Multi-objective optimization method, system, device and medium for hybrid control system

Abstract

The application discloses a multi-objective optimization method, system, device and medium of a hybrid power control system, relates to the technical field of hybrid electric vehicles, and comprises the following steps: acquiring a fitting data set, the fitting data set comprising a plurality of sample data, each sample data comprising a historical value of an input parameter and a historical value of an output parameter; performing function fitting based on the fitting data set to obtain a multi-objective optimization model; and using a multi-objective optimization algorithm based on a Gaussian mixture model to optimize and solve the multi-objective optimization model to obtain the optimal value of the input parameter, so as to optimize the design of the hybrid power control system. The application can solve the problem that the optimization model established by a traditional method is too complex through function fitting, has high optimization efficiency, and the output parameter comprises a battery voltage, an operation conversion frequency, emission, noise and a battery state of charge, can consider the influence of other factors on the performance of the hybrid power control system, and quickly solves the optimal design scheme of the hybrid power control system.

Description

Technical Field

[0001] This invention relates to the field of hybrid electric vehicle technology, and in particular to a multi-objective optimization method, system, device and medium for a hybrid power control system based on Gaussian mixture model progeny generation. Background Technology

[0002] The power of a hybrid electric vehicle (HEV) is provided by an internal combustion engine (ICE) and an electric motor (EM). In the design process of a hybrid electric vehicle, the energy control manager plays a key role in ensuring the hybrid performance of the hybrid electric vehicle. The hybrid control system is the control component of the energy control manager. Therefore, optimizing the design of the hybrid control system is a key method to ensure the hybrid performance of a hybrid electric vehicle.

[0003] In the design of hybrid power control systems, the primary goal is to reduce fuel consumption. Therefore, most current research focuses on this aspect. To achieve this, classic single-objective optimization methods have been widely applied in the design of hybrid power control systems. These methods include modal analysis, such as finite element analysis and experimental modal analysis. However, the optimization problem of the aforementioned hybrid power control system is a black-box problem, where the specific functional expression is unknown. While modal analysis can approximate the actual physical model well, the established optimization model is complex and the solution is very time-consuming. Therefore, existing single-objective optimization methods have low optimization efficiency and do not consider the influence of other factors on the performance of the hybrid power control system. Consequently, existing single-objective optimization methods cannot obtain the optimal design solution that meets the requirements.

[0004] Therefore, there is an urgent need for a multi-objective optimization technique with high optimization efficiency that can quickly solve for the optimal design solution. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-objective optimization method, system, device, and medium for a hybrid power control system. This method has high optimization efficiency and, considering the influence of other factors on the performance of the hybrid power control system, can quickly solve for the optimal design scheme.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A multi-objective optimization method for a hybrid power control system, the multi-objective optimization method comprising:

[0008] Obtain a fitted dataset; the fitted dataset includes multiple sample data, each sample data includes historical values ​​of input parameters and historical values ​​of output parameters corresponding to the historical values ​​of input parameters; the input parameters include the maximum battery state of charge threshold for turning off the internal combustion engine, the minimum battery state of charge threshold for turning on the internal combustion engine, the minimum operating speed, the maximum operating speed, the internal combustion engine speed for the first operation, the internal combustion engine speed for the second operation, the internal combustion engine speed for the third operation, the torque value for the first operation, the torque value for the second operation, the torque value for the third operation, and the speed at which the internal combustion engine is turned off; the output parameters include battery voltage, number of operation transitions, emissions, noise, and battery state of charge;

[0009] Based on the fitted dataset, a function fitting is performed to obtain a multi-objective optimization model that characterizes the functional relationship between the input parameters and the output parameters;

[0010] The multi-objective optimization model is optimized and solved using a multi-objective optimization algorithm based on Gaussian mixture model to obtain the optimal values ​​of the input parameters.

[0011] In some embodiments, the function fitting based on the fitted dataset specifically includes: performing function fitting using a Kriging model based on the fitted dataset.

[0012] In some embodiments, the multi-objective optimization algorithm based on Gaussian mixture model specifically includes: generating a subpopulation based on Gaussian mixture model every k iterations.

[0013] In some embodiments, the step of using a multi-objective optimization algorithm based on a Gaussian mixture model to optimize and solve the multi-objective optimization model to obtain the optimal values ​​of the input parameters specifically includes:

[0014] An initial population is randomly generated; the initial population includes multiple individuals, each of which represents a value of the input parameter;

[0015] Determine if the number of iterations has reached k; if yes, set the number of iterations to 0 and generate the offspring population based on the Gaussian mixture model; if no, increment the number of iterations by 1 and generate the offspring population based on crossover mutation; the initial value of the number of iterations is 0.

