Method, device, equipment, medium and program product for determining suspension system parameters

Abstract

The application provides a method, device, equipment, medium and program product for determining suspension system parameters. In the method, after obtaining an initial crossover rate, an initial mutation rate and a multi-objective optimization model of a suspension system, an initial population is determined according to a constraint condition of the multi-objective optimization model, and then the multi-objective optimization model is solved according to a non-dominated sorting genetic algorithm, an algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate to obtain target suspension system parameters. The algorithm parameter dynamic optimization model is a preselected deep Q network model for determining a crossover rate increment and a mutation rate increment according to state data of the population. In the solving process, the algorithm parameter dynamic optimization model dynamically adjusts the crossover rate and the mutation rate of the non-dominated sorting genetic algorithm according to the state data of the population, which can improve the solving efficiency and thus improve the efficiency of determining the suspension system parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicles, and in particular to a method, device, equipment, medium and program product for determining parameters of a suspension system. BACKGROUND

[0002] The suspension system is used to isolate the vibration and noise transmission between the power assembly, the vehicle body and the chassis, and can improve the comfort, reliability and durability of the vehicle. In order to design a suspension system with high isolation effect and structural strength, the parameters of the suspension system need to be determined, and then the suspension system is constructed.

[0003] In the prior art, the parameters of the suspension system are usually determined by constructing a multi-objective optimization model according to the parameters of the suspension system, and then solving the multi-objective optimization model to obtain the parameters of the suspension system.

[0004] However, the parameters in the algorithm used to solve the multi-objective optimization model are fixed, resulting in low efficiency in determining the parameters of the suspension system. SUMMARY

[0005] The method, device, equipment, medium and program product for determining the parameters of the suspension system provided by the embodiments of the present application solve the problem of low efficiency in determining the parameters of the suspension system caused by the fact that the parameters in the algorithm used to solve the multi-objective optimization model are fixed.

[0006] In a first aspect, the embodiments of the present application provide a method for determining parameters of a suspension system, comprising:

[0007] obtaining an initial crossover rate, an initial mutation rate and a multi-objective optimization model of the suspension system;

[0008] determining an initial population according to a constraint condition of the multi-objective optimization model;

[0009] solving the multi-objective optimization model according to a non-dominated sorting genetic algorithm, an algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate to obtain target parameters of the suspension system;

[0010] The algorithm parameter dynamic optimization model is a pre-selected deep Q network model for determining a crossover rate increment and a mutation rate increment according to state data of the population, and the state data of the population includes a population diversity index, a convergence index and a dominance relationship distribution index.

[0011] In a possible implementation, the determining the initial population according to the constraint condition of the multi-objective optimization model comprises:

[0012] selecting a preset number of individuals from all individuals satisfying the constraint condition of the multi-objective optimization model to form the initial population.

[0013] In a possible implementation, the solving of the multi-objective optimization model according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate comprises:

[0014] According to the initial population and the non-dominated sorting genetic algorithm, state data of the initial population is calculated.

[0015] According to the state data of the initial population and the algorithm parameter dynamic optimization model, a crossover rate increment and a mutation rate increment are determined.

[0016] According to the crossover rate increment, the mutation rate increment, the initial crossover rate and the initial mutation rate, a target crossover rate and a target mutation rate are calculated.

[0017] According to the target crossover rate, the target mutation rate, an objective function of the multi-objective optimization model and the non-dominated sorting genetic algorithm, a new initial population is generated.

[0018] It is judged whether a convergence condition is established.

[0019] If the convergence condition is not established, the target crossover rate is taken as a new initial crossover rate, the target mutation rate is taken as a new target mutation rate, and the above steps are repeated until the convergence condition is established, and a target suspension system parameter is selected from the new initial population.

[0020] In a possible implementation, the calculating of the state data of the initial population according to the initial population and the non-dominated sorting genetic algorithm comprises:

[0021] Euclidean distances of each two individuals in the initial population are calculated to obtain a first Euclidean distance data set.

[0022] A variance of the first Euclidean distance data set is taken as a population diversity index.

