A hierarchical learning-based energy storage converter control system optimization method

Abstract

This invention provides an optimization method for an energy storage converter control system based on hierarchical learning. First, a set of particles is randomly generated, each representing a set of possible controller parameters. Next, the fitness of each particle is evaluated based on the system's performance indicators, and the particles are divided into multiple levels according to their fitness values. Different learning strategies are applied to particles at different levels; lower-level particles focus more on exploring new parameter spaces, while higher-level particles focus more on developing within known optimal parameter regions. By randomly selecting two particles from higher levels as samples, the learning of other particles is guided, thereby updating the particle positions. This process is repeated until the required number of iterations is met or a satisfactory solution is found. This invention employs the LPSO algorithm, and through the hierarchical learning strategy, significantly improves the algorithm's global search capability and local development capability, enabling it to find higher-quality solutions to complex optimization problems.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control optimization of energy storage converters, and particularly relates to an optimization method for energy storage converter control systems based on hierarchical learning. Background Technology

[0002] In today's rapidly developing field of power electronics, power converters serve as a crucial bridge connecting energy sources and loads, and their performance directly impacts the stability and efficiency of the entire power system. The core of a power converter system lies in its controller design, and optimizing controller parameters is paramount for system performance. With the continuous development of new energy technologies, such as the increasing demand for grid connection of renewable energy sources like solar and wind power, the control strategies and parameter optimization of power converters have become more complex and urgent issues.

[0003] Traditional controller parameter optimization methods, including classic PID control, fuzzy control, and genetic algorithms, often have certain limitations. For example, PID control parameter tuning relies on the engineer's experience and is difficult to adapt to complex and ever-changing system dynamics; while fuzzy control has a certain degree of adaptability, it lacks a systematic optimization method for parameter tuning.

[0004] Evolutionary algorithms, as a method that can break free from dependence on engineering experience and optimize parameter tuning, offer a solution for the control system optimization of energy storage converter systems. However, traditional evolutionary algorithms also have some drawbacks:

[0005] (1) Traditional evolutionary algorithms may require a large number of populations and multiple generations of iterations to ensure the quality and diversity of solutions, which may result in high computational costs.

[0006] (2) Traditional evolutionary algorithms require users to manually set and adjust multiple parameters, such as crossover rate, mutation rate and population size, which may be a challenge for users.

[0007] (3) When dealing with complex problems with many local optima, traditional evolutionary algorithms may converge to local optima prematurely. Summary of the Invention

[0008] Objective: The technical problem this invention aims to solve is to address the shortcomings of existing technologies by providing a hierarchical learning-based optimization method for energy storage converter control systems. This method employs a hierarchical learning-based particle swarm optimization (LPSO) algorithm to optimize the controller parameters of the energy storage converter. Compared to the traditional particle swarm optimization (PSO) algorithm, LPSO achieves more efficient search and optimization by dividing particles into different levels and applying differentiated learning strategies at each level.

[0009] The method includes the following steps:

[0010] Step 1: Collect historical power grid data and preprocess it;

[0011] Step 2: Calculate the fitness value using the fitness function and evaluate the fitness value of each particle to measure the performance of each particle;

[0012] Step 3: Divide the particles into different levels according to their fitness. Divide the particles into low-level particles and high-level particles according to their fitness to form a hierarchical learning structure. The low-level particles focus on exploring new parameter spaces, while the high-level particles focus on developing within the known excellent parameter regions.

[0013] Step 4: Through a hierarchical learning strategy, lower-level particles enhance their exploration capabilities by learning from higher-level particles; higher-level particles focus on development within the known region of excellent parameters.

[0014] Step 5: Randomly select two particles from the high level as samples, which are used to update the individual's best pbest and the neighborhood's best nbest, respectively, to guide the learning of low-level particles.

[0015] Step 6: Set the parameters for Layered Particle Swarm Optimization (LPSO) algorithm based on hierarchical learning;

[0016] Step 7: Adaptively adjust the inertia weights of the hierarchical learning-based particle swarm optimization algorithm LPSO.

