Multi-Agent Modeling and State Optimization Method for Hydrogenerator Regulation

Abstract

The multi-agent modeling and state optimization method for regulating a hydrogenerator in this application, based on the characteristics of a turbine governor and the controlled object, establishes a mathematical model of the turbine regulation system. A step disturbance signal is introduced during model construction, and multi-agent is used to perform model identification on the governor part of the regulation system and the turbine and its water intake system part; it can ensure polynomial-time convergence, avoid premature convergence, and the improved multi-agent algorithm increases the global search ability in the solution space, breaks the local solution with the help of external forces, and avoids falling into the problems of premature convergence and genetic stability. The results confirm the effectiveness of the algorithm. Establishing an accurate mathematical model of the turbine regulation system is of great significance for power system analysis and calculation and improving the control performance of the governor, improves the power generation efficiency and quality of the hydrogenerator set, and its performance improvement has a positive effect on the hydropower plant, its power supply quality, economic benefits and safe operation.

Description

Technical Field

[0001] This application relates to a method for hydroturbine regulation modeling and state optimization, and particularly to a method for multi-agent modeling and state optimization of hydrogenerator regulation, belonging to the technical field of power regulation models. Background Art

[0002] The problem of the safe operation of the current power grid system has become increasingly prominent, and the control performance requirements for power equipment are also getting higher and higher. The modeling and parameter testing of the hydroturbine regulation system are one of the difficult problems in power generation regulation modeling that need to be solved urgently. The hydroturbine regulation system uses a hydroturbine governor as the controller and is a closed-loop control system composed of a hydrogenerator set and a water diversion system as the controlled object. It is a complex system with nonlinearity and non-minimum phase that is affected by water, machinery, and electricity. The basic task of the hydroturbine regulation system is to continuously regulate the active power of the hydrogenerator set according to the change of the load to maintain the unit speed (frequency) within the specified range, thereby controlling the frequency of the power grid to remain stable. Hydrogenerator sets often perform frequency regulation and peak shaving tasks in the power grid and start and stop frequently. The quality of their performance and the level of automation directly affect the normal operation of the units. Therefore, the performance of the hydroturbine regulation system has a significant impact on the power quality and safe and reliable operation of the power system.

[0003] With the increasing development of large-scale interconnected power systems for ultra-high voltage long-distance power transmission, the power consumption departments have continuously raised their requirements for power quality. The power consumption departments have put forward higher requirements for the regulation quality, regulation stability boundary, hydroturbine power generation efficiency, and optimal control of water-machine-electricity combination of the hydroturbine regulation system. The model parameters of the controlled object of the hydroturbine regulation system are related to the operating conditions of the unit itself and the load it carries. The hydroturbine model parameters vary greatly with different operating conditions, and the changes in the generator operating conditions and load will also affect some parameters of the generator model. With the expansion of the production scale of hydropower plants and the widespread application of medium and large-sized units, the improvement of the complexity of the power system, and the increasing number of various control means, the control and stability problems of the power system have become increasingly complex, which also poses higher and higher requirements for the control of hydropower plant units. Such requirements have not only been limited to the requirements for the control quality of a single control system, but also the coupling effects between control processes must be considered to obtain overall system stability and overall performance improvement.

[0004] Whether it is for the improvement of the performance of the turbine control system itself or for the analysis of the power system, it must be based on an accurate and reliable mathematical model. The testing and dynamic modeling of the four major parameters of the power system (including the generator, excitation system, prime mover and control system, and comprehensive load model) has become an urgent problem to be solved. The prior art has little research on the modeling method and state optimization for turbines and control systems, one of the four major parameters, especially for power system analysis and calculation. There is no complete, practical and unified method. The mathematical model of the turbine controller used in the power system analysis software is very different from the actual controller in terms of model structure, and it is far from meeting the needs of accurate analysis of the power turbine system. The prior art algorithm converges slowly, and the disadvantage model of premature local optimality cannot achieve parameter optimization.

[0005] The problems that need to be solved in the existing technology of turbine control modeling and state optimization and the key technical difficulties of this application include:

[0006] (1) Modeling and parameter testing of the turbine control system is a difficult problem in power generation control that needs to be solved urgently. The existing technology cannot realize the testing and dynamic modeling of the four major parameters of the power system (including generator, excitation system, prime mover and control system, and comprehensive load model) based on an accurate and reliable mathematical model. The existing technology has little research on the modeling method and state optimization for turbines and control systems, one of the four major parameters, especially for power system analysis and calculation. There is no complete, practical and unified method. The mathematical model of the turbine controller used in the power system analysis software is very different from the actual controller in operation, even in the model structure, which is far from meeting the needs of accurate analysis of the power turbine system. The existing algorithm converges slowly, the local optimum is too early, and the defect model cannot realize parameter optimization. Since the turbine generator set often performs frequency regulation and peak regulation tasks in the power grid, it is frequently started and stopped. At present, its performance is not good enough and the automation level is not high, which directly affects the normal operation of the unit. The poor performance of the turbine control system is very unfavorable to the power quality and safe and reliable operation of the power system.

[0007] (2) The existing technology for the control of hydropower plant units can no longer meet the requirements of the power consumption department for continuously improving power quality. The control quality, control stability boundary, turbine power generation efficiency, and optimal combined control of water turbine, generator, and power system cannot fully meet the requirements. Since the traditional genetic algorithm cannot guarantee polynomial-time convergence, it is difficult to avoid premature convergence. Its global search ability in the solution space is poor, and it is prone to premature convergence and genetic stability problems. The changes in the generator operating conditions and load affect the parameters of the generator model. With the expansion of the production scale of hydropower plants, the wide application of medium and large-sized units, the increasing complexity of the power system, and the growing variety of control means, the control and stability problems of water turbines have become increasingly complex. It is not only limited to the requirements for the control quality of a single control system, but also necessary to consider the coupling effects between various control processes. The existing technology cannot meet these requirements, cannot establish an accurate mathematical model of the water turbine control system, realize power system analysis and calculation, and improve the control performance of the governor, which affects the power supply quality, economic benefits, and safe operation of hydropower plants and the entire power system.

[0008] (3) The existing technology lacks the modeling and analysis of the water turbine control system, including the mathematical models of the water conveyance system and the water turbine governor, the modeling and analysis of the main hydraulic amplification component and the electro-hydraulic servo system, and the establishment of a unified governor model with output speed limit and pure delay for the servo system. It is unable to construct a unified model of the water turbine and the water conveyance system and a unified model of the parallel PID governor to support the model identification of the water turbine control system. It lacks the model identification of the water turbine control system, cannot realize the governor model identification and the model identification of the monotonic water turbine and its water conveyance system, and has not established a master control and identification model for the water turbine control system. It is unable to optimize the model state of the water turbine control system based on improved multi-agent technology, lacks parameter identification based on improved multi-agent technology, cannot construct a governor model, and cannot construct a model of the water turbine and its penstock. The existing technology for the modeling and state optimization algorithm of hydrogenerator control has poor effects, cannot diagnose water turbine power generation faults in a timely and accurate manner, cannot handle faulty equipment in a timely manner, and is not conducive to ensuring the safety and stability of the power system. Summary of the Invention

[0009] Based on the characteristics of the turbine governor and the controlled object, this application establishes a mathematical model of the turbine control system. A step disturbance signal is introduced during model construction, and multi-agent is used to perform model identification on the governor part of the control system and the turbine and its water intake system part. It can ensure polynomial-time convergence and avoid premature convergence. This application proposes an improved multi-agent, which increases the global search ability in the solution space, breaks the local solution with the help of external forces. Its essence is to simulate the cataclysm phenomenon in which a large number of species become extinct and only a few species survive during the biological renewal process, retains the current best solution, randomly generates other individuals again, and avoids premature convergence and genetic stability problems. The results confirm the effectiveness of the algorithm. Establishing an accurate mathematical model of the turbine control system is of great significance for power system analysis and calculation and improving the control performance of the governor. The control system has greatly improved the power generation efficiency and quality of the hydro-generator set, and its performance improvement has a positive effect on the power supply quality, economic benefits and safe operation of the hydropower plant and the entire power system.

