Primary frequency regulation control method and device based on intelligent solution of hydraulic turbine simulation
By using the Gray Wolf Optimization Algorithm and BP neural network to form a simulation model in the turbine speed regulation system, the shortcomings of the PID algorithm in nonlinear and time-varying systems are solved, more efficient frequency and power coordination is achieved, and the system's reliance is improved Rodability and control accuracy.
Patent Information
- Application Number
- CN202111566236.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-20
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-12-20
AI Technical Summary
In the existing turbine speed regulation system, PID algorithms are difficult to effectively control nonlinear and time-varying systems, which makes it difficult to establish a model and lacks robustness, which cannot meet the requirements of primary frequency regulation.
Using a method based on turbine simulation intelligent solution, the simulation model is formed using the gray wolf optimization algorithm and the BP neural network, and the opening control is optimized. Combined with the social hierarchical stratification of the gray wolf optimization algorithm, the mathematical model of surrounding prey, hunting and attacking prey, the mathematical model of the gray wolf optimization algorithm is constructed. GWO-BP neural network realizes adaptive control.
It improves the control accuracy and robustness of the turbine speed regulation system, can effectively coordinate frequency and power control under different heads and loads, reduces the role of PID in primary frequency regulation, enhances anti-interference ability, and improves the working efficiency of the speed regulator.
Smart Images

Figure CN114357867B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of speed governor control, and in particular relates to a primary frequency regulation control method and device based on intelligent solution of turbine simulation. Background Art
[0002] The existing technology, namely the PID algorithm, is difficult to meet the primary frequency regulation requirements, and its robustness is obviously insufficient in the face of the changeable working conditions of the hydropower plant. The existing patent document CN111997825A discloses a turbine speed governor power / frequency control method, the disclosed content is: the frequency deviation signal is converted into a target power signal and superimposed with the power given signal to form an actual target given power, and then compared with the actual power signal to obtain the control deviation for control; in this way, the frequency power and power control method and process are completely consistent; therefore, the coordination of primary frequency regulation (frequency control) and AGC (power control) can be well achieved; in order to achieve better indicators of regulation control performance under different water heads, a reference model control method is proposed; according to the target control power and the current operating water head, the required control output is directly generated by the reference model, which speeds up the control response speed; in order to obtain better control accuracy, the target power is compared with the actual power, and the deviation signal is introduced into the correction controller output to correct the control output; the reference model is corrected by using the correction link output to make the reference model have a self-improvement function;
[0003] The existing technology focuses on the use of PID to perform closed-loop control of the speed control system, and lacks intelligent adaptive control. At present, PID is generally used to control the system through three links: proportional, integral, and differential based on the error between the given value and the actual measured value. The PID controller has a better control effect than a linear system with a simple structure, but for a system like a turbine regulation system that is nonlinear, time-varying, and has a non-minimum phase, the control effect is not so ideal. Summary of the invention
[0004] In view of the technical problems existing in the background technology, the primary frequency regulation control method and device based on intelligent solution of turbine simulation provided by the present invention overcome the shortcomings of the prior art, effectively solve the problems existing in the traditional PID algorithm of turbine primary frequency regulation, and focus on solving the defect that the turbine speed regulation system is difficult to establish a model due to its nonlinearity and time-varying nature.
[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions to achieve:
[0006] A primary frequency modulation control method and device based on intelligent solution of turbine simulation, the steps are as follows:
[0007] Step 1: Obtain the power setting value P0 from the power setting module;
[0008] Step 2: The target power generation module adds the given power P0 to the target power Pf of the primary frequency modulation to obtain the actual target control power CP=P0+Pf;
[0009] Step 3: The water head measurement module measures the current working water head H;
[0010] Step 4: Use the Grey Wolf optimization algorithm to find the optimal opening under the same working conditions, and the optimization goal is to minimize the PID output value;
[0011] Step 5: After obtaining the optimal opening under the same working conditions, a simulation model is formed through the GWO-BP neural network.