[0016] The offspring population and the initial population are merged to obtain a merged population, and a number of individuals are selected from the merged population to form a selected population.

[0017] Determine whether the iteration termination condition has been met; if yes, end the iteration and determine the optimal value of the input parameter based on the selected population; if no, continue the iteration, using the selected population as the initial population for the next iteration, and return to the step of "determining whether the number of iterations has reached k times".

[0018] In some embodiments, the random generation of the initial population specifically includes:

[0019] Randomly select a value from the range of values ​​for each input parameter to form a value for the input parameter, thus obtaining an individual;

[0020] Determine if N individuals have been obtained; if yes, then the N individuals form the initial population; otherwise, return to the step of "randomly selecting a value within the range of each input parameter".

[0021] In some embodiments, generating the offspring population based on the Gaussian mixture model specifically includes:

[0022] Construct the current Gaussian sub-model based on the initial population;

[0023] Using the current Gaussian sub-model and the historical Gaussian sub-model as input, the weight parameters are calculated using the EM algorithm; the historical Gaussian sub-model is the current Gaussian sub-model constructed during the historical iteration process.

[0024] The current Gaussian sub-model and the historical Gaussian sub-model are fused according to the weight parameters to obtain a Gaussian mixture model;

[0025] The probability distribution of the Gaussian mixture model is downsampled to generate a progeny population.

[0026] In some embodiments, selecting a number of individuals from the merged population to form the selected population specifically includes:

[0027] For each individual in the merged population, calculate the distance between the individual and each remaining individual in the merged population, sort the distances in descending order, and select the distance at position A in the sort as the neighbor distance of the individual.

[0028] Each individual in the merged population is sorted in descending order of neighbor distance, and the individuals ranked in the top N positions are selected to form the selected population.

[0029] A multi-objective optimization system for a hybrid power control system, the multi-objective optimization system comprising:

[0030] A construction module is used to acquire a fitting dataset; the fitting dataset includes multiple sample data, each sample data including historical values ​​of input parameters and historical values ​​of output parameters corresponding to the historical values ​​of the input parameters; the input parameters include the maximum threshold of battery state of charge when the internal combustion engine is off, the minimum threshold of battery state of charge when the internal combustion engine is on, the minimum operating speed, the maximum operating speed, the internal combustion engine speed for the first operation, the internal combustion engine speed for the second operation, the internal combustion engine speed for the third operation, the torque value for the first operation, the torque value for the second operation, the torque value for the third operation, and the speed at which the internal combustion engine is off; the output parameters include battery voltage, number of operation transitions, emissions, noise, and battery state of charge;

[0031] The fitting module is used to perform function fitting based on the fitting dataset to obtain a multi-objective optimization model that characterizes the functional relationship between the input parameters and the output parameters.

[0032] The optimization module is used to optimize the multi-objective optimization model using a multi-objective optimization algorithm based on Gaussian mixture model to obtain the optimal values ​​of the input parameters.

[0033] A multi-objective optimization device for a hybrid power control system includes:

[0034] Processor; and

[0035] Memory, in which computer-readable program instructions are stored.

[0036] The multi-objective optimization method described above is executed when the computer-readable program instructions are run by the processor.

[0037] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described multi-objective optimization method.

[0038] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0039] This invention provides a multi-objective optimization method, system, device, and medium for a hybrid power control system. First, a fitting dataset is obtained, comprising multiple sample data points. Each sample data point includes historical values ​​of input and output parameters. Then, a function is fitted based on the fitting dataset to obtain a multi-objective optimization model. Finally, a multi-objective optimization algorithm based on a Gaussian mixture model is used to optimize and solve the multi-objective optimization model, obtaining the optimal values ​​of the input parameters to optimize the design of the hybrid power control system. This invention, through function fitting, solves the problem of overly complex optimization models established by traditional methods, achieving high optimization efficiency. Furthermore, the output parameters include battery voltage, number of operation transitions, emissions, noise, and battery state of charge, taking into account the impact of other factors on the performance of the hybrid power control system and quickly obtaining the optimal design scheme for the hybrid power control system. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A schematic diagram of the energy control manager provided by the present invention;

[0042] Figure 2 This is a flowchart of the multi-objective optimization method provided in Embodiment 1 of the present invention;

[0043] Figure 3 This is a schematic diagram of the multi-objective optimization method provided in Embodiment 1 of the present invention;

[0044] Figure 4 This is a schematic diagram of the solution process of the multi-objective optimization algorithm provided in Embodiment 1 of the present invention;

[0045] Figure 5 This is a system block diagram of the multi-objective optimization system provided in Embodiment 2 of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] The purpose of this invention is to provide a multi-objective optimization method, system, device, and medium for a hybrid power control system. This method has high optimization efficiency and, considering the influence of other factors on the performance of the hybrid power control system, can quickly solve for the optimal design scheme.