[0023] The initial population is subjected to non-dominated sorting processing and crowdedness processing by using the non-dominated sorting genetic algorithm to obtain crowdedness of each individual in the initial population.

[0024] According to the crowdedness of each individual in the initial population, a dominance relationship distribution index is calculated.

[0025] If the current loop is the first loop, a convergence index is determined as a preset value.

[0026] If the current loop is not the first loop, the convergence index is calculated according to the initial population of the current loop and the initial population of the last loop of the current loop.

[0027] In a possible implementation, the calculating the convergence index according to the initial population of the current cycle and the initial population of the last cycle of the current cycle comprises:

[0028] The initial population of the current cycle is subjected to non-dominated sorting processing by using the non-dominated sorting genetic algorithm to obtain a first Pareto front sequence;

[0029] The initial population of the last cycle of the current cycle is subjected to non-dominated sorting processing by using the non-dominated sorting genetic algorithm to obtain a second Pareto front sequence;

[0030] The first hyper-volume of the first Pareto front in the first Pareto front sequence and the second hyper-volume of the first Pareto front in the second Pareto front sequence are calculated;

[0031] The difference between the first hyper-volume and the second hyper-volume is divided by the second hyper-volume to obtain the convergence index.

[0032] In a possible implementation, before the obtaining the initial crossover rate, the initial mutation rate and the system data of the suspended system, the method further comprises:

[0033] According to the non-dominated sorting genetic algorithm and the obtained training population, the initial deep Q network model, the training crossover rate and the training mutation rate, a training data set is generated, each training data in the training data set comprising first state data, action data, reward data and second state data, the action data comprising a crossover rate increment and a mutation rate increment;

[0034] The initial deep Q network model is subjected to model training according to the training data set to obtain the algorithm parameter dynamic optimization model.

[0035] In a possible implementation, the generating the training data set according to the non-dominated sorting genetic algorithm and the obtained training population, the initial deep Q network model, the training crossover rate and the training mutation rate comprises:

[0036] The first state data is calculated according to the training population and the non-dominated sorting genetic algorithm;

[0037] The first state data is input into the initial deep Q network model to obtain action data;

[0038] The new training crossover rate and the new training mutation rate are calculated according to the action data, the training crossover rate and the training mutation rate;

[0039] The target population is generated according to the new training crossover rate, the new training mutation rate, the training population and the non-dominated sorting genetic algorithm;

[0040] calculate reward data according to the training population and the target population;

[0041] calculate second state data according to the target population and the non-dominated sorting genetic algorithm;

[0042] compose the first state data, the action data, the reward data and the second state data into one training data;

[0043] update the iteration number;

[0044] if the updated iteration number is less than a preset iteration number threshold, repeat the above steps by taking the target population as a new training population until the updated iteration number is equal to the preset iteration number threshold, and select a preset training number of data from all training data to compose the training data set.

[0045] In a possible implementation, the calculating reward data according to the training population and the target population comprises:

[0046] calculate Euclidean distances between each two individuals in the target population to obtain a second Euclidean distance data set;

[0047] take a variance of the second Euclidean distance data set as a diversity parameter;

[0048] perform non-dominated sorting processing on the training population by using the non-dominated sorting genetic algorithm to obtain a third Pareto front sequence;

[0049] perform non-dominated sorting processing on the target population by using the non-dominated sorting genetic algorithm to obtain a fourth Pareto front sequence;

[0050] calculate a third hyper volume of a first Pareto front in the third Pareto front sequence and a fourth hyper volume of a first Pareto front in the fourth Pareto front sequence;

[0051] take a difference between the fourth hyper volume and the third hyper volume as a convergence quality parameter;

[0052] calculate the reward data according to the diversity parameter and the convergence quality parameter.

[0053] In a second aspect, an embodiment of the present application provides a parameter determination apparatus of a suspension system, comprising:

[0054] an acquisition module configured to acquire an initial crossover rate, an initial mutation rate and a multi-objective optimization model of the suspension system;

[0055] a processing module configured to:

[0056] determine an initial population according to a constraint condition of the multi-objective optimization model;

[0057] solve the multi-objective optimization model according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate, to obtain a target suspension system parameter;

[0058] The algorithm parameter dynamic optimization model is a pre-selected and trained deep Q network model for determining a crossover rate increment and a mutation rate increment according to state data of a population. The state data of the population includes a population diversity index, a convergence index and a dominance relationship distribution index.