[0012] In the preferred solution, in step 4, a mathematical model is constructed based on the social hierarchy of gray wolves and the hunting process; the mathematical model of the GWO algorithm includes the following parts: social hierarchy stratification, surrounding prey, hunting, attacking prey and searching prey, and the model is constructed as follows:
[0013] 1) Social stratification:
[0014] In the GWO algorithm, first, a hierarchical model is constructed according to the social hierarchy of the gray wolf population, and the fitness value of each gray wolf in the population is calculated, and the fitness values are sorted from large to small. Then, the three gray wolves with the best fitness are marked as α, β and δ, and the remaining gray wolves are marked as ω. The ω gray wolf completes the position update around α, β and δ. The optimization process of the GWO algorithm is mainly guided by the three best individuals in each generation of the population, namely α, β and δ.
[0015] 2) Surround the prey:
[0016] The mathematical model of the gray wolf population's behavior of searching for and approaching prey during hunting and finally surrounding the prey is as follows:
[0017] D=C·X p (txt)
[0018] X(t+1)=X p (t)-A·D
[0019] A=2a·r1-a
[0020] C=2r2
[0021] Where: t is the current cycle number, A and C are coefficient vectors; X p is the position vector of the prey; X(t) represents the current position vector of the gray wolf; during the entire iteration of the algorithm, the convergence factor a will decrease linearly from 2 to 0; r1 and r2 are random vectors in the interval [0, 1];
[0022] 3) Hunting:
[0023] During the hunting process, the gray wolf population can identify the location of the prey and surround the prey; in this process, α will lead β and δ to guide the wolf pack to surround the prey; the mathematical model of the wolf pack's hunting behavior can be expressed as follows:
[0024] D α =C1·X α -X,D β =C2·X β -X,D δ =C3·X δ -X,
[0025] X1=X α -A1·D α , X2=X β -A2·D β , X3=X δ -A3·D δ ,
[0026]
[0027] Where: X α , X β , X δ Respectively represent the position vectors of α, β, and δ in the current population; X represents the current position vector of the ω gray wolf; D α , D β , D δ They represent the distances between the current candidate gray wolf and the top three gray wolves with the best fitness; C1, C2, and C3 are random vectors; X(t+1) is the position of the final prey determined based on X1, X2, and X3;
[0028] 4) Attacking prey
[0029] The last step of the gray wolf hunting is to attack the prey. In order to simulate this process, the adjustment of the individual behavior of the gray wolf is achieved by controlling the value of parameter a. a is gradually reduced, so the fluctuation range of A is also reduced. In the process of algorithm iteration, when the value of a decreases linearly from 2 to 0, the corresponding A value changes in the interval [-a, a]. The value of a is adjusted according to the following formula:
[0030]
[0031] When the value of A is within the interval, the position of the gray wolf at the next moment can be anywhere between the current gray wolf and the prey; when |A| < 1, the wolf pack will attack the prey;
[0032] 5) Finding prey:
[0033] The gray wolf population mainly relies on the positions of α, β, and δ to search and locate prey; in the process of searching for prey, the wolf pack first searches for the location information of the prey in a dispersed manner, and then gathers together to attack the prey; for the establishment of the dispersed model, |A|>1 is used to make the gray wolves disperse as much as possible in various areas and search for prey, so that GWO can perform a global search and thus find the best prey; another search coefficient in the GWO algorithm is C; C is a random vector composed of the interval [0,2], which represents the random weight of the influence of the gray wolf's location on the prey. C>1 means that the influence weight is large, otherwise, it means that the influence weight is small; this mechanism helps to improve the randomness of GWO's algorithm search during the optimization process, thereby avoiding the algorithm from falling into the local optimum; C decreases nonlinearly; during the entire iteration process, C adjusts the global search of the algorithm, and the randomness of C can effectively avoid the algorithm from stagnating in search and falling into the local optimum;
[0034] The optimization process of the Gray Wolf Optimization Algorithm is as follows:
[0035] In the process of GWO algorithm optimization, a random population is first generated; then, in the process of algorithm iteration, the three best gray wolves α, β and δ are used to determine the possible location of the prey, and the positions of other gray wolf individuals are updated according to the distance between α, β and δ and the prey; at the same time, the exploration and development process of the algorithm is balanced by controlling the parameter a; when |A|>1, the algorithm performs global exploration; when |A|<1, the algorithm performs local development.
