1. A method for setting air flow of a fire grate in an urban domestic garbage incineration process is characterized by comprising the following steps:

(1) establishing a set case library according to historical data of the urban life incineration process; a characteristic variable x₁～x₉I.e. primary air heater outlet air temperature x₁Left inner side speed x of drying grate₂Left outer side speed x of drying grate₃Left outer side speed x of drying grate₄And the right outer side speed x of the drying grate₅And the temperature x of the inlet at the left side of the grate of the drying section₆And the temperature x of the left outlet of the grate of the drying section₇And the temperature x of the right inlet of the grate of the drying section₈And the temperature x of the right outlet of the grate of the drying section₉The historical data and the air flow y of the drying grate are normalized and then expressed into a characteristic vector form to form A source cases which are stored in a set case library; record each source case as C_tExpressed in the following form:

C_t＝(X_t；y_t),t＝1,2,…,A (1)

wherein A is the total number of source cases; y is_tIs the t-th source case C_tThe flow value of the air in the middle drying section furnace exhaust; x_tIs a problem description of the tth source case, X_tCan be expressed as:

X_t＝(x_1,t,…,x_λ,t,…,x_9,t) (2)

wherein x is_λ,t(λ ═ 1, …,9) represents C_tThe central lambda characteristic variable value; y is_tRespectively representing the air flow values of the furnace discharge of the drying section;

(2) initializing parameters; characteristic weight upper and lower limits omega of selfish herd simulation-annealing algorithm_dAnd ω_uTotal iteration number itern, current iteration number k and initial temperature T₀Final temperature T_minTemperature control parameter T_SAAnd the temperature decay coefficient τ;

(3) distributing characteristic weight based on selfish herd-simulated annealing algorithm; firstly, before the weight distribution is optimized by using the SHO-SA algorithm, N groups of randomly generated weight objects need to be divided into two groups of 'prey' and 'predator', the size of the prey group is larger than that of the predator, the member set of the prey group is represented by H, and the number of the member set is N_h(ii) a The other part is predator, the member set is represented by P, and the number is N_p(ii) a Secondly, carrying out weight optimization by using a Selfish Herd Optimizer (SHO) algorithm, wherein the SHO algorithm carries out optimization through a motion stage, a predation stage and a recovery stage; the detailed process is as follows:

a. and (3) a motion stage:

according to the structure of the prey, the weight of the prey group is divided into three roles to analyze the motion phase: captain, followers and fleets moving independently of the prey group; head collarRepresenting the weight with the best fitness in the prey group of the kth iteration; the three definitions are as follows:

wherein the content of the first and second substances,representing data h in the kth iteration_iWeight of (2)Representing followers in the k-th iteration process;representing an escaper in the k-th iteration process; rand (0,1) represents a (0,1) random number and does not affect algorithm performance; based on the above description, the movement patterns of the three characters are expressed as follows:

head and neck movement:

wherein the content of the first and second substances,representing the weight corresponding to the optimal fitness of the (k + 1) th iteration; c. C^kAnd s^kFor the motion vector, the calculation is as shown in formula (7) and formula (8); SV is a survival value, expressed as follows:

wherein the content of the first and second substances,the weight with the highest elimination risk in the k iteration is calculated as shown in a formula (10);the rejection factor, representing the weight in the "prey" group versus the weight in the "predator" group, is calculated as shown in equation (11);representing the optimal weight in the k iteration process;the attraction factor of the optimal weight to other weights of the 'prey' group at the kth iteration is represented, and the calculation is shown as the formula (12); alpha and epsilon are random numbers between intervals (0,1) and do not influence algorithm performance; f (Ω (i)) represents the fitness corresponding to the ith group of the N groups of weights, and Ω (i) ═ ω (ω)_i,1,ω_i,2,…,ω_i,j,…,ω_i,J)，ω_i,jRepresents the jth feature weight in Ω (i), J being the total number of features; f. of_bestAnd f_worstRespectively representing the best fitness and the worst fitness found in the SHO algorithm evolution process, and respectively defined as shown in formula (13) and formula (14):

