of fitness function: The fitness function is the only indicator for evaluating the quality of individuals in the search optimization process. It is also the link between the SOA algorithm and the control system, guiding the algorithm to evolve towards the control target. The selection principles are as follows: In order to obtain excellent dynamic characteristics of regulation, the time integral performance index of the absolute value of the error is used as the minimum objective function; in order to prevent the controller output from being too large and damaging the field equipment, the control input square term is introduced; in order to avoid overshoot, moderate penalty control is used, and the overshoot is used as an indicator. In this way, the objective function is shown in formula (7). Among them, the values of each weight are generally: ω ₁ = 0.999, ω ₂ = 0.001, ω ₃ = 100.

Determination of search step length: The uncertain reasoning behavior of SOA uses the approximation ability of fuzzy systems to simulate the intelligent search behavior of people to establish the connection between the objective function value and the step length. The Gaussian membership function is used to represent the search step length fuzzy variable.

In the above formula (8), u _A is the Gaussian membership; x is the input variable; u and δ are the membership function parameters. When the output variable exceeds [u-3δ, u+3δ], if the membership is less than 0.0111, it can be ignored, so u _min = 0.0111 is set, and in order to obtain a faster convergence speed, u _max = 0.95 is set.

Using a linear membership function, the best position has a maximum membership u _max = 1, the worst position has a minimum membership u _min = 0.0111, and u < 1.0 at other positions, as shown in formula (9).

In the above formula (9), u _ij is the membership degree of the objective function value i in the j-dimensional search space; α _ij is the search step size in the j-dimensional search space; δ _ij is the Gaussian membership function parameter, and its value is determined by formulas (10) and (11):

ω＝(iter _max -iter)/iter _max (11)

In the above formula, x _min and x _max are the positions of the minimum and maximum function values in the same population, respectively; ω is the inertia weight, which decreases linearly from 0.9 to 0.1 with the increase of evolutionary generations; iter and iter _max are the current and maximum iteration numbers, respectively.

Determination of search direction: Through the analysis and modeling of people's selfish behavior, altruistic behavior and pre-action behavior, the expressions of three action directions are obtained as shown in equations (12)-(14):

Taking the above factors into consideration, the search direction is determined by randomly weighted geometric averaging of three directions, as shown in formula (15):

in, and They are The best position in is the collective best historical position of the neighborhood of the i-th search individual, is the best position experienced by the i-th search individual so far; sign(*) is the sign function; and is a constant in [0,1]; ω is the inertia weight.

Update of individual position: Based on the above analysis, the individual position can be updated according to the search direction and step size, as shown in the following formula.

Δx _ij (t+1) = α _ij (t) d _ij (t) (16)

x _ij (t+1) = x _ij (t) + Δ x _ij (t+1) (17)

Among them, Δx _ij (t+1) is the position change of the i-th search individual in the j-dimensional search space at the next moment, and x _ij (t) is the position of the i-th search individual in the j-dimensional search space at moment t.

Furthermore, the improvement of the crowd search algorithm in step (3): The standard crowd search algorithm relies on the mutual coordination between the empirical gradient direction and the uncertain reasoning behavior, and conducts extensive global search in the entire iterative process, thereby accumulating rich prior knowledge for mining the global optimal solution. However, the position update method guided by the fuzzy step size can easily induce the algorithm to fall into the local optimum, thereby causing the search to deviate from the optimal solution; the lack of information exchange among the crowd can easily reduce the diversity of the crowd, which will further limit the scope of algorithm development. In response to the above problems, the reverse differential evolution mechanism and the adaptive t distribution strategy are used to jointly improve the original crowd search algorithm to further enhance its ability to escape from the local optimum.

For any individual x _ij , the inverse solution is defined as

x _i ' _j = ub _j + lb _j - x _ij (18)

In the above formula, ub _j and lb _j are the upper and lower limits of the j-th dimension search space, respectively. In order to further enhance the diversity of the population, it is proposed to use differential evolution algorithm (DE) to update the position of the population, find high-quality individuals through mutation, crossover and selection strategies, and strengthen the information exchange between individuals.