[0010] To achieve the above technical effects, the technical solutions adopted in this application are as follows:

[0011] A multi-agent modeling and state optimization method for hydro-generator control, based on the characteristics of the turbine governor and the controlled object, establishes a mathematical model of the turbine control system. A step disturbance signal is introduced during model construction, and multi-agent is used to perform model identification on the governor part of the control system and the turbine and its water intake system part.

[0012] A Modeling and analysis of the turbine control system: Establish the mathematical models of the water intake system and the turbine governor respectively. The modeling analysis of the turbine governor includes the modeling analysis of the main hydraulic amplification element and the electro-hydraulic servo system, and establishes a unified governor model with the output speed limit and pure delay of the servo system. Construct a unified model of the turbine and the water intake system and a unified model of the parallel PID governor to support the model identification of the turbine control system.

[0013] B Model identification of the turbine control system: It is divided into governor model identification and monotonic turbine and its water intake system model identification, including both discrete models and continuous models. The monotonic turbine and its water intake system model identification includes a linear PID governor identification model and a PID governor identification model with a servomotor speed limit. Finally, establish the total control and identification model of the turbine control system.

[0014] C State optimization of the turbine control system model based on the improved multi-agent: One is the parameter identification based on the improved multi-agent, including: rule definition of the improved multi-agent, improvement method of the multi-agent, and operator construction of the improved multi-agent. The second is the construction of the governor model. The third is the construction of the turbine and its water pipeline model.

[0015] Preferably, for the modeling analysis of the water diversion system: When the water flow in the pipeline is disturbed and accompanied by changes in flow rate and water head, and the elastic deformation generated by the water body and the pipe wall cannot be ignored for the water hammer pressure value and the entire transition process, the elastic deformation propagates the disturbance along the pipeline in the form of a finite wave velocity α. Still starting from the fluid motion equation and continuity equation for solution, ignoring the frictional losses and minor terms in the pipeline, the continuity equation and motion equation of a certain fluid isolation body in the pipeline are derived:

[0016] Continuity equation:

[0017] Motion equation:

[0018] Q is the flow rate (m 3 / S) of a certain cross-section in the pipeline at time t, H is the water head (m) of a certain cross-section in the pipeline at time t, x is the distance (m) from the corresponding cross-section to the specified point, D is the pipeline diameter (m), F is the cross-sectional area of the pipeline (m 2 ), a is the water hammer wave velocity (m / S), f is the hydraulic friction resistance coefficient of the pipeline, and we get:

[0019]

[0020] Let the distance from the water turbine inlet to the reservoir water intake be L, the relative value of the water head increment be h, and the water hammer phase length T r = 2L / a. Taking the Laplace transform of Equations 3 and 4, the mutual relationship between the flow rate and water head at the water turbine inlet is obtained as:

[0021]

[0022] In the formula, h w is the pipeline constant, T r = 2L / a is the water hammer phase length; a is the water hammer wave velocity. The hyperbolic tangent function in the formula is expanded using a series as:

[0023]

[0024] Expanding the first few terms gives:

[0025]

[0026] Considering elastic water hammer, taking the first two terms for the numerator and the first three terms for the denominator, we get:

[0027]

[0028] If only rigid water hammer is considered, only the first term can be taken for both the numerator and the denominator, that is:

[0029]

[0030] Adopt a rigid model. It is determined that when the pressure pipeline is less than 600 - 800, it belongs to the small fluctuation situation. In some necessary cases including the draft tube, calculate the elastic water.

[0031] Preferably, the mathematical model of the main hydraulic amplification element: The main hydraulic amplification element of the governor is a hydraulic amplification stage composed of a main distributor valve and a main servomotor, regarded as a simple integral link, that is:

[0032]

[0033] Among them, Y(s) is the position of the servomotor, S(s) is the opening of the main distributor valve, and T y is the servomotor time constant. The main hydraulic amplification element is a series connection of an integral link and a second-order link.

[0034] Preferably, the linear mathematical model of the electro-hydraulic servo system: Adopt the structure of the governor + electro-hydraulic servo system. The required control law is generated by an analog circuit or a microcomputer, and the hydraulic amplification part forms its own closed loop to form a relatively independent electro-hydraulic servo system. Assume: 1) All elements with integral properties except the main hydraulic amplification element are self-closed loop by feedback; 2) The time constants of the remaining elements in all hydraulic amplification parts are much smaller than the main servomotor time constant; 3) All transfer coefficients from the electro-hydraulic converter input to the auxiliary servomotor output are considered together in the servomotor time constant T; 4) Ignore the influence of high-order factors and nonlinear factors;

[0035] Y C is the input of the servo system, is the electrical signal output by the governor, Y is the position of the servomotor, that is, the output of the servo system, and T y is the servomotor time constant. Staticially, Y follows Y C and has a one-to-one correspondence with it. It is a servo system, and the corresponding transfer function is:

[0036]

[0037] Establish the mathematical model of the electro-hydraulic servo system with output speed limit and pure delay: Introduce the corresponding saturation nonlinear link and pure delay link, and add the settings of the delay link and nonlinear link. Among them, the upper and lower saturation points of the saturation nonlinear link before the integrator are set at s max and -s max respectively used to adjust the maximum movement speed of the servomotor in the opening and closing directions. Set 0 and 1 to simulate the upper and lower limit numerical requirements of the nonlinear link, simulate the available stroke interval of the servomotor, and the pure delay link introduced at the system input end simulates the delay characteristics of the system.

[0038] Preferably, for the water turbine governor identification model: the governor mathematical model is jointly composed of the governor and the servomechanism model. Regarding them as a whole, the input excitation and output of the identification model are taken as the frequency input of the governor and the output displacement of the servomotor.