wherein the content of the first and second substances,representing the weight of a predator group in the k iteration process, wherein k represents the current iteration number in the SHO evolution process;

movement of followers and fleets:

wherein the content of the first and second substances,andrepresents the set of followers and escapes, when Ω (h)_i) When identified as a primary or secondary follower, f_i ^kA motion vector representing a follower; when omega (h)_i) When identified as an evacuee,a motion vector representing an escaper;f_i ^kandis represented as follows:

wherein the content of the first and second substances,as weights in the k-th iterationFor the weightThe attraction factor of (c);as weights in the k-th iterationFor the weightThe attraction factor of (c);is the weight omega in the k iteration process_bestFor the weightThe attraction factor of (c);represents the weight Ω (h) in the k-th iteration_i) A survival value of (c); chi, kappa andis a random number between intervals (0,1) and does not influence the performance of the algorithm;representing relative weightsIs preferably weighted by the weight ofThe survival value of (2) is larger, and the calculation formula is as follows:

wherein r is_i,jRepresenting weights during the kth iterationAnd weightEuclidean distance between; omega (H)^k) All weights representing the "prey" group during the kth iteration;

the predator's movement depends only on the individual distance to the prey, and the predator usually attacks individuals that are closer, and the formula for the movement can be expressed as:

wherein the content of the first and second substances,weight of predator group at kth iteration, I ∈ (1, N)_p) (ii) a ρ represents a value in the interval [0,1 ]]A random number within;selecting the weight in the prey group by adopting a roulette mode; r ∈ (1, 2.,. N.)_h)，N_hNumber of prey group weights;

b. a predation stage:

in the predation phase, the eliminated weight population Ω (K) is defined as follows:

wherein N is_kRepresents the total number of weights, Ω (h), rejected during the predation phase_n) And Ω (h)_w) The eliminated weight individuals are selected;

c. and (3) a recovery stage:

in order to ensure that the weight scale of each iteration is the same, the weight scale is restored by roulette, and the restoration weight is defined as follows:

wherein the content of the first and second substances,representing candidate weightsA characteristic element of (a);

after the recovery stage, recovering the population scale to obtain a new iteration weight; obtaining the weight corresponding to the minimum RMSE in the population through the SHO algorithm, and combining the weight and the weightThe fitness is recorded as omega_bestAnd f (omega)_best) While being stored at omega_BestAnd f (omega)_Best) In (1), as follows:

Ω_Best＝Ω_best (22)

f(Ω_Best)＝f(Ω_best) (23)

then, continuously optimizing the weight by adopting a Simulated Annealing (SA) algorithm; the detailed process is as follows:

will omega_bestAs an initial value of the simulated annealing algorithm, the weight Ω_bestA random search is performed nearby, resulting in a new weight Ω_newAnd calculating its fitness f (omega)_new) Then adopting Metropolis criterion to receive the optimal weight; when the fitness of the new weight is better than omega_BestIs directly accepted, otherwise a new weight, namely zeta, is accepted through probability xi<Xi is accepted, otherwise, the xi is not accepted, wherein the calculation formula of xi is shown as a formula (25); if the new weight is accepted, then the new solution Ω_newAnd its fitness f (omega)_new) Imparting omega_BestAnd f (omega)_Best) And updating the annealing temperature to the lowest temperature T_minThe update temperature equation and Metropolis criteria are expressed as follows:

T_n+1＝τ·T_n (24)

wherein, T_nThe annealing temperature of the nth annealing; t is_n+1The annealing temperature is the annealing temperature of the (n + 1) th annealing; τ represents the temperature decay rate, generally taking a value between 0.8 and 0.99, here 0.95; (ii) a ζ is the interval [0,1 ] generated by rand]A random number within; omega_newAs a new weight, Ω_BestThe weight corresponding to the current optimal fitness; t is_SAIs a temperature control parameter, and T_SA＝T_n；