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] like Figure 1 The diagram shows the energy control manager of a hybrid electric vehicle, which mainly consists of an internal combustion engine, a hybrid power control system, a battery, and an electric motor. Before the hybrid electric vehicle starts moving, the battery is charged via the power grid. During braking, the electric motor acts as a generator, recovering the energy generated during braking back into the battery. The energy control manager uses the battery's state of charge (SOC(t)) at time t as input to the hybrid power control system to determine its specific control method. The implementation of this specific control method affects the hybrid performance of the hybrid electric vehicle; therefore, optimizing the hybrid power control system is a key method to ensure the hybrid performance of the hybrid electric vehicle. Considering that vehicle speed and torque are controlled by the driver or external devices, this invention only considers the optimization of the control management of the internal combustion engine (i.e., the ICE part).

[0050] Existing single-objective optimization methods for optimizing hybrid power control systems mainly employ modal analysis methods, such as finite element analysis, to simulate and model the actual physical model of the optimization problem. However, the optimization models established by these methods are complex, and the solution cost is very high, especially the computation of the objective function, which is extremely time-consuming. This poses a severe challenge to solving such problems using general solvers. Furthermore, existing single-objective optimization methods often only consider designing a hybrid power control system with the single objective of minimizing fuel consumption, neglecting the impact of other objectives on the performance of the hybrid power control system, thus failing to obtain the optimal solution set. Therefore, the purpose of this invention is to provide a knowledge-learning-based multi-objective optimization technique to address the problems of time-consuming function computation and the inability to obtain the optimal solution set inherent in single-objective optimization methods such as finite element analysis.

[0051] Example 1:

[0052] This embodiment provides a multi-objective optimization method for a hybrid power control system, such as... Figure 2 and Figure 3 As shown, the multi-objective optimization method includes:

[0053] S1: Obtain the fitted dataset; the fitted dataset includes multiple sample data, each sample data includes historical values ​​of input parameters and historical values ​​of output parameters corresponding to the historical values ​​of input parameters; the input parameters include the maximum battery state of charge threshold for turning off the internal combustion engine, the minimum battery state of charge threshold for turning on the internal combustion engine, the minimum operating speed, the maximum operating speed, the internal combustion engine speed for the first operation, the internal combustion engine speed for the second operation, the internal combustion engine speed for the third operation, the torque value for the first operation, the torque value for the second operation, the torque value for the third operation, and the speed at which the internal combustion engine is turned off; the output parameters include battery voltage, number of operation switching times, emissions, noise, and battery state of charge;

[0054] Generally, hybrid power control systems need to operate according to the following six rules, specifically:

[0055] Rule 1: If the speed is lower than V off Then turn off the internal combustion engine.

[0056] Rule 2: If the battery state of charge is greater than SOC max Then turn off the internal combustion engine. Note: SOC max It is not the maximum value of the battery's state of charge, but rather the maximum threshold value at which the battery stops charging.

[0057] Rule 3: If the battery state of charge is less than SOC min Then turn off the internal combustion engine. Note: Rule 3 is more important than Rule 1.

[0058] Rule 4: If the internal combustion engine is on and the speed is less than V1, then perform operation 1.

[0059] Rule 5: If the internal combustion engine is on and the speed is between V1 and V2, then perform operation 2.

[0060] Rule 6: If the internal combustion engine is on and the speed is greater than V2, then perform operation 3.

[0061] Based on the aforementioned rules required for the operation of a hybrid power control system, the types and value ranges of the input parameters selected in this embodiment for optimizing the hybrid power control system are shown in Table 1 below:

[0062] Table 1

[0063]

[0064] It should be noted that, in actual practice, the operation using the internal combustion engine speed and torque value of operation 1 as defined in Table 1 is the first operation described in this embodiment; the operation using the internal combustion engine speed and torque value of operation 2 as defined in Table 1 is the second operation described in this embodiment; and the operation using the internal combustion engine speed and torque value of operation 3 as defined in Table 1 is the third operation described in this embodiment.