[0059] In a third aspect, an embodiment of the present application provides an electronic device, comprising:

[0060] a processor, a memory and a communication interface;

[0061] The memory is configured to store executable instructions of the processor.

[0062] The processor is configured to execute the executable instructions to implement the method for determining the suspension system parameter according to any one of the first aspect.

[0063] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon. The computer program is executed by a processor to implement the method for determining the suspension system parameter according to any one of the first aspect.

[0064] In a fifth aspect, an embodiment of the present application provides a computer program product comprising a computer program. The computer program is executed by a processor to implement the method for determining the suspension system parameter according to any one of the sixth aspect.

[0065] The method, device, equipment, medium and program product for determining the suspension system parameter provided by the embodiments of the present application, after obtaining the initial crossover rate, the initial mutation rate and the multi-objective optimization model of the suspension system, determine the initial population according to the constraint condition of the multi-objective optimization model, and then solve the multi-objective optimization model according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate, to obtain the target suspension system parameter. The algorithm parameter dynamic optimization model is a pre-selected and trained deep Q network model for determining a crossover rate increment and a mutation rate increment according to state data of a population. The present solution dynamically adjusts the crossover rate and the mutation rate of the non-dominated sorting genetic algorithm according to the state data of the population in the solving process through the algorithm parameter dynamic optimization model, which can improve the solving efficiency and thus improve the efficiency of determining the suspension system parameter. BRIEF DESCRIPTION OF DRAWINGS

[0066] The accompanying drawings, which are incorporated herein and constitute part of this specification, illustrate embodiments consistent with the application and, together with the description, further serve to explain the principles of the application.

[0067] Figure 1 A flowchart of a method for determining parameters of a suspension system according to an embodiment of the application;

[0068] Figure 2 A flowchart of a method for determining parameters of a suspension system according to an embodiment of the application;

[0069] Figure 3 A structure diagram of a device for determining parameters of a suspension system according to an embodiment of the application;

[0070] Figure 4 A structure diagram of an electronic device according to an embodiment of the application.

[0071] The above-described drawings show certain embodiments of the application in accordance with the present disclosure, and will hereinafter be described in detail. These drawings and the associated description herein are to be read in conjunction with the patent claims. The application illustratively disclosed herein suitably can be practiced in the absence of any element(s) not specifically disclosed herein. The terms "comprising," "including," "containing," etc. shall be read expansively and without limitation. The terms "comprising," "having," "including," and "containing" are to be construed as open-ended terms (i.e., meaning "including, but not limited to,") unless explicitly provided otherwise in this specification. DETAILED DESCRIPTION

[0072] Reference will now be made to the exemplary embodiments illustrated, and specific language will be used herein to describe the same. It will, however, be apparent that no limitation of the scope of the application is intended by this specific language. Alterations and further modifications of the inventive features described herein, and additional applications of these features, which would occur to one skilled in the art are to be considered within the scope of the application as described in such disclosure and as recited in the claims.

[0073] The terms "first," "second," "third," "fourth," and the like in the description and in the claims, where they occur, are not used to denote or otherwise connote any ordinal, sequential, or high-low relationship. It is to be understood that the terms "comprising," "including," "containing," and "having" are intended to be open-ended and, in addition to the specific items or acts listed, encompassing any processes, methods, systems, compositions of matter, or apparatuses that fall within the scope of the claims. The terms "comprising," "including," "containing," and "having" are to be construed as open-ended terms (i.e., meaning "including, but not limited to,") unless explicitly provided otherwise in this specification or claims.

[0074] In a vehicle, a suspension system is essential for isolating the vibration and noise transmission between the power assembly, the vehicle body and the chassis, which can improve the comfort, reliability and durability of the vehicle. In order to design a suspension system with high isolation effect and structural strength, the suspension system parameters need to be determined, and then the suspension system is constructed.