Finally, when the temperature in the simulated annealing algorithm reaches the lowest temperature, indicating that the SHO-SA one-time iteration optimization is finished; selecting the optimal weight obtained by the iteration to enter next iteration optimization until the iteration times are reached, thereby obtaining an approximate solution of the optimal weight;

(4) acquiring a dry section grate target case from a test case library, carrying out normalization processing, calculating the similarity between the dry section grate target case and a source case through a case retrieval model, and retrieving a similar case; (ii) a Wherein, the similarity measurement method based on Euclidean distance is adopted to calculate the current data (x)_1,A+1,…,x_λ,A+1,…,x_9,A+1) And source case C_tS similarity value of_t：

(5) Obtaining an average value of M similar case solutions through case reuse, so as to obtain a value of a set value M of the target case solution, wherein the value of the set value M is 3;

(6) forming a case by the target case and the set value, storing the case into a set database, and solving the next suboptimal setting;

(7) and (5) repeating the steps (3) to (6) to realize the optimal setting process of the air flow of each grate in the incineration process of the urban domestic garbage.

Background

In the process of burning garbage, the control result of the flue gas temperature of a combustion chamber of the urban domestic garbage incinerator is one of key indexes for judging whether the incineration process is normally operated, and particularly the temperature of a hearth is controlled to be 850-950 ℃. The stable and accurate furnace temperature is beneficial to the full drying of the garbage and the separation of volatile matters, and promotes the burnout of residual carbon, thereby improving the combustion degree of the garbage, reducing the emission of harmful pollutants (such as dioxin and the like) and more effectively realizing the resource utilization of the garbage. However, the operation variables of the flue gas temperature of the first combustion chamber are more, the coupling property is provided, the decision of the controller is not facilitated, and according to the mechanism analysis, the logic relationship which can be determined is that the air flow of the partition fire grate is set according to the speed of each section of fire grate, the inlet air temperature and the air pressure condition, and the control is realized by setting the air flow through whether the combustion condition reaches the standard or not. Therefore, the realization of the optimized setting of the air flow of the furnace grate in the incineration process has important practical significance.

At present, the research on the set model of the air flow of the fire grate of the urban domestic garbage incinerator mainly comprises mechanism modeling and data-driven modeling. The urban domestic garbage incineration process has comprehensive complexity of numerous interference, serious adjustable parameter coupling and the like, so that the mechanism modeling is difficult to realize. However, the modeling method based on data driving realizes the prediction of the target by collecting data which is related to the target variable and is easy to monitor on line by means of an artificial intelligence technology, and becomes a research hotspot in the set field. At present, the modeling method for engineering mainly includes a neural network, a support vector machine and the like. The modeling method based on the neural network usually needs a large number of samples with enough representativeness, and is easy to fall into local optimization, and the training result is unstable. In contrast, the support vector machine method can avoid possible over-learning and local optimization, has global optimization capability and generalization capability better than that of a neural network, but is applicable to small samples and lacks self-learning capability. Therefore, the setting methods are not effective.

Case reasoning, which has been widely used over the last forty years, has a main idea of solving the current problem by evaluating the similarity between the current problem and the source cases stored in the case base, then retrieving the same or similar cases in the case base, and using the solution of the source case(s). The method has the advantages that the case retrieval performance is influenced by the quality of the case retrieval performance, and the distribution result of the feature weight directly influences the case retrieval performance in the case retrieval process. Therefore, starting from the distribution method of the feature weight in the case retrieval process, the model is carried out on the air grate flow rate to realize the optimized setting.

Disclosure of Invention

Aiming at the problems, the invention provides a method for setting the air flow of the fire grate in the process of incinerating the municipal solid waste, which can lead the municipal solid waste incinerator to stably operate by optimally setting the operation variables (the air flow of each fire grate).