For each individual vector _xi in the population, three different individual vectors are randomly selected and combined to generate mutant individuals:

_vi (t+1)= ^xr1 (t)+F( ^xr2 (t) ^-xr3 (t)) (19) In the above formula, _r1 , _r2 , _r3 are the numbers of three different differential individuals in [1,0.5N], and the scaling factor F is a random number in the range of [0,1].

The crossover operation is used to construct the test individual u _ij (t+1), and the construction method is:

Among them, rand generates random numbers between [0,1], and the range of the crossover probability factor CR is between [0,1]. f(*) is the fitness function, which calculates the fitness value of the new individual after the crossover operation. If the new individual has a better fitness, the original individual is replaced:

The global optimal position is adaptively updated using a mutation factor with the number of iterations i as the system parameter. A higher quality position is selected based on the greedy selection algorithm and participates in the next iteration. The adaptive t distribution is a random parameter group that combines the advantages of Gaussian distribution and Cauchy distribution. When it is used as a mutation factor to perturb the feasible solution, it can enable the algorithm to have a certain local random search ability to avoid falling into the local optimum. The adaptive mutation process of the optimal individual is shown in the following formula:

x _best (t+1)＝x _best (t)+x _best (t)*trnd(t) (22)

In the above formula, trnd is the parameter function of the adaptive t distribution, which can be called by the MATLAB function library; _xbest (t) is the current optimal solution, and _xbest (t+1) is the new solution after the adaptive t distribution is mutated.

In summary, the crowd search algorithm SOA optimizes the parameters of the main steam pressure control PID according to the above analysis process. In order to verify the steady-state and dynamic performance of the obtained regulation system, the set value step signal is used to excite the designed closed-loop control system.

Advantages of the present invention: Compared with other intelligent optimization algorithms, the improved SOA algorithm adopted in the present invention has clear meanings and simple parameter settings in the design process of fitness function, optimization step size and direction, individual position iterative update, etc. involved in the process of adjusting PID parameters. In addition, the present invention can directly use actual on-site data for modeling, so that the designed PID controller is more targeted to the production process and has more engineering practical value.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a schematic diagram of the control principle of the present invention;

FIG2 is a main steam pressure identification modeling curve diagram according to an embodiment of the present invention;

Fig. 3 is an algorithm flow chart of the present invention;

FIG4 is a PID parameter variation curve in an embodiment of the present invention;

5 is a curve diagram showing changes in the fitness function during SOA optimization in an embodiment of the present invention;

FIG6 is a graph showing a step disturbance response curve of a main steam pressure setting value in an embodiment of the present invention;

7 is a comprehensive comparison curve diagram of the step response of the main steam pressure setting value in the example of the present invention (SOA and improved SOA);

FIG8 is a main steam pressure regulation process curve of a thermal power unit in deep regulation condition using the algorithm of the present invention.

Detailed ways

In conjunction with the accompanying drawings, a 660MW deep peak-shaving thermal power unit is taken as an implementation object, and the parameters of the main steam pressure control system are optimized according to the technical method of the present invention, and the application of the intelligent algorithm based on crowd search optimization in the main steam pressure control system of the thermal power unit is described in detail. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Referring to Figure 1, the overall control principle of the present invention is introduced. The main steam pressure-fuel control object in the figure includes the generalized control objects of fuel command and actual fuel quantity control. When the fuel main control parameters are constant, the modeling of the main steam pressure can be made simpler. Considering the complexity and computational complexity of the crowd search algorithm, an offline parameter setting scheme is adopted while the original control logic of the DCS and the PID controller structure remain basically unchanged. Under the offline condition, the control system design is carried out for the controlled object of the identified main steam pressure transfer function, and the fitness function that has been reasonably constructed is continuously iterated and calculated through the crowd search algorithm to obtain a set of optimal PID control parameter values.