[0075] In the prior art, the determination of the suspension system parameters is usually based on the suspension system parameters to construct a multi-objective optimization model, and then the suspension system parameters are obtained by solving the multi-objective optimization model. Since the parameters in the algorithm used to solve the multi-objective optimization model are fixed, it will cause the problem of low efficiency in determining the suspension system parameters.

[0076] In view of the problems in the prior art, the inventors found that in order to improve the efficiency of determining the suspension system parameters, the solving efficiency can be improved. The non-dominated sorting genetic algorithm is used for solving, which generates a new population according to the existing population, combines the crossover rate and the mutation rate, and then generates a new population again. The new population is used as the existing population to generate a new population again. After convergence, the target suspension system parameters are selected from the new population. Therefore, each time a new population is generated, the dynamic optimization model can be updated according to the algorithm parameters, the state data of the existing population is updated, the crossover rate and the mutation rate are updated, and a new population is generated according to the updated crossover rate and the mutation rate. By adjusting the crossover rate and the mutation rate through the state data, the convergence efficiency, that is, the solving efficiency, can be improved, and then the efficiency of determining the suspension system parameters can be improved. Based on the above invention concept, the suspension system parameter determination scheme in the present application is designed.

[0077] The execution subject of the suspension system parameter determination method in the present application can be a computer, and can also be a server, a terminal device, etc. The present application does not limit it, and the following will be described taking a computer as an example.

[0078] The application scenario of the suspension system parameter determination provided in the present application is exemplarily illustrated as follows.

[0079] Exemplarily, in this application scenario, the staff needs to develop a new suspension system, and a multi-objective optimization model of the suspension system is constructed according to the vehicle condition.

[0080] In order to determine the suspension system parameters, the staff inputs the initial crossover rate, the initial mutation rate and the multi-objective optimization model of the suspension system into the computer.

[0081] The computer determines the initial population according to the constraint conditions of the multi-objective optimization model.

[0082] Then, the multi-objective optimization model is solved according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate, and the target suspension system parameters are obtained.

[0083] That is, according to the initial population and non-dominated sorting genetic algorithm, the state data of the initial population is calculated; according to the state data of the initial population and the algorithm parameter dynamic optimization model, the crossover rate increment and the mutation rate increment are determined.

[0084] According to the crossover rate increment, the mutation rate increment, the initial crossover rate and the initial mutation rate, the target crossover rate and the target mutation rate are calculated; and then according to the target crossover rate, the target mutation rate, the objective function of the multi-objective optimization model and the non-dominated sorting genetic algorithm, a new initial population is generated.

[0085] Further, it is judged whether the convergence condition is established; if the convergence condition is not established, the target crossover rate is taken as the new initial crossover rate, the target mutation rate is taken as the new target mutation rate, and the process is repeated until the convergence condition is established, and the target suspension system parameters are selected from the new initial population.

[0086] The subsequent staff constructs the suspension system according to the suspension system parameters.

[0087] It should be noted that the above scenario is only an example of an application scenario provided by the embodiments of the present application, and the actual form of various devices included in the scenario is not limited by the embodiments of the present application. In the specific application of the scheme, the actual demand can be set.

[0088] Next, the technical scheme of the present application will be described in detail through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in detail in some embodiments.

[0089] Figure 1 The flowchart of the suspension system parameter determination method embodiment provided by the present application is shown in the figure, and the embodiments of the present application explain the case that the computer dynamically adjusts the crossover rate and mutation rate in the non-dominated sorting genetic algorithm during solving by using the non-dominated sorting genetic algorithm according to the state data of the population and the algorithm parameter dynamic optimization model. The method in this embodiment can be realized by software, hardware or a combination of software and hardware. As Figure 1 shown, the suspension system parameter determination method specifically includes the following steps:

[0090] S101: Obtain the initial crossover rate, the initial mutation rate and the multi-objective optimization model of the suspension system.

[0091] In order to obtain a suspension system with high isolation effect and structural strength, after constructing the multi-objective optimization model of the suspension system, the staff inputs it to the computer, and at the same time, in order to solve it, the initial crossover rate and the initial mutation rate also need to be inputted.