[0065] Considering that in addition to fuel consumption, the performance of the actual hybrid power control system also needs to take into account other factors, such as driving experience, noise, vibration, and acoustic roughness, the output parameters used in this embodiment for optimizing the design of the hybrid power control system are shown in Table 2 below:

[0066] Table 2

[0067] Target number Physical meaning of the target Target 1 Battery stress (battery voltage, BS) Target 2 Operation changes (OPC) Target 3 Emission Target 4 Noise Target 5 Battery state of charge (SOC)

[0068] The purpose of this embodiment is to obtain the optimal values ​​of the 11 input parameters (i.e., design variables) in Table 1 when the five output parameters (objectives) in Table 2 achieve the overall optimality through multi-objective optimization. To achieve this, experiments are conducted on the hybrid power control system. First, the historical values ​​of the 11 input parameters are determined. Then, the historical values ​​of the five output parameters under the historical values ​​of the 11 input parameters are determined, resulting in a sample data set. The experimental process is repeated multiple times to obtain multiple sample data sets, which together form the fitting dataset. In practical applications, the sample data can come from the sampling data obtained by the Design of Experiment (DOE). One sample data set includes the corresponding historical values ​​of the 11 input parameters and the historical values ​​of the five output parameters.

[0069] S2: Based on the fitted dataset, perform function fitting to obtain a multi-objective optimization model that characterizes the functional relationship between the input parameters and the output parameters;

[0070] Because the optimization problem involved in the design of the hybrid electric vehicle's hybrid power control system is a black-box optimization problem, the objective function is difficult to express mathematically analytically and even lacks directly usable gradient information, resulting in very time-consuming calculation of the objective function. To solve this problem, this embodiment determines the objective function through function fitting to construct a multi-objective optimization model. Specifically, after obtaining the fitting dataset using S1, this embodiment uses the fitting dataset as input and performs function fitting using any existing function fitting method. This yields five function expressions with 11 input parameters as independent variables and each output parameter as a dependent variable, thus constructing the multi-objective optimization model. More specifically, this embodiment can use the Kriging model for function fitting. Therefore, in S2, function fitting based on the fitting dataset can include: performing function fitting using the Kriging model based on the fitting dataset. Of course, this embodiment is not limited to the Kriging model and can also use other methods for function fitting.

[0071] Kriging models predict the function value at an estimated point by calculating the weighted average of the function values ​​at multiple observation points in the neighborhood of the estimated point. This method is mathematically closely related to regression analysis. Kriging interpolation is a linear, unbiased, optimal estimation interpolation method that fully considers not only the positional relationship between the observation point and the estimated point but also their spatial relationship. For the five output parameters in this embodiment, the process of fitting the output parameter to the eleven input parameters to establish the functional expression between the output parameter and the eleven input parameters is consistent. Below, this embodiment uses any output parameter as an example to discuss the function fitting process in detail:

[0072] For the regional change Z(x), where Z(x1), Z(x2), ... Z(x... n Let x1, x2, ..., xn be the true function values ​​(i.e., the historical values ​​of the output parameters corresponding to the observation points) for n observation points (each observation point corresponds to a sample data point in the fitted dataset). n Let the coordinates of the n observation points be (i.e., the historical values ​​of the 11 input parameters corresponding to the observation points). Then, the estimated function value Z(x0) of the estimated point x0 at any location within the fitting region can be estimated using a linear combination, namely:

[0073]

[0074] In equation (1), Z(x0) is the estimated function value of the estimated point, x0 represents the estimated point; n is the number of observation points, that is, the number of sample data in the fitted dataset obtained by S1; λ i Z(x) represents the weighting coefficients used when estimating the estimated point from the i-th observation point;i ) represents the true function value (i.e., the historical value of the output parameter) at the i-th observation point, x i This represents the historical value of the i-th observation point (i.e., the 11 input parameters).

[0075] The unknown values ​​in equation (1) only include the weighting coefficients. Assuming that the regional variation Z(x) satisfies second-order continuity, according to the principle of unbiased estimators and minimum estimation variance in Kriging interpolation, the weighting coefficients can be calculated by constructing Lagrange conditional extrema. The constructed Lagrange equation is as follows:

[0076]

[0077] In equation (2), γ(x) i x j ) represents the value of the variation function, indicating x i and x j Distance, x j Represents any observation point other than the i-th observation point; μ is the Lagrange daily number; γ(x) i x0) is the value of the variogram, representing x i Distance to x0.

[0078] The formula for calculating the value of the variation function is:

[0079]

[0080] In equation (3), γ(h) is the variogram value at lag distance h; N(h) is the total number of all observation point pairs with a spacing of h.

[0081] To estimate the unknown value of the regional variation, a theoretical model of the variogram needs to be fitted based on the measured values. Commonly used variogram models include the exponential model, the spherical model, the Gaussian model, and the power function model. Among them, the Gaussian model can realize the mapping of the input vector to the nonlinear space and accurately fit the local properties of the function. Therefore, the variogram model used in this embodiment is the Gaussian model. The formula of the Gaussian model is:

[0082]

[0083] In equation (4), C0 is the nugget constant, C is the arch height, C0+C is the sill value, and a is the range.