[0017] Step 8: Update particle positions based on system feedback and adjust controller parameters;

[0018] Step 9: Repeat the evaluation, stratification, learning, and updating process until the optimal solution is found;

[0019] Step 10: Apply the parameters corresponding to the particles with the best fitness to the energy storage converter system.

[0020] Step 1 includes: the historical power grid data includes voltage V(t) at time t, current I(t) at time t, and power factor. .

[0021] Step 2 includes: calculating fitness using the following formula. :

[0022] ,

[0023] in, , , These are weighting coefficients; f1 is the output voltage tracking speed index; f2 is the output voltage overshoot; and f3 is the output voltage static error.

[0024] Step 3 includes: In the hierarchical learning-based particle swarm optimization algorithm LPSO, NP particles are divided into NL levels, each containing the same number of particles, called the level size LS, which is calculated using the following formula:

[0025] ,

[0026] Particle fitness value This determines the position of the particle in the hierarchy; the i-th particle The mathematical expression for being assigned to the l-th layer is:

[0027] ,

[0028] in, It is the i-th particle The fitness value ranks the particles among all particles. This indicates rounding up to the nearest integer.

[0029] In step 4, the LPSO algorithm is layered using a hierarchical learning strategy. The specific steps include:

[0030] Particle hierarchical partitioning: Particles are sorted from high to low fitness values, and the sorted particles are divided into partitions from L1 to L2. NL L1 is the highest level, L NL This is the lowest level, and each level contains the same number of particles;

[0031] Hierarchical learning strategy: Lower-level particles (L3, L4) enhance their exploration capabilities by learning from particles from higher levels (such as L1, L2); for higher-level particles, first-level particles directly enter the next generation, and the two samples of second-level particles are randomly selected from the first level.

[0032] For example, if the total number of particles is 100, they are divided into 4 layers, with 25 particles in each layer. L1 and L2 are higher layers (higher fitness values), and L3 and L4 are lower layers (lower fitness values).

[0033] By employing this hierarchical strategy, the particle swarm optimization algorithm can balance local search and global exploration capabilities, thereby improving overall optimization performance.

[0034] Step 5 includes: from higher levels respectively and Randomly selected particles and As exemplar particles, they are used to update the individual's best pbest and the neighborhood's best nbest, respectively, to guide the learning of lower-level particles.

[0035] Higher-level judgments are based on the fitness ranking of particles. This indicates that in the hierarchical learning-based particle swarm optimization algorithm LPSO, from the first... A level randomly selected from the higher levels. It contains two or more particles, and the position and velocity of each particle are respectively represented by... and Let i represent the particle's number in the swarm (the i-th particle), and j represent the dimension of the problem space (the j-th dimension). express The current position of the i-th particle in the j-th dimension of the solution space. express The velocity of the i-th particle in the j-th dimension (i.e., the step size and direction of the position change); This indicates that in the hierarchical learning-based particle swarm optimization algorithm LPSO, from the first... A level randomly selected from the higher levels. It contains two or more particles, and the position and velocity of each particle are respectively represented by... and express; express The current position of particle i in the j-th dimension of the solution space. express The velocity of the i-th particle in the j-th dimension;

[0036] Indicates from The first one randomly selected from the middle One particle; Indicates from The first one randomly selected from the middle particles; among which , and The index is randomly selected for the particles; superior example particles are used to guide particles toward better solution regions, enhancing exploration capabilities; relatively inferior example particles are used to increase the randomness and diversity of particles, improving exploration capabilities.

[0037] In step 6, the parameters include the number of particles (NP), the number of levels (NL), the level size (LS), and control parameters. Typically, NP=20, NL=4, LS=5. .