[0039] 1 - Linear PID governor identification model

[0040] The model to be identified for the linear PID governor identification model is:

[0041]

[0042] b 1 、b 2 、a 1 、a 2 、b p 、T y are model coefficients, or:

[0043]

[0044] Y(s) is the position of the servomotor, S(s) is the opening of the main relay valve. After the model identification is completed, β 1 、β 2 、α 0 、α 1 、α 2 、α 3 can all be regarded as known. First, substitute the equations:

[0045] b 2 =T d T n ,b 1 =T d +T n

[0046] a 2 =T′ n T d (b p +b t ),a 1 =b p T′ n +T d (b p +b t ) Equation 13

[0047] into the relational expressions: β 1 =b 1 ,β 2 =b 2 ,α 0 =b p ,α 1 =b p T y+a 1 ,α 2 =a 1 T y +a 2 α 3 =a 2 T y We get from:

[0048] β 1 =T d +T n

[0049] β 2 =T d T n

[0050] α 0 =b p

[0051] α 1 =b p T y +b p T′ n +T d (b p +b t )

[0052] α 2 =(b p T′ n +T d (b p +b t ))T y +T′ n T d (b p +b t )

[0053] α 3 =T y T′ n T d (b p +b t Equation 14

[0054] For the intermediate servomotor, by solving the above non - linear equations, we obtain its parameters b t , T d , b p , T y , T n and T′ n , which are calculated by the following formula if expressed in terms of the parameters of the parallel - type governor:

[0055]

[0056] If the pure time delay T is considered D , the pure delay time added to the parameters to be identified is independent, and the calculation of the regulator parameters still uses Equation 15;

[0057] 2 - Identification Model of PID Regulator with Servomotor Speed Limit

[0058] The linear part of the regulator model is:

[0059]

[0060] The servomechanism becomes a saturated nonlinear system. After the model identification is completed, b 2 , b 1 , a 2 , a 1 , b p , T y , S max and - S max are regarded as known, and the corresponding regulator parameters are calculated and solved by the following formula:

[0061] b 2 = T d T n

[0062] b 1 = T d + T n

[0063] a 2 = T′ n T d (b p + b t )

[0064] a 1 = b p T′ n + T d (b p + b t ) Equation 17

[0065] For the parallel PID - type regulator structure, its parameters are also obtained from Equation 15.

[0066] Preferably, the identification model of the monotonic turbine and the water conveyance system: If the monotonic turbine is considered to operate in the grid - connected state and the grid frequency fluctuation is assumed to be negligible, its identification model is written in the unified model form as:

[0067]

[0068] Where:

[0069] a 0 = 1 / emy

[0070] a 1 = e qh T w / e my

[0071] b 1 = -e y T w Equation 19

[0072] The torque M of the water turbine t and the flow rate Q vary with the guide vane opening a y , the runner vane opening , the head H and the rotational speed n. M r , Q r , n r and H r are the values of the torque, flow rate, rotational speed and head under the rated condition respectively. a M and y M are the maximum values of the guide vane opening and the servomotor displacement. is the maximum value of the runner vane opening. After the model identification is completed, a 0 , a 1 and b 1 are regarded as known;

[0073] Parameter calculation of the water turbine and its water intake system: There are only 3 independent coefficients a 0 , a 1 , b 1 in the model. Three relevant equations are listed. The number of actual parameters of the water turbine and its water intake system is greater than 3. Only the transfer coefficient e my = 1 / a 0 of the opening to the torque can be determined.

[0074] Preferably, the rule definition of the improved multi-agent: When obtaining an optimal solution, the optimal solution will be retained, and then new individuals will be randomly generated. In the next update, there will be individuals with more diversity in meaning than the previous individuals. The small-scale agents obtain the diversity of the large-scale agents and break away from the limitation of the local optimal solution;

[0075] Perform multi-agent processing on the agents according to the catastrophe requirements of the natural biological update history. This agent is the stage optimal solution. The specific process of the improved multi-agent includes:

[0076] (a) Initialize the agents and the target values;

[0077] (b) Evaluate the fitness and perform genetic operations;

[0078] (c) Judge whether the requirements of the improved multi-agent are met and perform multi-agent operations;

[0079] (d) Perform multi-agent operations and fitness evaluation at this stage;

[0080] (f) Increment the multi-agent count by 1;

[0081] (g) If the result reaches the expected situation, end; otherwise, save the best solution and continue with the catastrophe;

[0082] Adopt improved agents to enhance the solution space. When performing genetic update operations, locally optimal solutions are reserved in advance to prevent random algorithm search.

[0083] Preferably, the multi-agent improvement method: increasing the agent scale M improves the global performance of the multi-agent, but the cost is a corresponding increase in the solution time, and the solution time T is proportional to the product of M and the number of iterations k, i.e.: T = MKT, where T 0 is the average computation time for generating a new individual through genetic operations and fitness evaluation;

[0084] Based on premature convergence, the current optimal solution before the catastrophe implies the local optimal traits of the current agents in the i-th stage. After the multi-agent operation, it is saved as the 0-th individual, whose fitness is in a relatively high position compared to the fitness of other randomly generated individuals Adopt the tournament selection mechanism to ensure that it can participate in the new round of updates;

[0085] Specifically, before and after performing the multi-agent operation to generate the first-generation agents, among the 2M - 1 individuals in two consecutive stages, representing the advanced directions and optimal traits of the M individuals in the i-th stage, are more likely to be selected into the breeding pool than the other M - 1 individuals in the i-th stage. The tournament selection mechanism ensures that it will not be lost in the first iteration after the catastrophe operation, thus ensuring that the M individuals in the previous stage can be implicitly updated in this stage. The role of the best solution retained in the new round of update process is the same as the effect of directly adding the M individuals before the operation to participate in the update operation, enabling the algorithm to implicitly expand the agent scale under the condition of unchanged operation scale;

[0086] Assume the number of catastrophes is c, and the number of update iterations in each stage is k'. Then, in each stage, the computation time is: T (i) = Mk′T 0 , i = 0, 1, 2,..., C, and the total solution time: T = M(C + 1)k′T 0 ;

[0087] The implied operation scale at each stage is successively \(M+(M - 1)i\). If all the implied agents are allowed to participate in the update, the computing time at the \(i\)-th stage is: \(T'\) (i) =\([M+(M - 1)i]k'T\) 0 , the total solution time:

[0088] Obtained: When \(C>0\), there is \(T < T'\), and the greater the \(C\), the greater the gap between the two. As long as it will not be lost in the first iteration after multi-agent operation, the multi-agent operator can exert the implicit parallelism of multi-agents and improve the multi-agent efficiency.

[0089] Preferably, improve the construction of the multi-agent operator: when the genetic operation update condition does not meet the expected situation, perform the catastrophe operation of the improved multi-agent. Based on the stability of mutation and the feasibility of operation, adopt binary search coding;

[0090] This application uses the minimum fitness to represent the global optimal fitness, sets a pregen-c as the stagnation generation number. If the current generation number is greater than the set generation number, perform multi-agent operation to ensure that the generation number is between the second generation and the maximum generation number. At the same time, when the current optimal solution is less than the historical optimal solution, it is necessary to trigger the multi-agent operation. After the multi-agent operation, while retaining the optimal solution, set an arbitrary proportion to replace the original part, set an integer-taking environment to prevent the appearance of the decimal part;

[0091] After each multi-agent operation, it is necessary to initialize the new agents, set up a space to save the new fitness function, reserve a place to fill in numbers, define a new objective function, select the Griewank function as the expression of the objective function, sort the new agents and their fitness after the catastrophe in descending order for the next genetic and multi-agent operations. Each multi-agent operation has a catastrophe mark for easy observation until the maximum generation number of the genetic operation is reached, starting from the overall update of the agents and considering the convergence and stability of the algorithm.