[0084] The standardized form of equation (4) is:

[0085]

[0086] In this embodiment, after obtaining the fitted dataset, the distance between the historical values ​​of the input parameters in any two sample data is first calculated. This distance can be Euclidean distance, resulting in multiple distances. This distance is used as the lag distance. For each distance, multiple observation point pairs (including two sample data whose historical values ​​of the input parameters are at the same distance) are selected and substituted into equation (3) to obtain the variogram value corresponding to the distance (i.e., the lag distance). This process is repeated to obtain the variogram value corresponding to each distance. Then, the variogram value corresponding to each distance is substituted into equation (4) or equation (5) for fitting to obtain the variogram model. After determining the coordinates of the estimated point, the variogram value in equation (2) is calculated using the variogram model, and the weighting coefficients are obtained. Then, the weighting coefficients are substituted into equation (1) to obtain the estimated function value of the estimated point.

[0087] This embodiment uses the Kriging model to fit the function to the multi-objective black-box optimization problem, obtaining a multi-objective optimization model. The multi-objective optimization model includes 5 function expressions, each representing the function expression between an output parameter and 11 input parameters. The multi-objective optimization model is used to evaluate the individual (including the values ​​of the 11 input parameters) and obtain the objective function value of the individual (including the values ​​of the 5 output parameters), thus solving the complex problem of solving the optimization model of the hybrid power control system.

[0088] S3: The multi-objective optimization model is optimized and solved using a multi-objective optimization algorithm based on Gaussian mixture model to obtain the optimal values ​​of the input parameters.

[0089] In this embodiment, the multi-objective optimization algorithm based on Gaussian mixture model specifically includes: generating a progeny population based on Gaussian mixture model every k iterations, thereby accelerating the convergence speed of the multi-objective optimization algorithm and improving optimization efficiency.

[0090] like Figure 4 As shown, S3 may include:

[0091] (1) Randomly generate an initial population, which includes multiple individuals, each of which represents a value of the input parameter.

[0092] In this embodiment, the population is initialized based on the upper and lower bounds of each input parameter (i.e., the value range of each input parameter in Table 1). N individuals that satisfy the upper and lower bounds are randomly generated to form the initial population, where N can be 100. Specifically, randomly generating the initial population may include: randomly selecting a value within the value range of each input parameter to form one value of the input parameter, thus obtaining one individual; determining whether N individuals have been obtained; if so, the N individuals form the initial population; if not, returning to the step of "randomly selecting a value within the value range of each input parameter" until N individuals are obtained.

[0093] (2) Determine whether the number of iterations has reached k; if yes, set the number of iterations to 0 and generate the offspring population based on the Gaussian mixture model; if no, increment the number of iterations by 1 and generate the offspring population based on crossover mutation; the initial value of the number of iterations is 0.

[0094] (2) This is the offspring population generation step. The multi-objective optimization algorithm based on Gaussian mixture model used in this embodiment improves this offspring population generation step by introducing an offspring generation strategy based on Gaussian mixture model. In order to increase the accuracy of model prediction, this embodiment controls the frequency of using the offspring generation strategy based on Gaussian mixture model by the number of iterations. The offspring population is generated once every k iterations using the offspring generation strategy based on Gaussian mixture model. In other iterations, algorithms such as crossover and mutation can be used to generate the offspring population.

[0095] In this embodiment, a threshold Δ can be set for the number of iterations. Assuming the termination condition is M iterations, the threshold for the number of iterations used in the offspring generation strategy of the Gaussian mixture model is... The value of k can be determined by presetting a threshold number of attempts.

[0096] To accelerate the convergence speed of the multi-objective optimization algorithm, this embodiment fully utilizes the optimal solution set (i.e., the selected population) generated by the iteration of the multi-objective optimization algorithm to learn about the distribution of the optimal solution set. Specifically, this embodiment uses the optimal solution set generated during the iteration process to train a Gaussian mixture model, and uses the trained Gaussian mixture model to generate the offspring population, thereby promoting population convergence. Generating the offspring population based on the Gaussian mixture model may include:

[0097] 1) Construct the current Gaussian sub-model based on the initial population.

[0098] The Gaussian sub-model is a Gaussian distribution. The characteristics of the Gaussian distribution (including the mean and covariance) are calculated from the optimal solution set generated during the iteration process (i.e., the post-selection population generated in the previous iteration, including multiple optimal values ​​of the 11 input parameters). After determining the characteristics of the Gaussian distribution, the current Gaussian sub-model can be constructed. The current Gaussian sub-model follows the following probability density function:

[0099]

[0100] In equation (6), x is a variable (i.e., 11 input parameters); D is the dimension of the variable, which is 11 in this embodiment; Σ is the covariance; and μ is the mean.