[0038] In step 7, the inertia weights of the hierarchical learning-based particle swarm optimization algorithm LPSO are adaptively adjusted using the following formula: ,

[0039] in, Represents particles The inertia weight at iteration d, and These represent the pre-defined minimum and maximum inertia coefficients, typically taken as 0.4 and 0.9 respectively. Let represent the fitness of the particle in the d-th iteration. It is the average fitness of all particles at the d-th iteration. , This represents the maximum fitness of all particles at the d-th iteration.

[0040] Step 8 includes updating the particle velocity and position using the following formula: ,

[0041] ,

[0042] in, It is a random number between 0 and 1. It is a particle The update speed after d iterations It is a particle The position updated after d iterations; This indicates that after d iterations, starting from the th... The first one randomly selected from the 1st level The position vectors of particles, where... It is a randomly selected higher-level index. That is, this level is superior to the i-th level where the current particle is located. From that level The index of a randomly selected particle; This indicates that after d iterations, starting from the th... The first one randomly selected from the 1st level The position vectors of particles, where... It is a randomly selected higher-level index. ,and ), From that level The index of a randomly selected particle, where , and For randomly selected particle indices, and These are two exemplars learned in the current example, each from a different higher level; Indicates from the first The first one randomly selected from the 1st level The position vectors of each particle. Indicates from the first The first one randomly selected from the 1st level The position vectors of each particle. fitness value Greater than or equal to fitness value , Used to enhance the development capabilities of particles. Used to enhance particle detection capabilities;

[0043] These are control parameters for hierarchical learning, ranging between 0 and 1, and influence the second paradigm. .

[0044] Step 9 includes: During the iteration of the LPSO algorithm, the following steps are repeated until the termination condition is met:

[0045] Particle fitness evaluation: For each particle in the current population, the fitness value is calculated using the fitness function. The fitness function takes into account multiple performance indicators, such as the tracking speed of the output voltage, overshoot and static error, and obtains the fitness value of each particle by weighted summation through weighting coefficients.

[0046] Particle stratification: Based on their fitness values, particles are assigned to different strata; particles with higher fitness values ​​are assigned to higher strata (e.g., L1), and particles with lower fitness values ​​are assigned to lower strata (e.g., LNL). The purpose of stratification is to apply differentiated learning strategies at different strata, thereby improving the algorithm's search efficiency and solution quality.

[0047] Paradigm selection and learning: Two particles are randomly selected from higher levels (L1, L2) as paradigm particles, used to update the individual's optimal pbest and the neighborhood's optimal nbest, respectively. The better paradigm particles are used to guide the particles closer to the excellent solution region, enhancing the exploration capability; the relatively worse paradigm particles are used to increase the randomness and diversity of the particles, improving the exploration capability. In this way, the algorithm can achieve a balance between global search and local search.

[0048] Updating particle velocity and position: Based on example particles and control parameters, the velocity and position of each particle are updated. The velocity update formula considers the particle's current velocity, the velocity increment learned from better example particles, and the velocity increment learned from relatively worse example particles. The position update formula adjusts the particle's position based on the updated velocity. By continuously updating the particle's velocity and position, the algorithm can gradually approach the optimal solution in the search space.

[0049] Check the termination condition: After each iteration, check if the termination condition is met. The termination condition can be reaching a preset maximum number of iterations. If the termination condition is met, the iteration process ends; otherwise, continue to the next iteration.

[0050] Step 10 includes:

[0051] Determine the optimal solution: During the iteration process of the LPSO algorithm, record the particle with the best fitness value and the corresponding parameters; the best fitness value indicates that the particle has the best performance after comprehensively considering multiple performance indicators.

[0052] Parameter application: The parameters corresponding to the particles with optimal fitness are extracted and used as the optimized parameters for the energy storage converter controller. These parameters are then applied to the actual energy storage converter system to improve system performance, stability, response speed, and overall efficiency.

[0053] The present invention has the following beneficial effects: the LPSO algorithm significantly improves the global search capability and local exploitation capability of the algorithm through a hierarchical learning strategy, and can find higher quality solutions in complex optimization problems.