In order to achieve the purpose, the invention adopts the following technical scheme:

a method for setting air flow of a fire grate in an urban domestic garbage incineration process is characterized by comprising the following steps: (1) establishing a set case library according to historical data of the incineration process of the municipal solid waste; (2) initializing parameters; (3) distributing characteristic weight based on selfish herd-simulated annealing algorithm (SHO-SA); (4) obtaining solutions of M similar cases through a case retrieval model; (5) and obtaining the average value of the M similar case solutions through case reuse, thereby obtaining the set value of the target case solution. (6) Forming a case by the target case and the set value thereof and storing the case in a set case library; and (7) repeating the steps (3) to (6) to realize the optimal setting process of the grate air flow in the municipal solid waste incineration process. The method further comprises the following steps:

(1) establishing a set case library according to historical data of the urban life incineration process; a characteristic variable x₁～x₉(Primary air Heater Outlet air temperature x₁Drying furnaceRow left inner velocity x₂Left outer side speed x of drying grate₃Left outer side speed x of drying grate₄And the right outer side speed x of the drying grate₅And the temperature x of the inlet at the left side of the grate of the drying section₆And the temperature x of the left outlet of the grate of the drying section₇And the temperature x of the right inlet of the grate of the drying section₈And the temperature x of the right outlet of the grate of the drying section₉) The historical data and the air flow y of the drying grate are expressed into a characteristic vector form after normalization processing to form A source cases which are stored in a set case library. Record each source case as C_tExpressed in the following form:

C_t＝(X_t；y_t),t＝1,2,…,A (1)

wherein A is the total number of source cases; y is_tIs the t-th source case C_tThe air flow value of the grate 1 of the middle drying section; x_tIs a problem description of the tth source case, X_tCan be expressed as:

X_t＝(x_1,t,…,x_λ,t,…,x_9,t) (2)

wherein x is_λ,t(λ ═ 1, …,9) represents C_tThe value of the lambda characteristic variable.

(2) Initializing parameters; make selfish herd model-simulated annealing algorithm characteristic weight upper and lower limits omega_dAnd ω_uTotal iteration number itern, current iteration number k and initial temperature T₀Final temperature T_minTemperature control parameter T_SAAnd a temperature decay coefficient τ.

(3) Distributing characteristic weight based on selfish herd-simulated annealing algorithm; referring to FIG. 2, first, before optimizing the weight distribution using the SHO-SA algorithm, it is necessary to divide the randomly generated N groups of weight objects into two groups of "prey" and "predator", the size of the prey group being larger than that of the predator, the member set of the prey group being represented by H and the number of the member set being N_h(ii) a The other part is predator, the member set is represented by P, and the number is N_p. Secondly, carrying out weight optimization by using a Selfish Herd Optimizer (SHO) algorithm, wherein the SHO algorithm carries out optimization through a motion stage, a predation stage and a recovery stage; detailed description of the inventionThe process is as follows:

a. and (3) a motion stage:

according to the structure of the prey, the weight of the prey group is divided into three roles to analyze the motion phase: headrail, follower, and fleets moving independently of the prey group. Head collarRepresents the weight of best fitness in the prey set of the kth iteration. The three definitions are as follows:

wherein the content of the first and second substances,representing data h in the kth iteration_iThe weight of (a) is determined,representing followers in the k-th iteration process;representing an escaper in the current iterative process; rand (0,1) represents a (0,1) random number and does not affect algorithm performance. Based on the above description, the movement patterns of the three characters are expressed as follows:

head and neck movement:

wherein the content of the first and second substances,weight corresponding to the best fitness representing the (k + 1) th iteration, c^kAnd s^kFor the motion vector, the calculation is as shown in formula (7) and formula (8); SV is a survival value, expressed as follows:

wherein the content of the first and second substances,the weight with the highest elimination risk in the k iteration is calculated as shown in a formula (10);the rejection factor, representing the weight in the "prey" group versus the weight in the "predator" group, is calculated as shown in equation (11);representing the optimal weight in the k iteration process;the attraction factor representing the optimal weight to the other weights of the 'prey' group at the kth iteration is calculated as shown in formula (12); alpha and epsilon are random numbers between intervals (0,1) and do not influence algorithm performance; f (Ω (i)) represents the fitness corresponding to the ith group of the N groups of weights, and Ω (i) ═ ω (ω)_i,1,ω_i,2,…,ω_i,j,…,ω_i,J)，ω_i,jRepresents omega(i) J is the total number of features; f. of_bestAnd f_worstRespectively representing the best fitness and the worst fitness found in the SHO algorithm evolution process, and respectively defined as shown in formula (13) and formula (14):

wherein the content of the first and second substances,represents the weight of the set of "predators" during the kth iteration, and k represents the current number of iterations during the SHO evolution.