[0092] Preferably, construct the regulator model: first construct a model parameter initialization program to initialize the parameters of the regulator model, the turbine and the water diversion system model. Start the model construction under SIMULINK. All the data recorded in the construction are stored in the file construction result file PC.mat in matrix form, with the variable name Rdt. The data contained in this file is provided to the identification algorithm;

[0093] The identification model is divided into two major parts, namely: the governor model (governor + servo system) and the water turbine and penstock system model. Among them, the governor model includes PI-type governors and PID-type governors. The identification data used for the governor model are frequency disturbances and servomotor displacements, and the identification data used for the water turbine and penstock system model are servomotor displacements and water turbine output torques. The corresponding ideal governor model is:

[0094]

[0095] Introduce a unified model discrimination criterion (this criterion is adopted by all the following methods). The calculation method of the evaluation function value is:

[0096]

[0097] The main parameters of the multi-agent are set as follows: agent size = 80; crossover probability = 0.7; mutation probability according to the stochastic universal sampling strategy; maximum number of genetic generations = 100, and one optimal solution is retained for each generation of mutation.

[0098] Compared with the prior art, the innovation points and advantages of this application are:

[0099] (1) Based on the characteristics of the water turbine governor and the controlled object, this application establishes a mathematical model of the water turbine control system. A step disturbance signal is introduced in the model construction, and a multi-agent is used to identify the governor part of the control system and the water turbine and its penstock system part. It can ensure polynomial-time convergence and avoid premature convergence. This application proposes an improved multi-agent. This algorithm increases the global search ability in the solution space, breaks the local solution with the help of external forces. Essentially, it simulates the cataclysm phenomenon in which a large number of species become extinct and only a few species survive in the biological renewal process, retains the current best solution, randomly generates other individuals again, and avoids falling into the problems of premature convergence and genetic stability. The results confirm the effectiveness of this algorithm. Establishing an accurate mathematical model of the water turbine control system is of great significance for power system analysis and calculation and improving the control performance of the governor. The control system greatly improves the power generation efficiency and quality of the water turbine generator set, and its performance improvement has a positive effect on the power supply quality, economic benefits and safe operation of the hydropower plant and the entire power system.

[0100] (2) This application proposes a modeling and analysis method for the hydroturbine governing system. The mathematical models of the water conveyance system and the hydroturbine governor are established respectively. The modeling and analysis of the hydroturbine governor include the modeling and analysis of the main hydraulic amplification element and the electro-hydraulic servo system, establishing a unified governor model with output speed limit and pure delay of the servo system, constructing a unified model of the hydroturbine and the water conveyance system and a unified model of the parallel PID governor, to support the model identification of the hydroturbine governing system; realizing the modeling and parameter testing of the hydroturbine governing system, and on the basis of an accurate and reliable mathematical model, realizing the testing and dynamic modeling of the four major parameters of the power system (including generators, excitation systems, prime movers, governing systems, and composite load models), and proposing a complete, practical and unified method. It meets the requirements of accurately analyzing the power hydroturbine system, with fast algorithm convergence, solves the shortcoming of premature local optimum of the model, and realizes parameter optimization. Hydrogenerator sets often undertake frequency regulation and peak shaving tasks in the power grid, with frequent start-up and shutdown. The improvement of their performance and automation level helps the normal operation of the units. The improvement of the performance of the hydroturbine governing system is very useful for the power quality and safe and reliable operation of the power system.

[0101] (3) This application proposes the model identification of the hydroturbine governing system, which is divided into governor model identification and monotonic hydroturbine and its water conveyance system model identification. The monotonic hydroturbine and its water conveyance system model identification includes a linear PID governor identification model and a PID governor identification model with servomotor speed limit. Finally, a total control and identification model of the hydroturbine governing system is established; realizing the model state optimization of the hydroturbine governing system based on improved multi-agent. Through parameter identification based on improved multi-agent, governor model construction, and hydroturbine and its penstock model construction, it meets the requirements of the power consumption department for continuously improving power quality. The regulation quality, regulation stability boundary, hydroturbine power generation efficiency, and optimal control of the hydro-mechanical and electrical integration of the hydroturbine governing system fully meet the requirements, ensuring polynomial time convergence, having strong global search ability in the solution space, and not falling into premature convergence and genetic stability problems. Meeting the requirements of the expansion of the production scale of hydropower plants and the wide application of medium and large-sized units, the increasing complexity of the power system, and the increasing variety of various control means, and the increasingly complex control and stability problems of hydroturbines. This application is not limited to the requirements for the control quality of a single control system, but also considers the coupling effects between control processes, establishes an accurate mathematical model of the hydroturbine governing system, and realizes the improvement of power system analysis and calculation and governor control performance. Description of the Drawings

[0102] Figure 1 It is a simplified block diagram of the electro-hydraulic servo system.

[0103] Figure 2 It is a simplified block diagram of the corresponding saturation nonlinear link of the servo system.

[0104] Figure 3 It is a unified model diagram of a speed governor considering the output speed limit and pure delay of a servo system.

[0105] Figure 4 It is an identification model diagram of a PID governor with a servomotor speed limit.

[0106] Figure 5 It is a schematic diagram summarizing the identification results of a PID governor.

[0107] Figure 6 It is a schematic diagram summarizing the identification results of a water turbine and its water intake system. Specific implementation manners

[0108] The following further describes the technical solution of the multi-agent modeling and state optimization method for water turbine generator regulation provided in this application with reference to the accompanying drawings, so that those skilled in the art can better understand this application and be able to implement it.

[0109] The control object of the water turbine control system is the water turbine generator set, which together with the water turbine governor as the system controller constitutes a closed-loop control system. Its performance is closely related to the power supply quality, economic benefits, and safe operation of the hydropower plant and the entire power system. Establishing an accurate mathematical model of the water turbine control system is of great significance for power system analysis and calculation and improving the control performance of the governor.

[0110] Since the traditional genetic algorithm cannot guarantee polynomial-time convergence and it is difficult to avoid premature convergence, this application proposes an improved multi-agent algorithm. This algorithm increases the global search ability in the solution space, breaks the local solution with the help of external forces, and essentially simulates the catastrophic phenomenon in which a large number of species become extinct and only a few species survive during the biological update process, retains the current best solution, randomly generates other individuals again, and avoids premature convergence and genetic stability problems. The results confirm the effectiveness of this algorithm.