[0101] 2) Using the current Gaussian sub-model and the historical Gaussian sub-model as input, the weight parameters are calculated using the EM algorithm. The historical Gaussian sub-model is the current Gaussian sub-model constructed during the historical iteration process.

[0102] In this embodiment, a child population is generated based on the Gaussian mixture model every k intervals. This is equivalent to generating a current Gaussian sub-model using 1) every k intervals. Here, the historical Gaussian sub-model refers to the previously generated current Gaussian sub-model. For example, if the current generation of the child population is the 3rd time using the Gaussian mixture model, there will be two historical Gaussian sub-models. However, if the current generation of the child population is the 1st time using the Gaussian mixture model, there are no historical Gaussian sub-models, and 2) and 3) are no longer needed. Instead, the current Gaussian sub-model obtained in 1) is directly used as the Gaussian mixture model in 4), and 4) is executed.

[0103] The weight parameters can be evaluated using maximum likelihood estimation. Specifically, the maximum likelihood function for a Gaussian mixture model is as follows:

[0104]

[0105] In equation (7), the parameter , respectively, represent the expectation, covariance, and probability of occurrence in the Gaussian mixture model for each sub-model (including the current Gaussian sub-model and historical Gaussian sub-models); N is the number of observation data (i.e., an optimal value for the 11 input parameters in the optimal solution set); x j For the j-th observation; K is the number of sub-models; α k The probability that the observed data belongs to the k-th sub-model, which is also the weight parameter described in this embodiment; φ(x|θ) k ) is the probability distribution function of the kth sub-model, which is calculated based on the probability density function of the kth sub-model.

[0106] Since it is not known in advance which sub-model each observation belongs to, directly taking the derivative of the maximum likelihood function cannot solve for the weight parameters. In this embodiment, the optimal weight parameters are calculated using the expectation-maximization (EM) algorithm. The EM algorithm is an existing technology, and the solution process will not be described in detail here.

[0107] 3) The current Gaussian sub-model and the historical Gaussian sub-model are fused according to the weight parameters to obtain the Gaussian mixture model.

[0108] Gaussian Mixture Model (GMM) is a popular unsupervised learning algorithm. A GMM can be viewed as a model composed of K sub-models, which are the latent variables of the GMM. The probability distribution of the GMM is shown below:

[0109]

[0110] 4) Downsample the probability distribution of the Gaussian mixture model to generate the offspring population.

[0111] In this embodiment, a Gaussian mixture model is used to approximate the distribution of the optimal solution set. Then, the probability distribution of the Gaussian mixture model is downsampled to generate a progeny population. Specifically, downsampling Equation (8) generates multiple individuals, thus creating a progeny population. Each individual in the progeny population also includes one value of each of the 11 input parameters.

[0112] (3) Merge the offspring population and the initial population to obtain the merged population, and select a number of individuals from the merged population to form the selected population.

[0113] In this embodiment, the generated offspring population and the initial population are merged. Both the initial population and the offspring population can include 100 individuals, resulting in a merged population of 200 individuals. Then, 100 superior individuals are selected from these 200 individuals to form the selected population. A superior individual is defined as the neighbor distance calculated for each individual in the target space. The neighbor distances of the 200 individuals are then sorted from largest to smallest, with higher-ranked individuals being more superior. For each individual, the neighbor distance is calculated as follows: the distances from individual 1 to the other 199 individuals are calculated in the target space, generating 199 distances. These 199 distances are then sorted from largest to smallest, and the A-th furthest distance is selected as the neighbor distance of individual 1. The distance is calculated as follows: using the individual as an input parameter, the multi-objective optimization model is used to calculate the five output parameters corresponding to that individual. Each individual corresponds to five output parameters, and the distance between two individuals is the Euclidean distance between the five output parameters corresponding to those two individuals. Of course, other methods can also be used for selection in this embodiment.

[0114] When using neighbor distance for selection, selecting several individuals from the merged population to form the selected population may include: for each individual in the merged population, calculating the distance between that individual and every remaining individual in the merged population, sorting the distances in descending order, and selecting the distance at the Ath position in the sort as the neighbor distance of that individual; sorting each individual in the merged population in descending order of neighbor distance, and selecting the individuals at the top N positions in the sort to form the selected population, where N can be 100.

[0115] (4) Determine whether the iteration termination condition has been met; if yes, end the iteration and determine the optimal value of the input parameters based on the selected population; if no, continue the iteration, use the selected population as the initial population for the next iteration, and return to the step of "determine whether the number of iterations has reached k".

[0116] In this embodiment, the iteration termination condition can be reaching the maximum number of iterations. Determining the optimal value of the input parameters based on the selected population can include: arbitrarily selecting an individual from the selected population as the optimal value of the input parameters, or selecting an individual from the selected population as the optimal value of the input parameters as needed. After obtaining the optimal value of the input parameters, the hybrid power control system can be optimized.