[0054] The dynamic sample selection mechanism of the LPSO algorithm increases the randomness and diversity of the algorithm, helps the algorithm escape local optima, and improves global search capabilities.

[0055] The adaptive inertia weight strategy of the LPSO algorithm dynamically adjusts the inertia weights based on the particle's performance, enabling the algorithm to better adapt to different search stages.

[0056] The LPSO algorithm exhibits higher convergence speed and accuracy in the optimization of energy storage converter control systems. It can be adjusted according to the characteristics of specific problems, improving the algorithm's versatility and adaptability, and significantly enhancing the system's stability and response speed. Attached Figure Description

[0057] Figure 1 This is a flowchart of the optimization process for the energy storage converter control system in this invention.

[0058] Figure 2 This is an optimized control system diagram of the energy storage converter in this invention.

[0059] Figure 3 This is a comparison chart of output voltage tracking under different algorithms. Detailed Implementation

[0060] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0061] like Figure 1 As shown in the figure, this embodiment provides an optimization method for an energy storage converter control system based on hierarchical learning, including the following steps:

[0062] Step 1: Collect historical power grid data, including fitting the parameters of the energy storage converter controller. , , , , Various combinations, and corresponding performance metrics (time). ), overshoot ( ), steady-state error ( ),in , , , , The selection of values ​​is random, but these values ​​are approximately in the range of [0.0001, 50].

[0063] Step 2 involves preprocessing the data, including data cleaning, removing unreasonable data, shuffling and rearranging the data, and normalizing the data.

[0064] Step 3: Design the fitness function of the LPSO algorithm. The fitness function includes multiple performance metrics. The detailed steps in this example are as follows:

[0065] Step 3.1, according to the experimental requirements, the parameters include the parameters of the fitted energy storage converter controller. , , , , , Various combinations of time (st), overshoot (ov), and steady-state error (sse).

[0066] Step 3.2 The specific fitness function is as follows:

[0067] ,

[0068] Where k1 and k2 are the weighting coefficients of performance index st, k3 and k4 are the weighting coefficients of performance index ov, and k5 and k6 are the weighting coefficients of performance index sse, with k1=50, k2=0.1, k3=50, k4=70, k5=50, and k6=550.

[0069] Step 4: Determine the fitness of the particles based on the fitness function. Particles are assigned to different levels according to their fitness values. Assume NP particles are divided into NL levels, each containing the same number of particles, called the "Level Size" (LS). The calculation formula can be defined as:

[0070] ,

[0071] Where NP=20, NL=4, LS=5.

[0072] Step 5: Particles are categorized into different levels based on their hierarchical learning of optimal strategies. Particles at different levels receive different treatments by learning from a different number of particles at higher levels. Specifically, the lower the level a particle belongs to, the more candidate examples it can learn. and The more you have, the more you can explore the search space, and vice versa. In this way, lower-level particles can focus on exploring the search space, while higher-level particles can focus on exploring the search space.

[0073] Step 6: Select particles from higher levels as samples to guide lower-level particles. Using two superior particles as examples to guide the learning of lower-level particles enhances exploration capabilities. Unlike the pbest, nbest, or gbest learning in most PSO algorithm variants, the LPSO algorithm directly utilizes two particles randomly selected from two different higher levels to guide particle learning. The two selected example particles are used to replace pbest (individual best) and nbest (neighborhood best) in the Particle Swarm Optimization (PSO) algorithm, respectively. Specifically, of the two randomly selected particles from higher levels, the superior one is used to update pbest, while the relatively inferior one is used to update nbest or gbest.