Movement of followers and fleets:

wherein the content of the first and second substances,andindicating follower and escapeSet of deaths, when Ω (h)_i) When identified as a primary or secondary follower, f_i ^kA motion vector representing a follower; when omega (h)_i) When identified as an evacuee,representing the motion vector of the evacuee. f. of_i ^kAndis represented as follows:

wherein the content of the first and second substances,as weights in the k-th iterationFor the weightThe attraction factor of (c);as weights in the k-th iterationFor the weightThe attraction factor of (c);is the weight omega in the k iteration process_bestFor the weightThe attraction factor of (c);represents the weight Ω (h) in the k-th iteration_i) A survival value of (c); chi, kappa andis a random number between the regions (0,1) and does not influence the performance of the algorithm;representing relative weightsIs preferably weighted by the weight ofThe survival value of (2) is larger, and the calculation formula is as follows:

wherein r is_i,jRepresenting weights during the kth iterationAnd weightEuclidean distance between; omega (H)^k) All weights of the "prey" group during the kth iteration are indicated.

The predator's movement depends only on the individual distance to the prey, and the predator usually attacks individuals that are closer, and the formula for the movement can be expressed as:

wherein the content of the first and second substances,weight of predator group at kth iteration, I ∈ (1, N)_p) (ii) a ρ represents a value in the interval [0,1 ]]A random number within;selecting the weight in the prey group by adopting a roulette mode; r ∈ (1, 2.,. N.)_h)，N_hRepresenting the number of prey group weights.

b. Predation phase

In the predation phase, the elimination weight is defined as follows:

wherein N is_kRepresents the total number of weights, Ω (h), rejected during the predation phase_n) And Ω (h)_w) The weight individuals are eliminated.

c. Recovery phase

In order to ensure that the weight scale of each iteration is the same, the weight scale is restored by roulette, and the restoration weight is defined as follows:

wherein the content of the first and second substances,representing candidate weightsThe characteristic elements of (1).

After the recovery phase, the population size is recovered to obtain a new iteration weight. Obtaining the weight corresponding to the minimum RMSE in the population through an SHO algorithm, and adding the weightsThe fitness of the weight is recorded as Ω_bestAnd f (omega)_best) While being stored at omega_BestAnd f (omega)_Best) In (1), as follows:

Ω_Best＝Ω_best (22)

f(Ω_Best)＝f(Ω_best) (23)

then, the above weights are further optimized by using a Simulated Annealing (SA) algorithm. The detailed process is as follows:

will omega_bestAs an initial value of the simulated annealing algorithm, the weight Ω_bestA random search is performed nearby, resulting in a new weight Ω_newAnd calculating its fitness f (omega)_new) And then adopting Metropolis criterion to receive the optimal weight. When the fitness of the new weight is better than omega_BestIs directly accepted, otherwise a new weight, namely zeta, is accepted through probability xi<Xi is accepted, otherwise not, where xi is calculated as shown in equation (25). If the new weight is accepted, then the new solution Ω_newAnd its fitness f (omega)_new) Is given omega_BestAnd f (omega)_Best) And updating the annealing temperature to the lowest temperature T_minThe update temperature equation and Metropolis criteria are expressed as follows:

T_n+1＝τ·T_n (24)

wherein, T_nThe annealing temperature of the nth annealing; t is_n+1The annealing temperature is the annealing temperature of the (n + 1) th annealing; τ represents the temperature decay rate, and is typically taken to be a value between 0.8 and 0.99, and is taken to be 0.95 herein. (ii) a ζ is the interval [0,1 ] generated by rand]A random number within; omega_newAs a new weight, Ω_BestThe weight corresponding to the current optimal fitness; t is_SAIs a temperature control parameter, and T_SA＝T_n。

And finally, when the temperature in the simulated annealing algorithm reaches the lowest temperature, indicating that the once iteration optimization of the SHO-SA algorithm is finished. And selecting the optimal weight obtained by the iteration and entering the next iteration optimization until the number of iterations is reached, thereby obtaining an approximate solution of the optimal weight.