[0117] This embodiment addresses the complex optimization problem involved in the design of a hybrid power control system by proposing a multi-objective optimization method based on a Gaussian mixture model. It establishes a multi-objective optimization model by fitting a function to a fitted dataset, and simultaneously trains the Gaussian mixture model to fit the distribution of the optimal solution set. A progeny generation strategy based on the Gaussian mixture model is designed, and a multi-objective optimization algorithm based on the Gaussian mixture model is constructed to optimize and solve the multi-objective optimization model, obtaining the optimal values ​​for the 11 input parameters of the hybrid power control system, thus yielding the optimal design scheme for the hybrid power control system. This multi-objective optimization method eliminates the need for establishing a complex optimization model and improves the multi-objective optimization algorithm, resulting in high optimization efficiency. Furthermore, it considers the influence of other factors on the performance of the hybrid power control system, enabling rapid solution to obtain the optimal design scheme for the hybrid power control system.

[0118] Example 2:

[0119] This embodiment provides a multi-objective optimization system for a hybrid power control system, such as... Figure 5 As shown, the multi-objective optimization system includes:

[0120] Module M1 is used to acquire a fitting dataset. The fitting dataset includes multiple sample data, each of which includes historical values ​​of input parameters and historical values ​​of output parameters corresponding to the historical values ​​of the input parameters. The input parameters include the maximum battery state of charge threshold for turning off the internal combustion engine, the minimum battery state of charge threshold for turning on the internal combustion engine, the minimum operating speed, the maximum operating speed, the internal combustion engine speed for the first operation, the internal combustion engine speed for the second operation, the internal combustion engine speed for the third operation, the torque value for the first operation, the torque value for the second operation, the torque value for the third operation, and the speed at which the internal combustion engine is turned off. The output parameters include battery voltage, number of operation transitions, emissions, noise, and battery state of charge.

[0121] The fitting module M2 is used to perform function fitting based on the fitting dataset to obtain a multi-objective optimization model that characterizes the functional relationship between the input parameters and the output parameters.

[0122] The optimization module M3 is used to optimize the multi-objective optimization model using a multi-objective optimization algorithm based on Gaussian mixture model to obtain the optimal values ​​of the input parameters.

[0123] Example 3:

[0124] This embodiment provides a multi-objective optimization device for a hybrid power control system, including:

[0125] Processor; and

[0126] Memory, in which computer-readable program instructions are stored.

[0127] The computer-readable program instructions are executed by the processor to perform the multi-objective optimization method as described in Example 1.

[0128] Example 4:

[0129] This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the multi-objective optimization method described in Embodiment 1.

[0130] Each embodiment in this specification focuses on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be found in the method section.

[0131] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A multi-objective optimization method for a hybrid power control system, characterized in that, The multi-objective optimization method includes: Obtain a fitted dataset; the fitted dataset includes multiple sample data, each sample data includes historical values ​​of input parameters and historical values ​​of output parameters corresponding to the historical values ​​of input parameters; the input parameters include the maximum battery state of charge threshold for turning off the internal combustion engine, the minimum battery state of charge threshold for turning on the internal combustion engine, the minimum operating speed, the maximum operating speed, the internal combustion engine speed for the first operation, the internal combustion engine speed for the second operation, the internal combustion engine speed for the third operation, the torque value for the first operation, the torque value for the second operation, the torque value for the third operation, and the speed at which the internal combustion engine is turned off; the output parameters include battery voltage, number of operation transitions, emissions, noise, and battery state of charge; Based on the fitted dataset, a function fitting is performed to obtain a multi-objective optimization model that characterizes the functional relationship between the input parameters and the output parameters; The multi-objective optimization model is optimized and solved using a multi-objective optimization algorithm based on a Gaussian mixture model to obtain the optimal values ​​of the input parameters; The step of using a multi-objective optimization algorithm based on a Gaussian mixture model to optimize and solve the multi-objective optimization model to obtain the optimal values ​​of the input parameters specifically includes: An initial population is randomly generated; the initial population includes multiple individuals, each of which represents a value of the input parameter; Determine if the number of iterations has reached k; if yes, set the number of iterations to 0 and generate the offspring population based on the Gaussian mixture model; if no, increment the number of iterations by 1 and generate the offspring population based on crossover mutation; the initial value of the number of iterations is 0. The offspring population and the initial population are merged to obtain a merged population, and a number of individuals are selected from the merged population to form a selected population. Determine whether the iteration termination condition has been met; if yes, end the iteration and determine the optimal value of the input parameter based on the selected population; if no, continue the iteration, using the selected population as the initial population for the next iteration, and return to the step of "determining whether the number of iterations has reached k times". The generation of offspring population based on the Gaussian mixture model specifically includes: Construct the current Gaussian sub-model based on the initial population; Using the current Gaussian sub-model and the historical Gaussian sub-model as input, the weight parameters are calculated using the EM algorithm; the historical Gaussian sub-model is the current Gaussian sub-model constructed during the historical iteration process. The current Gaussian sub-model and the historical Gaussian sub-model are fused according to the weight parameters to obtain a Gaussian mixture model; The probability distribution of the Gaussian mixture model is downsampled to generate a progeny population.