[0074] Step 7: Optimize the LPSO algorithm using MATLAB code. The particle velocity and position update formulas are as follows: ,

[0075] in, and It is a random number between 0 and 1. It is a particle At the velocity in dimension d, It is a particle The position updated after d iterations. This indicates that after d iterations, starting from the th... The first one randomly selected from the 1st level The position vectors of particles, where... It is a randomly selected higher-level index. That is, this level is superior to the i-th level where the current particle is located. From that level The index of a randomly selected particle; This indicates that after d iterations, starting from the th... The first one randomly selected from the 1st level The position vectors of particles, where... It is a randomly selected higher-level index. ,and ), From that level The index of a randomly selected particle, where , and For randomly selected particle indices, and These are two exemplars learned in the current example, each from a different higher level; Indicates from the first The first one randomly selected from the 1st level The position vectors of each particle. Indicates from the first The first one randomly selected from the 1st level The position vectors of each particle. fitness value Greater than or equal to fitness value , Used to enhance the development capabilities of particles. Used to enhance particle detection capabilities;

[0076] These are control parameters for hierarchical learning, ranging between 0 and 1, and influence the second paradigm. .

[0077] The particle velocity and position update formulas do not directly apply to the updates of particles in the first and second levels. To address this, we employ different additional techniques for the two levels. First, since the particles in the first level are the best particles in the entire population for the current generation, and better solutions are usually found near these particles, we simply keep these particles unchanged to preserve the most useful information and protect them from being diminished. Therefore, particles in the first level directly advance to the next generation. Second, a similar adaptation applies to particles in the second level. Instead of randomly selecting two samples from two higher levels, these two samples are randomly chosen from the first level.

[0078] Step 8, according to the requirements of Step 7, set , And set other parameters: number of particles N=20, number of iterations G=1000.

[0079] Step 9: Evaluate the fitness of each particle and update the global optimal position.

[0080] Step 10, Termination condition determination: The maximum number of evolution iterations set has been reached.

[0081] Step 11: If the convergence condition of the LPSO algorithm is met, output the optimal solution, i.e., the optimal controller. If the convergence condition is not met, return to step 7, update the particle's velocity and position, and continue iterating.

[0082] Step 12: The LPSO algorithm optimization program completes its execution, yielding the optimal parameter combination. =0.0001、 =7.7422、 =29.7730、 =22.6126、 =25.0926、 =50;

[0083] Step 13: Input the optimal parameter combination obtained in Step 12 into the system and compare it with the PSO algorithm. Plot a comparison graph of each performance index under different methods, as shown below. Figure 3 Output voltage tracking comparison chart.

[0084] like Figure 2 As shown in the figure, this embodiment provides an optimization diagram for an energy storage converter control system based on hierarchical learning. The core control structure of the system includes:

[0085] Outer loop DC voltage control: By comparing the actual DC voltage and target voltage The target current is generated by a PI regulator. This is further used as the input for inner loop current control.

[0086] Inner loop current control: using target current and (in ) and actual current and The error of f is used as input to generate the target voltage through an independent PI controller. and .

[0087] Coordinate transformation module:

[0088] Park Transform and Clarke Transform: Transforming Three-Phase Currents , , Converted to DC component and This allows for more efficient control.

[0089] Inverse transformation module: converts the target voltage and Convert to three-phase target voltage , , .

[0090] SPWM modulation module: Generates three-phase PWM signals through space vector pulse width modulation (SPWM) to drive the switching devices of the energy storage converter.

[0091] Intelligent optimization algorithm module: Employs a hierarchical learning-based optimization algorithm (LPSO) to dynamically adjust PI controller parameters, further improving system performance and adapting to complex operating environments.

[0092] like Figure 3 As shown in the figure, this embodiment provides a comparison chart of output voltage tracking under different algorithms. It can be seen from the figure that the LPSO algorithm can approach the steady-state output value relatively quickly in the initial stage. This rapid response capability is due to the hierarchical learning mechanism of the LPSO algorithm, which can more efficiently explore and utilize the search space, thereby obtaining a better solution in a short time.