(4) Acquiring a dry section grate target case from a test case library, carrying out normalization processing, calculating the similarity between the dry section grate target case and a source case through a case retrieval model, and retrieving a similar case; (ii) a Wherein, the method of similarity measurement based on Euclidean distance is adopted to calculate the current data (x)_1,A+1,…,x_λ,A+1,…,x_9,A+1) And source case C_tS similarity value of_t：

(5) Obtaining an average value of M similar case solutions through case reuse, thereby obtaining a set value of a target case solution;

(6) and forming a case by the target case and the set value, storing the case into a set database, and solving the next suboptimal setting.

(7) And (5) repeating the steps (3) to (6) to realize the optimal setting process of the grate air flow in the municipal solid waste incineration process.

Compared with the prior art, the invention has the following advantages: 1. the invention utilizes the historical data generated in the waste incineration process, establishes an optimized setting model by adopting a case reasoning method, has short required time and is beneficial to real-time application; 2. the subjectivity of optimization setting by experts through inspection is avoided; 3. the adoption of the characteristic weight optimal distribution method based on the selfish herd-simulated annealing algorithm is beneficial to avoiding the difficulties of weight distribution and distance traps, so that the set value of the air flow of the grate meets the requirement of operation optimization in the incineration process.

Drawings

FIG. 1 is a schematic diagram of a method for setting air flow of a grate in an incineration process of municipal solid waste according to the present invention;

FIG. 2 is a diagram of the structure of the SHO-SA algorithm

Detailed Description

The sample data is 1000 data generated in the combustion process of a waste incineration plant, and is randomly divided into 800 source cases and 200 test cases, and the specific implementation of the invention is further explained with reference to fig. 1.

A method for setting air flow of a fire grate in an urban domestic garbage incineration process is characterized by comprising the following steps:

(1) establishing a set case library according to historical data of the incineration process; the detailed process is as follows:

9 characteristic variables x₁～x₉(Primary air Heater Outlet air temperature x₁Left inner side speed x of grate of drying section₂Left outer side velocity x of fire grate of drying section₃Left outer side speed x of grate of drying section₄And the right outer side speed x of the fire grate of the drying section₅And the temperature x of the inlet at the left side of the grate of the drying section₆And the temperature x of the left outlet of the grate of the drying section₇And the temperature x of the right inlet of the grate of the drying section₈And the outlet temperature x of the right side of the grate of the drying section₉) The historical data and the corresponding set attribute, namely the air flow y of the grate of the drying section are normalized and then expressed into a characteristic vector form to form 800 source cases which are stored in a decision case library. Record each source case as C_tAnd can be expressed as follows:

C_t＝(X_t；y_t),t＝1,2,…,800 (1)

where 800 is the total number of source cases; y is_tIs the t-th source case C_tThe air flow value of the grate of the drying section is adjusted; x_tIs a description of the problem of the source case of item t, X_tCan be expressed as:

X_t＝(x_1,t,...,x_λ,t,...,x_9,t) (2)

wherein x is_λ,t(λ ═ 1, …,9) represents C_tThe central lambda characteristic variable value; y is_tA dry section grate air flow value;

(2) initializing parameters; according to the SHO algorithm principle, enabling upper and lower limits of SHO-SA algorithm characteristic weight to be 1 and 0; stackThe generation times are determined according to the experimental effect, and the total iteration times in the text is 100; according to the SA algorithm principle, the initial temperature is 100 ℃, the final temperature is 0.001 ℃, and the temperature control parameter T_SAIs equal to T_nAnd a temperature decay coefficient of 0.95.