[0036] In a preferred embodiment, step 5 includes the following steps:
[0037] 1) Sample collection: Collect opening data under various working conditions and water heads as the sample set of the simulation model;
[0038] 2) Feature set selection: select common feature sets and optimal hybrid feature sets as model inputs respectively;
[0039] 3) Sample grouping: Divide the existing samples into two groups, namely the training group and the test group. In the process of simulating the model, first use the data of the training group to train the model, and then use the data of the test group to verify the performance of the model.
[0040] 4) Sample normalization: Through the normalization method, all sample data values are adjusted to the range of [0,1], making the model calculation faster. The normalization conversion function is shown as follows:
[0041]
[0042] 5) Model construction: The steps of the simulation model based on GWO-BP neural network are as follows:
[0043] Set the initial parameters of the model. The initialization parameters mainly include population size, maximum number of iterations, variable dimension, scaling factor and crossover probability coefficient CR;
[0044] Initialize the population according to the formula in step 4). In order to ensure that the model error of the BP neural network is minimized after training, the average value of the absolute error between the actual output and the expected output of the BP neural network is selected as the objective function f. The optimization goal is to select the best BP neural network parameter value.
[0045]
[0046] Calculate the fitness of all gray wolf individuals in the population and arrange the fitness values in descending order. The top three gray wolf individuals in terms of fitness are defined as X α , X β , X δ ;
[0047] Update the position of the parent population individuals; Update the parent gray wolf individual P α , P β , P δ , and then sort the parent gray wolf individuals according to the size of the fitness; at the same time, the termination condition of the algorithm iteration is judged, and when the iteration termination condition is met, the optimal individual P of the parent generation is returned. α and f(P α ) is the optimal network parameter corresponding to the model to ensure the minimum PID output; the GWO-BP neural network simulation model of the optimal solution is completed.
[0048] In a preferred solution, a primary frequency regulation feedforward device based on intelligent solution of water turbine simulation includes a primary frequency regulation control method based on intelligent solution of water turbine simulation.
[0049] This patent can achieve the following beneficial effects:
[0050] 1. The present invention overcomes the shortcomings of the prior art, effectively solves the problems existing in the traditional PID algorithm of the primary frequency regulation of the turbine, focuses on solving the defect that the turbine speed regulation system is difficult to establish a model due to its nonlinearity and time-varying nature, and performs opening control by optimizing the simulation model to reduce the role of PID in the primary frequency regulation, with the purpose of having good anti-interference ability. The primary frequency regulation feedforward device based on the intelligent solution of turbine simulation adopts the common advantages of the Grey Wolf optimization algorithm and the BP neural network, and can thus well control a complex system such as the turbine speed regulation system.
[0051] 2. Convert the frequency deviation signal into a target power signal and superimpose it with the power given signal to form the actual target given power, which is then compared with the actual power signal to obtain the control deviation for control; in this way, the frequency power and power control methods and processes are completely consistent; therefore, the coordination between primary frequency modulation (frequency control) and AGC (power control) can be well achieved; in order to achieve better indicators of regulation and control performance under different water heads, a reference model control method is proposed; according to the target control power and the current operating water head, the required control output is generated by the neural network algorithm, which speeds up the control response speed; in order to obtain better control accuracy, the target power is compared with the actual power, and the deviation signal is introduced into the correction controller to output the correction control output.
[0052] 3. The use of the Grey Wolf optimization algorithm model and the neural network algorithm model can obtain more accurate control parameters and realize an adaptive control system to adapt to the nonlinearity, variability, and multi-operating conditions of the turbine regulation system.