The main steam pressure control method under deep peak regulation conditions of thermal power units based on crowd search optimization algorithm (SOA) specifically includes the following steps:

(1) Using the on-site step disturbance test method, based on the system identification principle, the transfer function model of the main steam pressure object is established. The unit boiler master control and steam turbine master control are both switched to manual mode, and the fuel master control instruction is changed stepwise (5%) to obtain the main steam pressure response curve, and the data is subjected to bad value elimination and zero mean processing. Select the known structural expression of a typical thermal process with self-balancing ability as shown in formula (1), and only need to identify the parameters therein. The identification algorithm can adopt the least squares, particle swarm optimization algorithm, etc. The transfer function of the obtained fuel quantity-main steam pressure process is in the form of formula (2), and the identification result curve is shown in Figure 2.

In the formula: k is the static gain of the main steam pressure transfer function; n is the order of the transfer function; τ is the delay time; s is the Laplace operator.

(2) Optimize the main steam pressure PID control parameters based on the crowd search algorithm.

It can be seen from formula (2) that the main steam pressure model of the thermal power unit under deep peak regulation conditions is a second-order inertia plus pure delay object. The on-site PID adjustment mostly adopts the trial and error method based on experience. The debugging process is time-consuming and the dynamic indicators of the adjustment process are not easy to meet the accuracy requirements.

In this project case, the unit master control DCS adopts parallel PID, and the control law is formed according to the deviation e(t) between the input value rin(t) and the output value yout(t) as shown in formula (3). Formula (5) is the transfer function form of PID, where _Kp is the proportional coefficient, _Ti is the integral time constant, and _Td is the differential time constant. The three are the parameters to be optimized.

e(t)＝rin(t)-yout(t) (3)

Select a suitable sampling period and discretize the PID controller and the controlled object. When using MATLAB for algorithm programming simulation, you can use its built-in function c2d for discretization.

Parameter encoding: The three parameters of PID are taken as a search individual, and the dimension of the position vector of each individual is D=3. It is defined that there are S individuals in the population P, so the population P can be expressed by formula (6):

Selection of fitness function: The fitness function is the only indicator for evaluating the quality of individuals in the search optimization process. It is also the link between the SOA algorithm and the control system, guiding the algorithm to evolve towards the control target. The selection principles are as follows: In order to obtain excellent dynamic characteristics of regulation, the time integral performance index of the absolute value of the error is used as the minimum objective function; in order to prevent the controller output from being too large and damaging the field equipment, the control input square term is introduced; in order to avoid overshoot, moderate penalty control is used, and the overshoot is used as an indicator. In this way, the objective function is shown in formula (7). Among them, the values of each weight are generally: ω ₁ = 0.999, ω ₂ = 0.001, ω ₃ = 100.

Determination of search step length: The uncertain reasoning behavior of SOA uses the approximation ability of fuzzy systems to simulate the intelligent search behavior of people to establish the connection between the objective function value and the step length. The Gaussian membership function is used to represent the search step length fuzzy variable.

In the above formula (8), u _A is the Gaussian membership; x is the input variable; u and δ are the membership function parameters. When the output variable exceeds [u-3δ, u+3δ], if the membership is less than 0.0111, it can be ignored, so u _min = 0.0111 is set, and in order to obtain a faster convergence speed, u _max = 0.95 is set.

Using a linear membership function, the best position has a maximum membership u _max = 1, the worst position has a minimum membership u _min = 0.0111, and u < 1.0 at other positions, as shown in formula (9).

In the above formula (9), α _ij is the search step size of the j-dimensional search space; δ _ij is the Gaussian membership function parameter, and its value is determined by formulas (10) and (11):

ω＝(iter _max -iter)/iter _max (11)

In the above formula, x _min and x _max are the positions of the minimum and maximum function values in the same population, respectively; ω is the inertia weight, which decreases linearly from 0.9 to 0.1 with the increase of evolutionary generations; iter and iter _max are the current and maximum iteration numbers, respectively.

Determination of search direction: Through the analysis and modeling of people's selfish behavior, altruistic behavior and pre-action behavior, the expressions of three action directions are obtained as shown in equations (12)-(14):

Taking the above factors into consideration, the search direction is determined by randomly weighted geometric averaging of three directions, as shown in formula (15):

in, and They are The best position in is the collective best historical position of the neighborhood of the i-th search individual, is the best position experienced by the i-th search individual so far; sign(*) is the sign function; and is a constant in [0,1]; ω is the inertia weight.