(3) Distributing characteristic weight based on SHO-SA algorithm; first, before using SHO-SA algorithm to optimize the distribution of weights, 100 randomly generated weight objects need to be divided into two groups of 'prey' and 'predator', the size of prey group is larger than that of predator, the member set of prey group is represented by H, and the number is N_hAccounting for 70-90% of the total amount; the other part is predator, the member set is represented by P, and the number is N_p. Secondly, carrying out weight optimization by using a Selfish Herd Optimizer (SHO) algorithm, wherein the SHO algorithm carries out optimization through a motion stage, a predation stage and a recovery stage; the detailed process is as follows:

a motion phase

According to the structure of the prey, the weight of the prey group is divided into three roles to analyze the motion phase: headrail, follower, and fleets moving independently of the prey group. Head collarRepresents the weight of best fitness in the k iteration prey set. The three definitions are as follows:

wherein the content of the first and second substances,representing data h in the kth iteration_iThe weight of (a) is determined,representing followers in the k-th iteration process;representing an escaper in the current iterative process; rand (0,1) represents a (0,1) random number and does not affect algorithm performance. Based on the above description, the movement patterns of the three characters are expressed as follows:

head and neck movement:

wherein the content of the first and second substances,represents the weight corresponding to the best fitness of the (k + 1) th iteration, c^kAnd s^kFor the motion vector, SV is calculated as a survival value as shown in equations (7) and (8), and is expressed as follows:

wherein the content of the first and second substances,the weight with the highest elimination risk at the k-th iteration is calculated as shown in formula (10),the rejection factor, representing the weight in the "prey" group versus the weight in the "predator" group, is calculated as shown in equation (11);representing the optimal weight in the k iteration process;the attractive factors of the optimal weight to other weights of the 'prey' group at the kth iteration are represented, and the calculation is shown as a formula (12); alpha and epsilon are random numbers between intervals (0,1) and do not influence algorithm performance; f (Ω (i)) represents the fitness corresponding to the ith group among 100 groups of weights, and Ω (i) ═ ω (ω)_i,1,ω_i,2,…,ω_i,j,…,ω_i,9)，ω_i,jRepresents the jth feature weight in Ω (i); f. of_bestAnd f_worstRespectively representing the best fitness and the worst fitness discovered in the SHO algorithm evolution process, and respectively defined as follows:

wherein the content of the first and second substances,represents the weight of the set of "predators" during the kth iteration, and k represents the current number of iterations during the SHO evolution.

Movement of followers and fleets:

wherein the content of the first and second substances,andrepresents the set of followers and escapes, when Ω (h)_i) When identified as a primary or secondary follower, f_i ^kA motion vector representing a follower; when omega (h)_i) When identified as an evacuee,representing the motion vector of the evacuee. f. of_i ^kAndis represented as follows:

wherein the content of the first and second substances,as weights in the k-th iterationFor the weightThe attraction factor of (c);as weights in the k-th iterationFor the weightThe attraction factor of (c);is the weight omega in the k iteration process_bestFor the weightThe attraction factor of (c);represents the weight Ω (h) in the k-th iteration_i) A survival value of (c); chi, kappa andis a random number between the regions (0,1) and does not influence the performance of the algorithm;representing relative weightsIs preferably weighted by the weight ofThe survival value of (2) is larger, and the calculation formula is as follows:

wherein r is_i,jRepresenting weights during the kth iterationAnd weightEuclidean distance between; omega (H)^k) All weights of the "prey" group during the kth iteration are indicated.

The predator's movement depends only on the individual distance to the prey, and the predator usually attacks individuals that are closer, and the formula for the movement can be expressed as:

wherein the content of the first and second substances,weights representing the kth iteration predator group, I ∈ (1, N)_p) (ii) a ρ represents a value in the interval [0,1 ]]A random number within;selecting the weight in the prey group by adopting a roulette mode; r ∈ (1, 2.,. N.)_h)，N_hThe number of prey group weights is indicated.

b predation phase

In the predation phase, the eliminated weight population Ω (K) is defined as follows:

wherein N is_kRepresents the total number of weights, Ω (h), rejected during the predation phase_n) And Ω (h)_w) The weight individuals are eliminated.

c recovery phase

In order to ensure that the weight scale of each iteration is the same, the weight scale is restored by roulette, and the restoration weight is defined as follows:

wherein the content of the first and second substances,representing candidate weightsThe characteristic elements of (1).