[0053] 4. The adaptability of the adjustment parameters is enhanced. The same set of control parameters can be used under different heads and loads, avoiding the frequent switching of adjustment parameters under different working conditions and the possible impact and disturbance caused by the switching.
[0054] 5. It has the function of model self-correction, better dynamic and static characteristics; the parameters are easy to adjust; it can improve the working efficiency of the speed regulator. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:
[0056] Figure 1 This is a flow chart of the control method of the present invention;
[0057] Figure 2 This is a schematic diagram of the position update of the GWO algorithm of the present invention;
[0058] Figure 3 It is an installation diagram of the primary frequency modulation feedforward device based on the intelligent solution of the hydraulic turbine simulation of the present invention;
[0059] Figure 4 This is the schematic diagram of the existing turbine speed governor PID control. DETAILED DESCRIPTION
[0060] The preferred solution is Figure 1 to Figure 2 As shown, a primary frequency modulation control method and device based on intelligent solution of turbine simulation includes the following steps:
[0061] Step 1: Obtain the power setting value P0 from the power setting module;
[0062] Step 2: The target power generation module adds the given power P0 to the target power Pf of the primary frequency modulation to obtain the actual target control power CP=P0+Pf;
[0063] Step 3: The water head measurement module measures the current working water head H;
[0064] Step 4: Use the Gray Wolf optimization algorithm to find the optimal opening under the same working conditions. The optimization goal is to minimize the PID output value; specifically:
[0065] According to the social hierarchy of gray wolves and the hunting process, a mathematical model is constructed; the mathematical model of the GWO algorithm includes the following parts: social hierarchy, surrounding prey, hunting, attacking prey and searching prey. The model can be constructed as follows:
[0066] 1) Social stratification:
[0067] In the GWO algorithm, first, a hierarchical model is constructed according to the social hierarchy of the gray wolf population, and the fitness value of each gray wolf in the population is calculated, and the fitness values are sorted from large to small. Then, the three gray wolves with the best fitness are marked as α, β and δ, and the remaining gray wolves are marked as ω. The ω gray wolf completes the position update around α, β and δ. The optimization process of the GWO algorithm is mainly guided by the three best individuals in each generation of the population, namely α, β and δ.
[0068] 2) Surround the prey:
[0069] The mathematical model of the gray wolf population's behavior of searching for and approaching prey during hunting and finally surrounding the prey is as follows:
[0070] D=C·X p (txt)
[0071] X(t+1)=X p (t)-A·D
[0072] A=2a·r1-a
[0073] C=2r2
[0074] Where: t is the current cycle number, A and C are coefficient vectors; X p is the position vector of the prey; X(t) represents the current position vector of the gray wolf; during the entire iteration of the algorithm, the convergence factor a will decrease linearly from 2 to 0; r1 and r2 are random vectors in the interval [0, 1];
[0075] 3) Hunting:
[0076] During the hunting process, the gray wolf population can identify the location of the prey and surround the prey; in this process, α will lead β and δ to guide the wolf pack to implement the prey encirclement; in order to simulate the hunting behavior of gray wolves, it is assumed that α, β, and δ wolves have a strong ability to identify the location of potential prey; therefore, during the iteration of the algorithm, the three best gray wolves (α, β, δ) in the current population are retained, and then the positions of these three are used to determine the location of the prey, and based on their positions, the position information of other gray wolf individuals (including ω) is updated, and the prey is gradually approached; the mechanism of gray wolf population individuals tracking the location of prey is as follows Figure 2 As shown;
[0077] The mathematical model of wolf pack hunting behavior can be expressed as follows:
[0078] D α =C1·X α -X,D β =C2·X β -X,D δ =C3·X δ -X,
[0079] X1=X α -A1·D α , X2=X β -A2·D β , X3=X δ -A3·D δ ,
[0080]
[0081] Where: X α , X β , X δ Respectively represent the position vectors of α, β, and δ in the current population; X represents the current position vector of the ω gray wolf; D α , D β , D δ They represent the distances between the current candidate gray wolf and the top three gray wolves with the best fitness; C1, C2, and C3 are random vectors; X(t+1) is the position of the final prey determined based on X1, X2, and X3;
[0082] 4) Attacking prey
[0083] The last step of the gray wolf hunting is to attack the prey. In order to simulate this process, the adjustment of the individual behavior of the gray wolf is achieved by controlling the value of parameter a. a is gradually reduced, so the fluctuation range of A is also reduced. In other words, during the algorithm iteration, when the value of a decreases linearly from 2 to 0, the corresponding A value changes in the interval [-a, a]. The value of a is adjusted according to the following formula:
[0084]
[0085] When the value of A is within the interval, the position of the gray wolf at the next moment can be anywhere between the current gray wolf and the prey; when |A| < 1, the wolf pack will attack the prey;
[0086] 5) Finding prey:
[0087] The gray wolf population mainly relies on the positions of α, β, and δ to search and locate prey; in the process of searching for prey, the wolf pack first searches for the location information of the prey in a dispersed manner, and then gathers together to attack the prey; for the establishment of the dispersed model, |A|>1 is used to make the gray wolves disperse as much as possible in various areas and search for prey, so that GWO can perform a global search and thus find the best prey (global optimum); another search coefficient in the GWO algorithm is C; C is a random vector composed of the interval [0,2], which represents the random weight of the influence of the gray wolf's location on the prey. C>1 means that the influence weight is large, otherwise, it means that the influence weight is small; this mechanism helps to improve the randomness of GWO's algorithm search during the optimization process, thereby avoiding the algorithm from falling into the local optimum; in addition, unlike A, C decreases nonlinearly; in this way, during the entire iteration process, it adjusts the global search of the algorithm, and the randomness of C can effectively avoid the algorithm from stagnating in search and falling into the local optimum;
[0088] The optimization process of the Gray Wolf Optimization Algorithm is as follows:
[0089] In the process of GWO algorithm optimization, a random population is first generated; then, in the process of algorithm iteration, the three best gray wolves α, β and δ determine the possible location of the prey, and the other gray wolf individuals update their positions according to the distance between α, β and δ and the prey; at the same time, the exploration and development process of the algorithm is balanced by controlling the parameter a; when |A|>1, the algorithm performs global exploration; when |A|<1, the algorithm performs local development;
[0090] Step 5: After obtaining the optimal opening under the same working conditions, a simulation model is formed through the GWO-BP neural network. Specifically:
[0091] 1) Sample collection: Collect opening data under various working conditions and water heads as the sample set of the simulation model;
[0092] 2) Feature set selection: select common feature sets and optimal hybrid feature sets as model inputs respectively;
[0093] 3) Sample grouping: Divide the existing samples into two groups, namely the training group and the test group. In the process of simulating the model, first use the data of the training group to train the model, and then use the data of the test group to verify the performance of the model.
[0094] 4) Sample normalization: Through the normalization method, all sample data values are adjusted to the range of [0,1], making the model calculation faster. The normalization conversion function is shown as follows:
[0095]
[0096] 5) Model construction: The steps of the simulation model based on GWO-BP neural network are as follows:
[0097] Set the initial parameters of the model. The initialization parameters mainly include population size, maximum number of iterations, variable dimension, scaling factor and crossover probability coefficient CR;
[0098] Initialize the population. In order to ensure that the model error of the BP neural network is minimized after training, the average value of the absolute error between the actual output and the expected output of the BP neural network is selected as the objective function f. The optimization goal is to select the best BP neural network parameter value.
[0099]
[0100] Calculate the fitness of all gray wolf individuals in the population and arrange the fitness values in descending order. The top three gray wolf individuals in terms of fitness are defined as X α , X β , X δ ;
[0101] Update the position of the individual in the parent population; introduce the crossover and mutation operations in the differential evolution algorithm to process the gray wolf individuals to generate the offspring population; introduce the selection operation in the differential evolution algorithm to update the parent population and calculate the values of C, A, and a respectively; update the parent gray wolf individual P α , P β , P δ , and then sort the parent gray wolf individuals according to the size of the fitness; at the same time, the termination condition of the algorithm iteration is judged, and when the iteration termination condition is met, the optimal individual P of the parent generation is returned. α and f(P α ) is the optimal network parameter corresponding to the model to ensure the minimum PID output; a GWO-BP neural network simulation model based on the optimal solution is established.