[0093] In the steady-state phase, the output voltage curve of the LPSO algorithm almost perfectly matches the target value (green dashed line). This indicates that the LPSO algorithm has higher accuracy in tracking the target output and can optimize with small errors. Although the LPSO algorithm curve shows a slight overshoot when reaching the steady-state value, this overshoot is small and decays rapidly, demonstrating its good dynamic response capability. Compared to the Particle Swarm Optimization (PSO) algorithm, this moderate overshoot allows for rapid attainment of the target value while avoiding excessive fluctuations. The hierarchical learning characteristic of the LPSO algorithm allows it to dynamically adjust the learning mode of the swarm during optimization, better balancing global and local search. It is this characteristic that enables the LPSO algorithm to achieve excellent performance in complex optimization problems.

[0094] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned method for optimizing an energy storage converter control system based on hierarchical learning.

[0095] This embodiment also provides a computer device, including:

[0096] Memory is used to store instructions.

[0097] A processor is configured to execute the instructions, causing the computer device to perform the operation of the hierarchical learning-based energy storage converter control system optimization method.

[0098] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0099] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0100] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0101] This invention provides an optimization method for an energy storage converter control system based on hierarchical learning. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. An optimization method for an energy storage converter control system based on hierarchical learning, characterized in that, Includes the following steps: Step 1: Collect historical power grid data and preprocess it; Step 2: Calculate the fitness value using the fitness function and evaluate the fitness value of each particle to measure the performance of each particle; Step 3: Divide the particles into different levels according to their fitness. Divide the particles into low-level particles and high-level particles according to their fitness to form a hierarchical learning structure. The low-level particles focus on exploring new parameter spaces, while the high-level particles focus on developing within the known excellent parameter regions. Step 4: Through a hierarchical learning strategy, lower-level particles enhance their exploration capabilities by learning from higher-level particles; higher-level particles focus on development within the known region of excellent parameters. Step 5: Randomly select two particles from the high level as samples, which are used to update the individual's best pbest and the neighborhood's best nbest, respectively, to guide the learning of low-level particles. Step 6: Set the parameters of the hierarchical learning-based particle swarm optimization algorithm LPSO; Step 7: Adaptively adjust the inertia weights of the hierarchical learning-based particle swarm optimization algorithm LPSO. Step 8: Update particle positions based on system feedback and adjust controller parameters; Step 9: Repeat the evaluation, stratification, learning, and updating process until the optimal solution is found; Step 10: Apply the parameters corresponding to the particles with the best fitness to the energy storage converter system.

2. The method of claim 1, wherein, Step 1 comprises: the power grid historical data comprises voltage V(t) at t moment, current I(t) at t moment and power factor .

3. The method of claim 2, wherein, Step 2 includes calculating the fitness using the following equation : ， wherein, , , are weight coefficients, f1 is a tracking speed index of the output voltage; f2 is an output voltage overshoot; f3 is an output voltage static error.

4. The method of claim 3, wherein, Step 3 includes: In the hierarchical learning-based particle swarm optimization algorithm LPSO, NP particles are divided into NL levels, each containing the same number of particles, called the level size LS, which is calculated using the following formula: ， fitness value of a particle The position of a particle in the hierarchy is determined by the fitness value of the particle, the ith particle The mathematical expression for assigning the ith particle to the lth level is: ， wherein, is the rank of the fitness value of the ith particle in all particles, denotes the ceiling function.

5. The method of claim 4, wherein, In step 4, the LPSO algorithm is layered using a hierarchical learning strategy. The specific steps include: Particle hierarchical partitioning: Particles are sorted from high to low based on their fitness values, and the sorted particles are then divided into partitions from L1 to L2. NL L1 is the highest level, L NL This is the lowest level, and each level contains the same number of particles; Hierarchical learning strategy: Lower-level particles enhance their exploration capabilities by learning from higher-level particles; higher-level particles focus on development within known areas of superior parameters.