After the recovery phase, the population size is recovered to obtain a new iteration weight. Obtaining the weight corresponding to the minimum RMSE in the population through an SHO algorithm, and respectively marking the weight and the fitness of the weight as omega_bestAnd f (omega)_best) While being stored at omega_BestAnd f (omega)_Best) In (1), as follows:

Ω_Best＝Ω_best (22)

f(Ω_Best)＝f(Ω_best) (23)

then, the above weights are further optimized by using a Simulated Annealing (SA) algorithm. The detailed process is as follows:

will omega_bestAs an initial value of the simulated annealing algorithm, the weight Ω_bestA random search is performed nearby, resulting in a new weight Ω_newAnd calculating its fitness f (omega)_new) And then adopting Metropolis criterion to receive the optimal weight. When the fitness of the new weight is better than omega_BestIs directly accepted, otherwise a new weight, namely zeta, is accepted through probability xi<Xi is accepted, otherwise not, where xi is calculated as shown in equation (25). . If the new weight is accepted, then the new solution Ω_newAnd its fitness f (omega)_new) Is given omega_BestAnd f (omega)_Best) And updating the annealing temperature to the lowest temperature T_minThe update temperature equation and Metropolis criteria are expressed as follows:

T_n+1＝τ·T_n (24)

wherein, T_nThe annealing temperature of the nth annealing; t is_n+1The annealing temperature is the annealing temperature of the (n + 1) th annealing; τ represents the temperature decay rate and is generally taken to be a value between 0.8 and 0.99, here 0.95. (ii) a ζ is the interval [0,1 ] generated by rand]A random number within; omega_newAs a new weight, Ω_BestThe weight corresponding to the current optimal fitness; t is_SAIs a temperature control parameter, and T_SA＝T_n。

And finally, when the temperature in the simulated annealing algorithm reaches 0.001 ℃, the SHO-SA algorithm is indicated to finish one-time iteration optimization. Selecting the optimal weight obtained by the iteration to enter the next iteration optimization until the iteration reaches 100 times, thereby obtaining an approximate solution of the optimal weight;

(4) acquiring a dry section grate target case from a test case library, carrying out normalization processing, calculating the similarity between the dry section grate target case and a source case through a case retrieval model, and retrieving a similar case; wherein, a similarity measurement method of Euclidean distance is adopted to calculate a similarity value S between current data of a grate at a dry section and a source case_t：

(5) Obtaining an average value of solutions of the previous M similar cases with large similarity values through case reuse, thereby obtaining a set value of a dry section air flow solution, wherein the value of M is 3 according to universality and accuracy;

(6) forming a case by the target case and the set value, storing the case into a set database, and solving the next suboptimal setting;

(7) repeating the steps (3) to (6) to realize the optimal setting process of the air flow of each grate in the incineration process of the municipal domestic waste;

the optimization control of the conventional waste incineration process is completely performed by operators according to experience, and once the operation is improper, the operation of the incineration process is influenced, so that great economic loss is caused. The invention combines a characteristic weight distribution method based on an SHO-SA algorithm with a case reasoning technology to realize the real-time setting of the air flow rate of the grate in the process of incinerating the urban domestic garbage, and in order to further verify the effect of the case reasoning (abbreviated as CBRSHO-SA) algorithm for distributing the weight by adopting the SHO-SA method to set the air flow rate of the grate in the process of incinerating the garbage, historical data is used in the comparison experiments of the method and other prediction methods, and the setting effect of the CBRSHO-SA can be seen to be more advantageous from the setting average fitting error of the air flow rate of a dry section, so the method has certain application value.