[0102] like Figure 3 As shown, a primary frequency regulation feedforward device based on intelligent solution of water turbine simulation includes the primary frequency regulation control method based on intelligent solution of water turbine simulation in this embodiment.
[0103] like Figure 4As shown in the figure, the existing turbine speed governor basically adopts PID structure or feedforward PID structure for primary frequency regulation control. Feedforward is mainly used to speed up the regulation speed by adopting a fixed rate, that is, a constant feedforward regulation method. The existing technology, namely the PID algorithm, is difficult to meet the primary frequency regulation requirements, and its robustness is obviously insufficient in the face of the variable working conditions of the hydropower plant.
[0104] The device uses the frequency adjustment amount of the primary frequency difference / ep, the unit working water head and the operating opening as the input of the device of the present invention. After being processed by the device, its output is the GV1 guide vane opening primary frequency adjustment given value obtained by intelligent solution. Its intelligent solution process adopts the Gray Wolf optimization algorithm and BP neural network technology based on turbine simulation conditions, which can solve the nonlinearity and time-varying nature of the turbine speed control system, thereby improving the stability of the control system.
[0105] The above-mentioned embodiments are only preferred technical solutions of the present invention and should not be regarded as limitations of the present invention. The protection scope of the present invention should be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims; that is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A primary frequency modulation control method based on intelligent solution of hydraulic turbine simulation, characterized in that The following steps are involved: Step 1: Obtain the power setting value P0 from the power setting module; Step 2: The target power generation module adds the given power P0 to the target power Pf of the primary frequency modulation to obtain the actual target control power CP=P0+Pf; Step 3: The water head measurement module measures the current working water head H; Step 4: Use the Grey Wolf optimization algorithm to find the optimal opening under the same working conditions, and the optimization goal is to minimize the PID output value; Step 5: After obtaining the optimal opening under the same working conditions, a simulation model is formed through the GWO-BP neural network; In step 4, a mathematical model is constructed based on the social hierarchy of gray wolves and the hunting process; the mathematical model of the GWO algorithm includes the following parts: social hierarchy stratification, surrounding prey, hunting, attacking prey and searching prey. The model is constructed as follows: 1) Social stratification: In the GWO algorithm, first, a hierarchical model is constructed according to the social hierarchy of the gray wolf population, and the fitness value of each gray wolf in the population is calculated and sorted from large to small. Then, the three gray wolves with the best fitness are marked as α, β and δ, and the remaining gray wolves are marked as ω. The ω gray wolf completes the position update around α, β and δ. The optimization process of the GWO algorithm is guided by the best three individuals in each generation of the population, namely α, β and δ. 2) Surround the prey: The mathematical model of the gray wolf population's behavior of searching for and approaching prey during hunting and finally surrounding the prey is as follows: D=C·X p (t)-X(t) X(t+1)=X p (t)-A·D A=2α·r1-α C=2r2 Where: t is the current cycle number, A and C are coefficient vectors; X p is the position vector of the prey; X(t) represents the current position vector of the gray wolf; during the entire iteration of the algorithm, the convergence factor a will decrease linearly from 2 to 0; r1 and r2 are random vectors in the interval [0, 1]; 3) Hunting: During the hunting process, the gray wolf population can identify the location of the prey and surround the prey; in this process, α will lead β and δ to guide the wolf pack to surround the prey; the mathematical model of the wolf pack's hunting behavior can be expressed as follows: D α =C1·X α -X,D β =C2·X β -X,D δ =C3·X δ -X, X1=X α -A1·D α ,X2=X β -A2·D β ,X3=X δ -A3·D δ , Where: X α , X β , X δ Respectively represent the position vectors of α, β, and δ in the current population; X represents the current position vector of the ω gray wolf; D α , D β , D δ They represent the distances between the current candidate gray wolf and the top three gray wolves with the best fitness; C1, C2, and C3 are random vectors; X(t+1) is the position of the final prey determined based on X1, X2, and X3; 4) Attacking prey The last step of the gray wolf hunting is to attack the