[0111] I. Modeling analysis of the water turbine control system

[0112] (I) Modeling analysis of the water intake system

[0113] If there are disturbances in the water flow in the pipeline accompanied by changes in flow rate and water head, and the elastic deformation generated by the water body and the pipe wall cannot be ignored for the water hammer pressure value and the entire transient process, the elastic deformation will propagate the disturbance along the pipeline in the form of a finite wave speed α. Still, starting from the fluid motion equation and continuity equation for solution, ignoring the friction loss and secondary terms in the pipeline, the continuity equation and motion equation of a fluid isolation body in the pipeline are derived:

[0114] Continuity equation:

[0115] Motion equation:

[0116] Let \(Q\) be the flow rate (\(m^3 / s\)) of a certain cross - section in the pipeline at time \(t\), \(H\) be the water head (\(m\)) of a certain cross - section in the pipeline at time \(t\), \(x\) be the distance (\(m\)) from the corresponding cross - section to the specified point, \(D\) be the pipeline diameter (\(m\)), \(F\) be the cross - sectional area of the pipeline (\(m^2\)), \(a\) be the water hammer wave speed (\(m / s\)), and \(f\) be the hydraulic friction resistance coefficient of the pipeline. Then we have: 3 / S), \(H\) is the water head (m) of a certain cross - section in the pipeline at time \(t\), \(x\) is the distance (m) from the corresponding cross - section to the specified point, \(D\) is the pipeline diameter (m), \(F\) is the cross - sectional area of the pipeline (\(m^2\)), \(a\) is the water hammer wave speed (m / S), \(f\) is the pipeline hydraulic friction resistance coefficient, and we get: 2 ),a is the water hammer wave speed (m / S), f is the pipeline hydraulic friction resistance coefficient, and we get:

[0117]

[0118] Let the distance from the water turbine inlet to the reservoir intake be \(L\), the relative value of the water head increment be \(h\), and the water hammer phase length \(T = 2L / a\). Taking the Laplace transform of equations 3 and 4, the relationship between the flow rate and water head at the water turbine inlet is obtained as: r =2L / a, taking the Laplace transform of equations 3 and 4, the relationship between the flow rate and water head at the water turbine inlet is obtained as:

[0119]

[0120] In the formula, \(h\) w is the pipeline constant, \(T\) r =2L / a is the water hammer phase length; \(a\) is the water hammer wave speed. The hyperbolic tangent function in the formula is expanded using a series as:

[0121]

[0122] Expanding the first few terms, we get:

[0123]

[0124] Considering elastic water hammer, taking the first two terms for the numerator and the first three terms for the denominator, we get:

[0125]

[0126] If only rigid water hammer is considered, only the first term can be taken for both the numerator and the denominator, that is:

[0127]

[0128] Adopting a rigid model, it is considered that the pressure pipeline with a diameter less than 600 - 800 belongs to the small - fluctuation situation. In some necessary cases including the draft tube, elastic water is calculated.

[0129] (2) Modeling and analysis of the water turbine governor

[0130] According to the change of the unit load, continuously regulate the active power output of the hydro - generator set to maintain the unit speed (frequency) within the specified range. To monitor the system state, predict the system performance, and diagnose the system faults, an accurate governor model is established.

[0131] 1. Modeling and Analysis of Main Hydraulic Amplifying Element and Electro-Hydraulic Servo System

[0132] (1) Mathematical Model of Main Hydraulic Amplifying Element

[0133] The main hydraulic amplifying element of the governor consists of a main distributor valve and a main servomotor, which is regarded as a simple integral link, that is:

[0134]

[0135] Among them, Y(s) is the position of the servomotor, S(s) is the opening of the main distributor valve, and T y is the servomotor time constant. The main hydraulic amplifying element is a series connection of an integral link and a second-order link.

[0136] (2) Linear Mathematical Model of Electro-Hydraulic Servo System

[0137] Adopting the structure of governor + electro-hydraulic servo system, the required control law is generated by analog circuit or microcomputer, and the hydraulic amplification part forms its own closed loop to form a relatively independent electro-hydraulic servo system. Assume: 1) All elements with integral properties except the main hydraulic amplifying element are self-closed loop by feedback; 2) The time constants of the remaining elements in all hydraulic amplification parts are much smaller than the servomotor time constant; 3) All transfer coefficients from the electro-hydraulic converter input to the auxiliary servomotor output are included in the servomotor time constant T for unified consideration; 4) The influence of high-order factors and non-linear factors is ignored. The simplified block diagram of the servo system is drawn as Figure 1 shown.

[0138] In the figure, Y C is the input of the servo system, is the electrical signal output by the governor, Y is the position of the servomotor, that is, the output of the servo system, and T y is the servomotor time constant. Staticly, Y follows Y C and has a one-to-one correspondence with it. For the servo system, the corresponding transfer function is:

[0139]

[0140] (3) Establishing the Mathematical Model of Electro-Hydraulic Servo System with Output Speed Limit and Pure Delay

[0141] Introduce the corresponding saturation non-linear link and pure delay link, as Figure 2 shown. Add the settings of the delay link and non-linear link. Among them, the upper and lower saturation points of the saturation non-linear link before the integrator are set at s max and -s maxThey are respectively used to adjust the maximum movement speeds of the servomotor in the opening and closing directions, set 0 and 1 to simulate the upper and lower limit numerical requirements of the non-linear link, simulate the available stroke interval of the servomotor, and the pure delay link introduced at the system input end simulates the delay characteristics of the system.

[0142] 2. Establish a unified model of the governor with output speed limit and pure delay for the follow-up system

[0143] Establish the output speed limit and pure delay of the electro-hydraulic follow-up system, and combine Figure 2 the follow-up system model with a saturation non-linear link (speed limit model) and a pure delay link introduced with the unified model of the hydraulic turbine governor to form a more applicable unified PID model of the hydraulic turbine governor under large disturbance conditions, as Figure 3 shown.

[0144] II. Model identification of the hydraulic turbine control system

[0145] (1) Identification model of the hydraulic turbine governor

[0146] The mathematical model of the governor consists of the governor and the follow-up system model. Regarding them as a whole, the input excitation and output of the identification model are taken as the frequency input of the governor (frequency measurement or frequency setting) and the output displacement of the servomotor;

[0147] 1. Identification model of the linear PID governor

[0148] The model to be identified for the identification model of the linear PID governor is:

[0149]

[0150] b 1 、b 2 、a 1 、a 2 、b p 、T y are model coefficients, or:

[0151]

[0152] Y(s) is the position of the servomotor, S(s) is the opening of the main pilot valve. After the model identification is completed, β 1 、β 2 、α 0 、α 1 、α 2 、α 3 can all be regarded as known. First, transform the equation:

[0153] b 2 =T d T n ,b 1 =Td +T n

[0154] a 2 = T' n T d (b p +b t ),a 1 = b p T' n +T d (b p +b t ) Equation 13

[0155] Substitute into the relationship: β 1 = b 1 ,β 2 = b 2 ,α 0 = b p ,α 1 = b p T y +a 1 ,α 2 = a 1 T y +a 2 α 3 = a 2 T y to get:

[0156] β 1 = T d +T n

[0157] β 2 = T d T n

[0158] α 0 = b p

[0159] α 1 = b p T y +b p T' n +T d (b p +b t )

[0160] α 2 = (b p T' n +T d (b p +b t ))T y +T' n Td (b p +b t )

[0161] α 3 =T y T′ n T d (b p +b t ) Equation 14

[0162] For the intermediate servomotor, solve the above nonlinear equations to obtain its parameters b t , T d , b p , T y , T n and T′ n , if expressed in terms of the parameters of the parallel-type controller, calculate using the following formula:

[0163]

[0164] If the pure time delay T D is considered, the additional pure delay time of the parameters to be identified is independent, and Equation 15 is still used to calculate the controller parameters.