Update of individual position: Based on the above analysis, the individual position can be updated according to the search direction and step size, as shown in the following formula.

Δx _ij (t+1) = α _ij (t) d _ij (t) (16)

x _ij (t+1) = x _ij (t) + Δ x _ij (t+1) (17)

Among them, Δx _ij (t+1) is the position change of the i-th search individual in the j-dimensional search space at the next moment, and x _ij (t) is the position of the i-th search individual in the j-dimensional search space at moment t.

Improvement of crowd search algorithm: The position update method guided by fuzzy step size can easily induce the algorithm to fall into local optimality, thus causing the search to deviate from the optimal solution; the lack of information exchange among the crowd can easily reduce the diversity of the crowd, which will further limit the scope of algorithm development. In response to the above problems, the reverse differential evolution mechanism and the adaptive t distribution strategy are used to jointly improve the original crowd search algorithm to further enhance its ability to escape from the local optimality.

For any individual x _ij , the inverse solution is defined as

x _i ' _j = ub _j + lb _j - x _ij (18)

In the above formula, ub _j and lb _j are the upper and lower limits of the j-th dimension search space, respectively. In order to further enhance the diversity of the population, it is proposed to use differential evolution algorithm (DE) to update the position of the population, find high-quality individuals through mutation, crossover and selection strategies, and strengthen the information exchange between individuals.

For each individual vector _xi in the population, three different individual vectors are randomly selected and combined to generate mutant individuals:

_vi (t+1)＝ ^xr1 (t)+F( ^xr2 (t) ^-xr3 (t)) (19)

In the above formula, r ₁ , r ₂ , r ₃ are the numbers of three different differential individuals in [1, 0.5N], and the scaling factor F is a random number in the range of [0, 1].

The crossover operation is used to construct the test individual u _ij (t+1), and the construction method is:

Among them, rand generates random numbers between [0,1], and the range of the crossover probability factor CR is between [0,1]. f(*) is the fitness function, which calculates the fitness value of the new individual after the crossover operation. If the new individual has a better fitness, the original individual is replaced:

The global optimal position is adaptively updated using a mutation factor with the number of iterations i as the system parameter. A higher quality position is selected based on the greedy selection algorithm and participates in the next iteration. The adaptive t distribution is a random parameter group that combines the advantages of Gaussian distribution and Cauchy distribution. When it is used as a mutation factor to perturb the feasible solution, it can enable the algorithm to have a certain local random search ability to avoid falling into the local optimum. The adaptive mutation process of the optimal individual is shown in the following formula:

x _best (t+1)＝x _best (t)+x _best (t)*trnd(t) (22)

In the above formula, x _best (t) is the current optimal solution, and x _best (t+1) is the new solution after the adaptive t distribution mutation.

According to the above steps, the improved crowd search algorithm SOA is implemented to optimize the parameters of the main steam pressure PID controller. The parameter change curve during the iteration process is shown in Figure 4. The parameters finally obtained by optimization are Kp=8.32, Ki=0.01, and Kd=52.1. Combined with the structure of the on-site DCS controller, and converted into actual differential form, the parameter sequence is placed in the on-site DCS logic configuration, and a relatively good control quality can be obtained. As shown in Figure 7, it is a comprehensive comparison curve of the step response of the main steam pressure set value in the example of the present invention. It can be seen that the control parameters obtained by the traditional SOA algorithm have a long step response output adjustment time under the control output condition constraint, and the static deviation is large; while the control parameters obtained by the improved SOA algorithm proposed in this paper have fast dynamic response, small static deviation, higher accuracy, and faster convergence in the set value step response process, which verifies the superiority of the improved algorithm. As shown in Figure 8, during the coordinated variable load test under the deep peak regulation condition of a supercritical unit using the method of the present invention, the actual main steam pressure has good performance in tracking the set value, and the control system indicators can meet the relevant industry technical standards.

The above is only a detailed description of a specific implementation scheme of the present invention, and does not limit the present invention. Any modifications, equivalent substitutions and improvements made on the design concept of the present invention should be included in the protection scope of the present invention.