6. The method of claim 5, wherein, Step 5 includes: from higher levels respectively and Randomly selected particles and As exemplar particles, they are used to update the individual's best pbest and the neighborhood's best nbest, respectively, to guide the learning of lower-level particles. Higher-level judgments are based on the fitness ranking of particles. This indicates that in the hierarchical learning-based particle swarm optimization algorithm LPSO, from the first... A level randomly selected from the higher levels. It contains two or more particles, and the position and velocity of each particle are respectively represented by... and Let i represent the particle's index in the swarm, and j represent the dimension of the problem space. express The current position of the i-th particle in the j-th dimension of the solution space. express The velocity of the i-th particle in the j-th dimension; This indicates that in the hierarchical learning-based particle swarm optimization algorithm LPSO, from the first... A level randomly selected from the higher levels. It contains two or more particles, and the position and velocity of each particle are respectively represented by... and express, express The current position of particle i in the j-th dimension of the solution space. express The velocity of the i-th particle in the j-th dimension; Indicates from The first one randomly selected from the middle One particle; Indicates from The first one randomly selected from the middle particles; among which , and For randomly selected particle indices; from higher levels Selected These are superior example particles, used to guide particles toward the region of excellent solutions; from a level lower than... of Selected These are relatively poor example particles, used to increase the randomness and diversity of the particles.

7. The method according to claim 6, characterized in that, In step 6, the parameters include the number of particles NP, the number of levels NL, the level size LS, and a control parameter .

8. The method of claim 7, wherein, In step 7, the inertia weights of the hierarchical learning-based particle swarm optimization algorithm LPSO are adaptively adjusted using the following formula: ， in, Represents particles The inertia weight at iteration d, and These represent the pre-defined minimum and maximum inertia coefficients, respectively. Let represent the fitness of the particle in the d-th iteration. It is the average fitness of all particles at the d-th iteration. , This represents the maximum fitness of all particles at the d-th iteration.

9. The method of claim 8, wherein, Step 8 includes updating the particle velocity and position using the following formula: ， ， in, It is a random number between 0 and 1. It is a particle The update speed after d iterations It is a particle The position updated after d iterations; This indicates that after d iterations, starting from the th... The first one randomly selected from the 1st level The position vectors of particles, where... It is a randomly selected higher-level index. , From that level The index of a randomly selected particle; This indicates that after d iterations, starting from the th... The first one randomly selected from the 1st level The position vectors of particles, where... It is a randomly selected higher-level index. ,and , From that level The index of a randomly selected particle, where , and For randomly selected particle indices, and These are two examples learned in the current example, each from a different higher level; Indicates from the first The first one randomly selected from the 1st level The position vectors of each particle. Indicates from the first The first one randomly selected from the 1st level The position vectors of each particle. fitness value Greater than or equal to fitness value , Used to enhance the development capabilities of particles. Used to enhance particle detection capabilities; These are the control parameters for hierarchical learning.

10. The method of claim 9, wherein, Step 9 includes: During the iteration of the LPSO algorithm, the following steps are repeated until the termination condition is met: Evaluate particle fitness: For each particle in the current population, calculate the fitness value using the fitness function; Particle stratification: Based on the particle's fitness value, particles are assigned to different strata; particles with higher fitness values ​​are assigned to higher strata, and particles with lower fitness values ​​are assigned to lower strata. Paradigm selection and learning: Two particles are randomly selected from the higher level as paradigm particles, which are used to update the individual's best pbest and the neighborhood's best nbest, respectively; the better paradigm particles are used to guide the particles to move closer to the excellent solution region and enhance the development capability; the relatively poor paradigm particles are used to increase the randomness and diversity of the particles and improve the exploration capability. Update particle velocity and position: Update the velocity and position of each particle based on the example particle and control parameters; Check the termination condition: After each iteration, check whether the termination condition is met. If the termination condition is met, the iteration process ends; otherwise, continue to the next iteration. Step 10 includes: Determine the optimal solution: During the iteration process of the LPSO algorithm, record the particles with the best fitness value and their corresponding parameters; Parameter application: Extract the parameters corresponding to the particles with the best fitness and use them as the optimized parameters of the energy storage converter controller.

Images