prey. In order to simulate this process, the adjustment of the individual behavior of the gray wolf is achieved by controlling the value of parameter a. a is gradually reduced, so the fluctuation range of A is also reduced. In the process of algorithm iteration, when the value of α decreases linearly from 2 to 0, the corresponding A value changes in the interval [-a, a]. The value of α is adjusted according to the following formula: When the value of A is within the interval, the position of the gray wolf at the next moment can be anywhere between the current gray wolf and the prey; when |A| < 1, the wolf pack will attack the prey; 5) Finding prey: The gray wolf population relies on the positions of α, β, and δ to search and locate prey; in the process of searching for prey, the wolf pack first searches for the location information of the prey in a dispersed manner, and then gathers together to attack the prey; for the establishment of the dispersed model, |A|>1 is used to make the gray wolves disperse as much as possible in various areas and search for prey, so that GWO can conduct a global search and thus find the best prey; another search coefficient in the GWO algorithm is C; C is a random vector composed of the interval [0,2], which represents the random weight of the influence of the gray wolf's location on the prey. C>1 means that the influence weight is large, otherwise, it means that the influence weight is small; it helps to improve the randomness of GWO's algorithm search during the optimization process, thereby avoiding the algorithm from falling into the local optimum; C decreases nonlinearly; during the entire iteration process, C adjusts the global search of the algorithm, and the randomness of C can effectively avoid the algorithm from stagnating in search and falling into the local optimum; The optimization process of the Gray Wolf Optimization Algorithm is as follows: In the process of GWO algorithm optimization, a random population is first generated; then, in the process of algorithm iteration, the three best gray wolves α, β and δ determine the possible location of the prey, and the other gray wolf individuals update their positions according to the distance between α, β and δ and the prey; at the same time, the exploration and development process of the algorithm is balanced by controlling the parameter a; when |A|>1, the algorithm performs global exploration; when |A|<1, the algorithm performs local development; Step 5 includes the following steps: 1) Sample collection: Collect opening data under various working conditions and water heads as the sample set of the simulation model; 2) Feature set selection: select common feature sets and optimal hybrid feature sets as model inputs respectively; 3) Sample grouping: Divide the existing samples into two groups, namely the training group and the test group. In the process of simulating the model, first use the data of the training group to train the model, and then use the data of the test group to verify the performance of the model. 4) Sample normalization: Through the normalization method, all sample data values are adjusted to the range of [0,1], making the model calculation faster. The normalization conversion function is shown as follows: 5) Model construction: The steps of the simulation model based on GWO-BP neural network are as follows: Set the initial parameters of the model, including population size, maximum number of iterations, variable dimension, scaling factor and crossover probability coefficient CR; Initialize the population according to the formula in step 4). In order to ensure that the model error of the BP neural network is minimized after training, the average value of the absolute error between the actual output and the expected output of the BP neural network is selected as the objective function f. The optimization goal is to select the best BP neural network parameter value. Calculate the fitness of all gray wolf individuals in the population and arrange the fitness values in descending order. The top three gray wolf individuals in terms of fitness are defined as X α , X β , X δ ; Update the position of the parent population individuals; Update the parent gray wolf individual P α , P β , P δ , and then sort the parent gray wolf individuals according to the size of the fitness; at the same time, the termination condition of the algorithm iteration is judged, and when the iteration termination condition is met, the optimal individual P of the parent generation is returned. α and f(P α ) is the optimal network parameter corresponding to the model to ensure the minimum PID output; the GWO-BP neural network simulation model of the optimal solution is completed.
2. A primary frequency regulation feedforward device based on intelligent solution of turbine simulation, characterized in that: The invention comprises a primary frequency regulation control method based on intelligent solution of turbine simulation according to claim 1.