2. The multi-objective optimization method according to claim 1, characterized in that, The function fitting based on the fitted dataset specifically includes: performing function fitting using a Kriging model based on the fitted dataset.

3. The multi-objective optimization method according to claim 1, characterized in that, The multi-objective optimization algorithm based on Gaussian mixture model specifically includes: generating a sub-population based on Gaussian mixture model every k iterations.

4. The multi-objective optimization method according to claim 1, characterized in that, The random generation of the initial population specifically includes: Randomly select a value from the range of values ​​for each input parameter to form a value for the input parameter, thus obtaining an individual; Determine if N individuals have been obtained; if yes, then the N individuals form the initial population; otherwise, return to the step of "randomly selecting a value within the range of each input parameter".

5. The multi-objective optimization method according to claim 1, characterized in that, The step of selecting a number of individuals from the merged population to form the selected population specifically includes: For each individual in the merged population, calculate the distance between the individual and each remaining individual in the merged population, sort the distances in descending order, and select the distance at position A in the sort as the neighbor distance of the individual. Each individual in the merged population is sorted in descending order of neighbor distance, and the individuals ranked in the top N positions are selected to form the selected population.

6. A multi-objective optimization system for a hybrid power control system, characterized in that, The multi-objective optimization system includes: A construction module is used to acquire a fitting dataset; the fitting dataset includes multiple sample data, each sample data including historical values ​​of input parameters and historical values ​​of output parameters corresponding to the historical values ​​of the input parameters; the input parameters include the maximum threshold of battery state of charge when the internal combustion engine is off, the minimum threshold of battery state of charge when the internal combustion engine is on, the minimum operating speed, the maximum operating speed, the internal combustion engine speed for the first operation, the internal combustion engine speed for the second operation, the internal combustion engine speed for the third operation, the torque value for the first operation, the torque value for the second operation, the torque value for the third operation, and the speed at which the internal combustion engine is off; the output parameters include battery voltage, number of operation transitions, emissions, noise, and battery state of charge; The fitting module is used to perform function fitting based on the fitting dataset to obtain a multi-objective optimization model that characterizes the functional relationship between the input parameters and the output parameters. The optimization module is used to optimize and solve the multi-objective optimization model using a multi-objective optimization algorithm based on a Gaussian mixture model, so as to obtain the optimal values ​​of the input parameters; The step of using a multi-objective optimization algorithm based on a Gaussian mixture model to optimize and solve the multi-objective optimization model to obtain the optimal values ​​of the input parameters specifically includes: An initial population is randomly generated; the initial population includes multiple individuals, each of which represents a value of the input parameter; Determine if the number of iterations has reached k; if yes, set the number of iterations to 0 and generate the offspring population based on the Gaussian mixture model; if no, increment the number of iterations by 1 and generate the offspring population based on crossover mutation; the initial value of the number of iterations is 0. The offspring population and the initial population are merged to obtain a merged population, and a number of individuals are selected from the merged population to form a selected population. Determine whether the iteration termination condition has been met; if yes, end the iteration and determine the optimal value of the input parameter based on the selected population; if no, continue the iteration, using the selected population as the initial population for the next iteration, and return to the step of "determining whether the number of iterations has reached k times". The generation of offspring population based on the Gaussian mixture model specifically includes: Construct the current Gaussian sub-model based on the initial population; Using the current Gaussian sub-model and the historical Gaussian sub-model as input, the weight parameters are calculated using the EM algorithm; the historical Gaussian sub-model is the current Gaussian sub-model constructed during the historical iteration process. The current Gaussian sub-model and the historical Gaussian sub-model are fused according to the weight parameters to obtain a Gaussian mixture model; The probability distribution of the Gaussian mixture model is downsampled to generate a progeny population.

7. A multi-objective optimization device for a hybrid power control system, characterized in that, include: processor; as well as Memory, in which computer-readable program instructions are stored. The multi-objective optimization method as described in any one of claims 1-5 is executed when the computer-readable program instructions are run by the processor.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the multi-objective optimization method according to any one of claims 1-5.

Images