[0165] 2. Identification Model of PID Controller with Servomotor Speed Limit

[0166] The identification model of the PID controller with servomotor speed limit is as Figure 4 shown. The linear part of the controller model is:

[0167]

[0168] The follow-up system becomes a saturated nonlinear system. After the model identification is completed, b 2 , b 1 , a 2 , a 1 , b p , T y , S max and -S max are regarded as known, and the corresponding controller parameters are calculated and solved using the following formula:

[0169] b 2 =T d T n

[0170] b 1 =T d +T n

[0171] a 2 =T′ n Td (b p +b t )

[0172] a 1 =b p T′ n +T d (b p +b i ) Equation 17

[0173] For the parallel PID type controller structure, its parameters are also obtained from Equation 15.

[0174] (2) Identification model of monotonic turbine and water conveyance system

[0175] If the monotonic turbine is considered to operate in the grid-connected state and the grid frequency fluctuation is assumed to be negligible, its identification model can be written in a unified model form as follows:

[0176]

[0177] Where:

[0178] a 0 =1 / e my

[0179] a 1 =e qh T w / e my

[0180] b 1 =-e y T w Equation 19

[0181] The turbine torque M t and the flow rate Q vary with the guide vane opening a y , the runner vane opening , the head H and the rotational speed n. M r , Q r , n r and H r are the values of the torque, flow rate, rotational speed and head under the rated condition respectively. a M and y M are the maximum values of the guide vane opening and the servomotor displacement. is the maximum value of the runner vane opening. After the model identification is completed, a 0 , a 1 and b 1 are regarded as known;

[0182] Calculation of the parameters of the turbine and its water conveyance system: There are only 3 independent coefficients a 0 , a 1 , b1 List three relevant equations. The number of actual parameters of the corresponding water turbine and its water intake system is greater than 3, and the actual parameters cannot be completely determined by the model parameters. Only the transfer coefficient e of the opening to the torque can be determined my = 1 / a 0 .

[0183] III. State optimization of the water turbine regulation system model based on improved multi-agent

[0184] (I) Parameter identification based on improved multi-agent

[0185] 1. Rule definition of the improved multi-agent

[0186] After obtaining an optimal solution, the optimal solution will be retained, and then new individuals will be randomly generated. In the next update, individuals with more diversity in meaning than the previous individuals will be available, and the small-scale agents can obtain the diversity of the large-scale agents, breaking away from the limitation of the local optimal solution.

[0187] Perform multi-agent processing on the agents according to the requirements of the catastrophe in the natural biological update history. This agent is the optimal solution at a certain stage. The specific process of the improved multi-agent includes:

[0188] (a) Initialize the agent and the target value;

[0189] (b) Evaluate the fitness and perform genetic operations;

[0190] (c) Judge whether the requirements of the improved multi-agent are met and perform multi-agent operations;

[0191] (d) Perform the multi-agent operation and fitness evaluation at this stage;

[0192] (f) Increment the multi-agent count by 1

[0193] (g) If the result meets the expected situation, end; otherwise, save the best solution and continue the catastrophe;

[0194] Adopt the improved agent to increase the solution space. Retain the local optimal solution in advance during the genetic update operation to prevent random algorithm search and ensure the stability of the multi-agent.

[0195] 2. Multi-agent improvement method

[0196] Increasing the agent scale M can improve the global performance of the multi-agent, but the cost is that the solution time increases accordingly. The solution time T is proportional to the product of M and the number of iterations k, that is: T = MKT, where T 0 is the average calculation time to generate a new individual through genetic operations and fitness evaluation;

[0197] Based on premature convergence, the current optimal solution before the catastrophe The local optimal traits of the current agent in the i-th stage are implied. After the multi-agent operation, it is saved as the 0-th individual, and its fitness is relatively high compared to the fitness of other individuals randomly generated again. The tournament selection mechanism is adopted to ensure that it can participate in the new round of update; Specifically, before and after the multi-agent operation to generate the first-generation agents, among the 2M - 1 individuals in two consecutive stages, the advanced directions and optimal traits of the M individuals representing the i-th stage

[0198] are more likely to be selected into the breeding pool than the other M - 1 individuals in the i-th stage. The tournament selection mechanism ensures that it will not be lost in the first iteration after the catastrophe operation, so as to ensure that the M individuals in the previous stage can be implicitly updated in this stage. The role of the best solution retained in the new round of update process is the same as the effect of directly adding the M individuals before the operation to participate in the update operation, enabling the algorithm to implicitly expand the agent scale under the condition of unchanged operation scale.

[0199] Assume that the number of catastrophes is c, and the number of update iterations in each stage is k'. Then in each stage, the calculation time is: T (i) 0 = Mk'T 0 , i = 0, 1, 2,..., C, and the total solution time: T = M(C + 1)kT 0 ;

[0200] The implicitly operating scales of each stage are M + (M - 1)i in turn. If all the implicitly generated agents are allowed to participate in the update, the calculation time in the i-th stage is: T' (i) = [M + (M - 1)i]k'T 0 , and the total solution time:

[0201] It is obtained that: When C > 0, there is T < T', and the larger C is, the greater the gap between the two. As long as it will not be lost in the first iteration after the multi-agent operation, the multi-agent operator can exert the implicit parallelism of multi-agents and improve the multi-agent efficiency.

[0202] 3. Improvement of the multi-agent operator construction

[0203] When the genetic operation update condition does not meet the expected situation, the catastrophe operation of the improved multi-agent is executed. Based on the stability of mutation and the feasibility of the operation, binary search coding is adopted;

[0204] This application uses the minimum fitness value to represent the global optimal fitness. A pregen - c is set as the stagnation generation number. If the current generation number is greater than the set generation number, multi - agent operations are performed, ensuring that the generation number is between the second generation and the maximum generation number. At the same time, only when the current optimal solution is less than the historical optimal solution is it necessary to trigger multi - agent operations. After multi - agent operations, while retaining the optimal solution, a random proportion is set to replace the original part, and an integer - valued environment is set up to prevent the appearance of decimal parts.

[0205] After each multi - agent operation, the new agents need to be initialized. Set up a space to save the new fitness function, reserve places for filling numbers, define a new objective function, and choose the Griewank function as the expression of the objective function. Sort the new agents and their fitness values after catastrophe in descending order for the next genetic and multi - agent operations. Each multi - agent operation has a catastrophe mark for easy observation until the maximum generation number of genetic operations is reached. Starting from the overall update of the agents, consider the convergence and stability of the algorithm.

[0206] (2) Construction of the regulator model

[0207] First, construct a model parameter initialization program to initialize the parameters of the regulator model, the water turbine, and the penstock system model. Start model construction under SIMULINK. All the data recorded during construction is stored in the file PC.mat in matrix form, with the variable name Rdt. The data contained in this file is provided to the identification algorithm.

[0208] The identification model is divided into two major parts, namely: the regulator model (regulator + servo system) and the water turbine and penstock system model. Among them, the regulator model includes PI - type regulators and PID - type regulators. The identification data used for the regulator model is frequency perturbation (as input data) and servomotor displacement (as output data). The identification data used for the water turbine and penstock system model is servomotor displacement (as input data) and turbine output torque (as output data). The corresponding ideal regulator model is:

[0209]

[0210] Introduce a unified model discrimination criterion (all the following methods use this criterion). The calculation method of the evaluation function value is:

[0211]

[0212] The main parameters of the multi - agent are set as: agent size = 80; crossover probability = 0.7; mutation probability according to the random walk sampling strategy; maximum genetic generation number = 100. One optimal solution is retained for each generation of mutation.