Claims (4)

1. A method for optimizing main steam pressure of a PID control thermal power depth peak shaver set by an improved crowd searching algorithm is characterized by comprising the following steps:

(1) Establishing a main steam pressure identification transfer function model based on field test data;

(2) PID parameter optimization based on crowd searching algorithm;

(3) Carrying out joint improvement on an original crowd searching algorithm by adopting a reverse differential evolution mechanism and a self-adaptive t distribution strategy, carrying out parameter optimization on a main steam pressure control PID, putting the optimized control parameters into a site DCS configuration, and checking a control effect through a constant disturbance test;

the specific steps in the step (2) are as follows:

(a) For the parallel PID, a control rule as shown in a formula (3) is formed according to the deviation e (T) of an input value rin (T) and an output value yout (T), wherein a formula (5) is a transfer function form of the PID, K _p is a proportionality coefficient, T _i is an integral time constant, T _d is a differential time constant, and the three are setting parameters to be optimized;

e(t)＝rin(t)-yout(t) (3)

e (t) -deviation of system output from desired output in the formula; u (t) -control output signal of PID;

Selecting a proper sampling period, discretizing a PID controller and a controlled object, and discretizing by adopting a built-in function c2d when using matlab to carry out algorithm programming simulation;

(b) Parameter coding: taking three parameters of PID as one search individual, the dimension of the position vector of each individual is d=3, defining S individuals in the population P, so the population P is represented by formula (6):

(c) And (3) selecting a fitness function: the fitness function is a unique index for evaluating the individual quality in the search optimization process, and is also a tie for combining the SOA algorithm with the control system, and the algorithm is guided to continuously evolve towards a control target; the selection principle is as follows: in order to obtain excellent regulation dynamic characteristics, a time integral performance index of an absolute value of an error is adopted as a minimum objective function; to prevent excessive controller output from damaging the field device, a square term for control input is introduced; in order to avoid the overshoot phenomenon, moderate punishment control is adopted, and the overshoot is used as one of indexes; thus, the objective function is shown as formula (7); wherein, the value of each weight is: ω ₁＝0.999,ω₂＝0.001,ω₃ =100;

(d) Determination of search step size: the uncertain reasoning behavior of the SOA is to simulate intelligent searching behavior of a person by utilizing approximation capability of a fuzzy system, so as to establish connection between an objective function value and a step length, and a Gaussian membership function is adopted to represent a searching step length fuzzy variable;

In the above formula (8), u _A is a Gaussian membership degree; x is an input variable; u and delta are membership function parameters, when the output variable exceeds [ u-3 delta, u+3 delta ], if the membership degree is smaller than 0.0111, u _min =0.0111 is set, and in order to obtain a faster convergence rate, u _max =0.95 is set;

A linear membership function is adopted, the maximum membership degree u _max =1 is arranged at the optimal position, the minimum membership degree u _min =0.0111 is arranged at the worst position, and u is smaller than 1.0 at other positions, as shown in a formula (9),

In the above formula (9), u _ij is the membership degree of the j-dimensional search space objective function value i; alpha _ij is the search step size of the j-dimensional search space; δ _ij is a gaussian membership function parameter whose value is determined by equations (10) and (11):

ω＝(iter_max-iter)/iter_max (11)

In the above formula, x _min and x _max are the positions of the minimum and maximum function values in the same population, respectively; omega is an inertial weight, linearly decreasing from 0.9 to 0.1 with increasing algebra; the item and item _max are the current iteration number and the maximum iteration number, respectively;

(e) Determination of search direction: by analyzing and modeling the human's Living, lithe and pre-acting behaviors, three directions of action are expressed as formulas (12) - (14):

Combining the above factors, determining the search direction by using three-direction random weighted geometric average, as in equation (15):

Wherein, AndRespectively isIs the best position in the model (a); For the i-th searching of the collective historical best position of the neighborhood of the individual Searching for the i-th best location that the individual has experienced so far; sign is a sign function; And Is a constant within [0,1 ]; omega is the inertial weight;

(f) Updating the individual position: according to the analysis, the individual position can be updated according to the searching direction and the step length, and the method is shown in the following formula;