[0213] (3) Construction of the Turbine and Its Water Conduit Model

[0214] Parameters of the turbine and its water conveyance system used in the construction: T w = 1.5 s, e my = 1.0, e qh = 0.5, e y = 1.0, corresponding to A 0 = 1.0, A 1 = 0.75, B 1 = -1.5, and the ideal model of the turbine and its water conveyance system is as follows:

[0215]

[0216] Corresponding model:

[0217]

[0218] Corresponding to VAal = 0.059442.

[0219] (4) Comparative Analysis of Identification Results

[0220] For the convenience of comparison, the identification results obtained by various methods are summarized in units of the identification object;

[0221] (1) Summary of Identification Results of PID Controllers

[0222] As Figure 5 , since the inertia of the servo system and its inherent speed limit non-linearly weaken the high-frequency components in the PID controller, by comparing the direct relationships among the three methods, it can be found that the improved multi-agent significantly improves in terms of error. The improved multi-agent model can better reflect the characteristics of the actual turbine governor system and is more in line with the actual situation.

[0223] (2) Turbine and Its Water Conduit System

[0224] As Figure 6 , the summary of identification results of the turbine and its water conveyance system shows that good identification effects are obtained for various methods (least squares identification-based, multi-agent-based model identification, improved multi-agent-based model identification) and step disturbance signals. Under the identification model of the improved multi-agent, the results are already very close to the ideal model, further verifying the feasibility of the improved multi-agent. Considering the feasibility of the on-site model, the time spent by the identification algorithm, the evaluation function value and error of the identification method, for the model identification of the turbine and its water conveyance system, the identification method based on the improved multi-agent can be recommended.

Claims

1. Multi-agent modeling and state optimization method for hydro-generator control, characterized by: Based on the characteristics of the turbine controller and the controlled object, a mathematical model of the turbine control system is established. A step disturbance signal is introduced into the model, and multi-agents are used to perform model identification on the controller part of the control system and the turbine and its water diversion system part. A. Modeling and analysis of turbine control system: mathematical models of water diversion system and turbine controller are established respectively. Modeling and analysis of turbine controller includes modeling and analysis of main hydraulic amplification element and electro-hydraulic servo system, establishment of unified model of controller with output speed limit and pure delay of servo system, construction of unified model of turbine and water diversion system and unified model of parallel PID controller, and support identification of turbine control system model; B. Model identification of hydraulic turbine control system: It is divided into model identification of the controller and model identification of monotonic hydraulic turbine and its water diversion system, which include both discrete model and continuous model. Model identification of monotonic hydraulic turbine and its water diversion system includes linear PID controller identification model and PID controller identification model with relay speed limit. Finally, the master control and identification model of hydraulic turbine control system is established. C State optimization of the turbine control system model based on improved multi-agent: First, parameter identification based on improved multi-agent, including: improved multi-agent rule definition, multi-agent improvement method, improved multi-agent operator construction; second, regulator model construction; third, turbine and water diversion pipeline model construction.

2. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Modeling and analysis of water diversion system: If the water flow in the pipeline is disturbed and accompanied by changes in flow rate and water head, and the elastic deformation of the water body and the pipe wall cannot be ignored in terms of the water hammer pressure value and the entire transition process, the elastic deformation propagates the disturbance along the pipeline in the form of a finite wave speed α. The solution is still based on the motion equation and continuity equation of the fluid, ignoring the friction loss and minor terms in the pipeline, and the continuity equation and motion equation of a fluid isolation body in the pipeline are derived: Q is the flow rate of a certain flow section in the pipeline at time t (m 3 / S), H is the water head of a certain flow section in the pipeline at time t (m), x is the distance from the corresponding flow section to the specified point (m), D is the pipeline diameter (m), and F is the cross-sectional area of ​​the pipeline (m 2 ), a is the water hammer wave velocity (m / S), f is the pipeline hydraulic friction resistance coefficient, and we get: Assume that the distance from the turbine inlet to the reservoir water intake is L, the relative value of the head increment is h, and the water hammer phase length is T. r =2L / a, and by performing Laplace transformation on equations 3 and 4, the relationship between the flow rate and the water head at the turbine inlet is obtained as follows: In the formula, h w is the pipeline constant, T r =2L / a is the water hammer phase; a is the water hammer wave velocity, and the hyperbolic tangent function in the formula is expanded using series as follows: Expand the first few items to get: Considering elastic water hammer, take the first two terms in the numerator and the first three terms in the denominator, and we get: If only rigid water hammer is considered, the numerator and denominator can only take the first term, that is: A rigid model is used to identify pressure pipes less than 600-800 as small fluctuations. In some necessary cases including tailwater pipes, elastic water is calculated.

3. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Mathematical model of the main hydraulic amplifier element: The main hydraulic amplifier element of the regulator is a hydraulic amplifier stage composed of the main pressure regulating valve and the main relay, which can be regarded as a simple integral link, namely: Where, Y(s) is the position of the relay, S(s) is the opening of the main pressure regulating valve, T y is the relay time constant, and the main hydraulic amplifier element is connected in series as an integral link and a second-order link.

4. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: The linear mathematical model of the electro-hydraulic servo system: the structure of the controller + electro-hydraulic servo system is adopted, the required control law is generated by the analog circuit or microcomputer, and the hydraulic amplification part is closed by itself, forming a relatively independent electro-hydraulic servo system, assuming that: 1) except for the main hydraulic amplification element, all the elements with integral properties are closed by feedback; 2) the time constants of the remaining elements in all hydraulic amplification parts are much smaller than the main relay time constant; 3) all the transfer coefficients from the electro-hydraulic converter input to the auxiliary relay output are taken into account in the relay time constant T; 4) the influence of high-order factors and nonlinear factors are ignored; Y C is the input of the servo system, is the electrical signal output by the regulator, Y is the position of the relay, that is, the output of the servo system, T y is the relay time constant. In static state, Y follows Y C , and there is a one-to-one correspondence with it, which is a follow-up system, and the corresponding transfer function is: Establish a mathematical model of the electro-hydraulic servo system with output speed limitation and pure delay: introduce the corresponding saturated nonlinear link and pure delay link, add the setting of delay link and nonlinear link, where the upper and lower saturation points of the saturated nonlinear link before the integrator are set at s max and -s max They are used to adjust the maximum movement speed of the relay in the opening and closing directions respectively, set 0 and 1 to simulate the upper and lower limit numerical requirements of the nonlinear link, simulate the available travel range of the relay, and the pure delay link introduced at the system input end simulates the delay characteristics of the system.

5. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Identification model of water turbine controller: The mathematical model of the controller is composed of the controller and the servo system model, which are classified as a whole. The input excitation and output of the identification model are taken as the frequency input of the controller and the output displacement of the relay; 1-Linear PID controller identification model The linear PID controller identification model to be identified is: b1, b2, a1, a2, bp, Ty are model coefficients, or: Y(s) is the position of the relay, S(s) is the opening of the main pressure regulating valve. After the model identification is completed, β1, β2, α0, α1, α2, and α3 can all be considered known. First, the formula: b2=T d T n ,b1=T d +T n a2 = T' n T d (b p + b t ), a1 = b p T' n + T d (b p + b t ) Equation 13 Substitute into the relationship: β1=b1,β2=b2,α0=b p , α1=b p T y +a1, α2 = a1T y +a2 α3=a2T y In the result: β1=T d +T n β2=T d T n α0＝b p α1=b p T y +b p T′ n +T d (b p +b t ) α2=(b p T′ n +T d (b p +b t ))T y +T′ n T d (b p +b t ) α3 = T y T' n T d (b p + b t ) Equation 14 For the intermediate actuator, solve the above nonlinear equations to obtain its parameter b t , T d 、b p , T y , T n and T′ n , if expressed as the parameters of a parallel controller, is calculated by the following formula: If we consider a pure time delay T D , the pure delay time added by the parameter to be identified is independent, and the calculation of the controller parameters still uses Equation 15; 2-PID controller identification model with relay speed limit The linear part of the regulator model is: The servo system becomes a saturated nonlinear system. After the model identification is completed, b2, b1, a2, a1, b p , T y , S max , and -S max Assuming it is known, calculate the corresponding controller parameters and solve them by the following formula: b2=T d T n b1=T d +T n a2=T′ n T d (b p +b t ) a1 = b p T' n + T d (b p + b t ) Equation 17 For the parallel PID controller structure, its parameters are also obtained by formula 15.

6. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Identification model of monotonic turbine and water diversion system: If the monotonic turbine is considered to be running in the grid-connected state and the grid frequency fluctuation is assumed to be negligible, its identification model can be written in the form of a unified model: Where: a0=1 / e my a1=e qh T w / e my b1=-e y T w Formula 19 Turbine torque M t and flow rate Q with guide vane opening a y , Blade opening Changes in water head H and speed n, M r , Q r 、n r and H r are the values ​​of torque, flow, speed and water head under rated working conditions, respectively. M and M is the maximum value of the guide vane opening and the servomotor displacement, is the maximum blade opening. After the model identification is completed, a0, a1 and b1 are considered known; Parameter calculation of the turbine and its water diversion system: There are only three independent coefficients a0, a1, b1 in the model, and three related equations are listed. The number of actual parameters of the corresponding turbine and its water diversion system is greater than 3, and only the transmission coefficient of the opening to the torque e can be determined. my =1 / a0.

7. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Improved multi-agent rule definition: When an optimal solution is obtained, it will be retained, and then new individuals will be randomly generated. In the next update, there will be individuals with more diversity than the previous individuals. The small scale of agents can obtain the diversity of the large scale of agents, breaking away from the limitation of local optimal solutions. According to the catastrophic requirements of the biological renewal history of nature, the intelligent agent is processed by multi-agent. This intelligent agent is the optimal solution for the stage. The specific process of improving the multi-agent includes: (a) Initialize the agent and target value; (b)) Evaluate fitness and perform genetic manipulations; (c) Determine whether the requirements for improving multi-agents are met and perform multi-agent operations; (d) Perform multi-agent operations and fitness evaluation at this stage; (f) The number of multi-agents increases by 1; (g) If the result meets the expected situation, then end; otherwise, save the best solution and continue the disaster; The improved agent is used to improve the solution space, and the local optimal solution is reserved in advance when performing genetic update operations to prevent random algorithm search.

8. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Improved method for multi-agents: Increasing the size of agents M improves the global performance of multi-agents, but the cost is a corresponding increase in solution time, and the solution time T is proportional to the product of M and the number of iterations k, that is, T = MKT, where T0 is the average computational time to generate a new individual through genetic operations and fitness evaluation; Based on premature convergence, the current optimal solution before the catastrophe Implicitly, the local optimal characteristics of the current agent in stage i, after the multi-agent operation occurs, Save as individual 0, Its fitness is the same as other randomly generated individuals The fitness of is at a higher position than that of the previous one, and the tournament selection mechanism is adopted to ensure that it can participate in the next round of updates; Specifically, before and after the multi-agent operation is performed to generate the first generation of agents, among the 2M-1 individuals in two consecutive stages, the advanced direction and optimal traits of the M individuals in the i-th stage are represented. It is easier to be selected into the breeding pool than the other M-1 individuals in stage i. The tournament selection mechanism ensures It will not be lost in the first iteration after the catastrophe operation, thus ensuring that the M individuals in the previous stage can be implicitly updated in this stage. The role played by the retained best solution in the new round of update process is the same as the effect of the M individuals before the operation directly joining in the update operation, so that the algorithm implicitly expands the scale of intelligent agents without changing the scale of operation. Assuming the number of catastrophes is c, and the number of update iterations in each stage is k', then in each stage, the calculation time is: T(i) = Mk'T0, i = 0, 1, 2, ..., C, and the total solution time is: T = M(C+1)k'T0; The implicit operation scale of each stage is M+(M-1)i. If all implicit agents are allowed to participate in the update, the calculation time of the i-th stage is: Total solution time: get: When C>0, T<T', and the larger C is, the greater the difference between the two. It will not be lost in the first iteration after the multi-agent operation. The multi-agent operator can give full play to the implicit parallelism of multi-agents and improve the efficiency of multi-agents.

9. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Improved multi-agent operator construction: When the genetic operation update conditions do not meet the expected situation, the improved multi-agent catastrophic operation is performed, and binary search coding is used based on the stability of mutation and the feasibility of operation; This application uses the minimum fitness value to represent the global optimal fitness, sets a pregen-c as the stagnant generation, and performs multi-agent operation if the current generation is greater than the set generation, ensuring that the generation is between the second generation and the maximum generation, and the current optimal solution is less than the historical optimal solution, which is necessary to trigger the multi-agent operation. After the multi-agent operation, while retaining the optimal solution, set an arbitrary ratio to replace the original part, set a number rounding environment, and prevent the occurrence of decimal parts; After each multi-agent operation is completed, the new agent must be initialized, space is set to save the new fitness function, place is reserved to fill in numbers, a new objective function is defined, and the Griewank function is selected as the expression of the objective function. The new agents and their fitness after the disaster are arranged in descending order to facilitate the next genetic and multi-agent operations. Each multi-agent operation has a disaster mark for easy observation until the set maximum generation of genetic operations is reached. Starting from the overall update of the agent, the convergence and stability of the algorithm are taken into consideration.

10. The multi-agent modeling and state optimization method for hydro-generator control according to claim 1 is characterized in that: Regulator model construction: First, a model parameter initialization program is constructed to initialize the parameters of the regulator model and the turbine and water diversion system model. The model construction is started under SIMULINK. All the data recorded in the construction are stored in the file construction result file PC.mat in the form of a matrix. The variable name is Rdt. The data contained in this file is provided to the identification algorithm. The identification model is divided into two parts, namely, the controller model (controller + follow-up system) and the turbine and water diversion system model. The controller model includes PI-type controller and PID-type controller. The identification data used by the controller model are frequency disturbance and relay displacement, and the identification data used by the turbine and water diversion system model are relay displacement and turbine output torque. The corresponding ideal model of the controller is: A unified model discrimination criterion is introduced (all the following methods adopt this criterion), and the evaluation function value calculation method is: The main parameters of multi-agent are set as follows: agent size = 80; crossover probability = 0.7; mutation probability based on random traversal sampling strategy; maximum genetic generations = 100, and an optimal solution is retained in each generation of mutation.

Images