Δx_ij(t+1)＝α_ij(t)d_ij(t) (16)

x_ij(t+1)＝x_ij(t)+Δx_ij(t+1) (17)

Wherein Deltax _ij (t+1) is the position change quantity of the ith searching individual at the next moment of the j-dimensional searching space, and x _ij (t) is the position of the ith searching individual at the moment of t in the j-dimensional searching space;

The specific steps of the step (3) are as follows:

for any individual x _ij, its inverse solution is defined as

x′_ij＝ub_j+lb_j-x_ij (18)

In the above formula, ub _j and lb _j are respectively the upper limit and the lower limit of the j-th dimension search space, in order to further enhance crowd diversity, it is proposed to update crowd positions by adopting differential evolution, and high-quality individuals are searched through mutation, intersection and selection strategies, so that information exchange among the individuals is enhanced;

For each individual vector x _i in the population, three different individual vectors are randomly selected for combination, resulting in variant individuals:

v_i(t+1)＝x^r1(t)+F(x^r2(t)-x^r3(t)) (19)

In the above formula, r ₁,r₂,r₃ is the number of three different differential individuals in [1,0.5N ], wherein the scaling factor F is a random number in the range of [0,1 ];

Test subjects u _ij (t+1) were constructed using a crossover operation, the construction method being:

Wherein, rand generates random numbers between [0,1], the range of the crossover probability factor CR is between [0,1], f (x) is an fitness function, the fitness value of a new individual after crossover operation is calculated, if the new individual has better fitness, the original individual is replaced:

The global optimal position is adaptively updated by using a variation factor taking the iteration number i as a system parameter, and a position with higher quality is selected based on a greedy selection algorithm and participates in next iteration; the self-adaptive t distribution is a random parameter set integrating the advantages of Gaussian distribution and Cauchy distribution, and when the random parameter set is used as a variation factor to disturb a feasible solution, the algorithm can have certain local random searching capacity so as to avoid sinking into local optimum; the adaptive mutation process of the optimal individual is shown in the following formula:

x_best(t+1)＝x_best(t)+x_best(t)*trnd(t) (22)

In the above formula, trnd is a self-adaptive t distribution parametric function, and can be called by a matlab function library; x _best (t) is the current optimal solution, and x _best (t+1) is the new solution after the adaptive t distribution variation.

2. The method for optimizing the main steam pressure of the PID control thermal power depth peak shaver set by the improved crowd searching algorithm according to claim 1, wherein the method comprises the following steps:

In the step (1), test modeling is carried out on steady-state operation conditions of the deep peak-shaving thermal power generating unit, the fuel quantity fed into the furnace is changed in a step mode, a main steam pressure response curve and data are obtained, and a main steam pressure transfer function model is obtained through identification by a least square method, wherein the main steam pressure transfer function model is shown in the following formula:

In the formula: k-static gain of the main steam pressure transfer function; n-transfer function order; τ -delay time; s-Laplace operator.

3. The method for optimizing PID control over main steam pressure of a thermal power depth peak shaver set by an improved crowd search algorithm according to claim 1, wherein the crowd search algorithm SOA is a bionic control algorithm, and is proposed by analyzing random search behaviors of people; the optimal solution to the target is completed by simulating experience gradient and uncertainty reasoning in intelligent searching behaviors of the person; i.e. by describing natural language and modeling uncertainty reasoning in order to determine search steps, while determining search direction by autonomous learning capabilities and group behavior, including Litsea behavior, lithe behavior, self-organizing aggregate behavior, preventive behavior and uncertainty reasoning behavior.

4. The method for optimizing the main steam pressure of the PID control thermal power depth peak shaver set by the improved crowd searching algorithm according to claim 1, wherein the method comprises the following steps: the fitness function value in the step (2) is used for explaining whether an individual or a solution is good or bad in the optimization process of the crowd search algorithm SOA, and meanwhile, provides a reference for the position update of the individual, adopts an error absolute value time integral performance index as a minimum objective function of parameter selection, and adds a square term of control input quantity into the objective function in order to prevent the control quantity from being too large.