[0092] In this step, the worker will obtain the initial crossover rate, the initial mutation rate and the multi-objective optimization model of the suspension system, and the computer can obtain these data.

[0093] It should be noted that the multi-objective optimization model of the suspension system is composed of a plurality of objective functions and at least one constraint condition, the suspension system parameters are vectors, the suspension system parameters are solutions of the multi-objective optimization model, the objective functions are composed of elements of the suspension system parameters, and each constraint condition is a limiting condition of at least one element of the suspension system parameters. The limiting condition of at least one element can be an upper limit or a lower limit of the element, and can also be an upper limit or a lower limit of a formula composed of the element. The formula composed of the element can represent engine displacement, natural frequency, etc.

[0094] The element of the suspension system parameters can be the stiffness, position and angle of each suspension point in the suspension system. The embodiments of the present application do not limit the element of the suspension system parameters, which can be determined according to actual conditions.

[0095] The objective function can be a function of maximizing energy decoupling rate, minimizing force transmitted by the suspension system to the vehicle body, minimizing the difference between the natural frequency and the set frequency, minimizing the material cost, etc. The embodiments of the present application do not limit the objective function, which can be determined according to actual conditions.

[0096] The constraint condition can be an upper limit and a lower limit of the stiffness of each suspension point, an upper limit and a lower limit of the engine displacement, an upper limit and a lower limit of the angle of each suspension point, etc. The embodiments of the present application do not limit the constraint condition, which can be determined according to actual conditions.

[0097] It should be noted that the multi-objective optimization model of the suspension system in the present application can also be any multi-objective optimization model of a suspension system in the prior art, which is not limited by the embodiments of the present application and can be determined according to actual conditions.

[0098] S102: Determine the initial population according to the constraint condition of the multi-objective optimization model.

[0099] In this step, after the computer obtains the multi-objective optimization model, in order to solve it, the initial population needs to be determined according to the constraint condition of the multi-objective optimization model.

[0100] The solution that meets the constraint condition of the multi-objective optimization model can be called an individual. From all the individuals that meet the constraint condition of the multi-objective optimization model, a preset number of individuals are selected to form the initial population.

[0101] S103: Solve the multi-objective optimization model according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate, and obtain the target suspension system parameters.

[0102] In this step, after the computer generates the initial population, the multi-objective optimization model is solved according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate, and the target suspension system parameters are obtained.

[0103] The algorithm parameter dynamic optimization model is a pre-selected deep Q network model for determining the crossover rate increment and the mutation rate increment according to the state data of the population, and the state data of the population includes the population diversity index, the convergence index and the dominance relationship distribution index. The non-dominated sorting genetic algorithm is NSGA-II (Non-dominated Sorting Genetic Algorithm II).

[0104] Specifically, in order to improve the solving efficiency, that is, to find individuals that meet the objective function faster, the state data of the initial population needs to be calculated according to the initial population and the non-dominated sorting genetic algorithm.

[0105] Further, the crossover rate increment and the mutation rate increment are determined according to the state data of the initial population and the algorithm parameter dynamic optimization model. That is, the state data of the initial population is input into the algorithm parameter dynamic optimization model to obtain the Q value of multiple actions, each action is composed of the crossover rate increment and the mutation rate increment, and then the action with the maximum Q value is determined as the determined crossover rate increment and mutation rate increment.

[0106] The target crossover rate and the target mutation rate are calculated according to the crossover rate increment, the mutation rate increment, the initial crossover rate and the initial mutation rate. That is, the initial crossover rate is added to the crossover rate increment to obtain the target crossover rate, and the initial mutation rate is added to the mutation rate increment to obtain the target mutation rate.

[0107] The new initial population is generated according to the target crossover rate, the target mutation rate, the objective function of the multi-objective optimization model and the non-dominated sorting genetic algorithm.

[0108] The non-dominated sorting genetic algorithm first performs non-dominated sorting processing and calculates the crowdedness processing on the initial population according to the objective function of the multi-objective optimization model, to obtain the Pareto front sequence of the initial population and the crowdedness of each individual in the initial population. The Pareto front sequence includes sequentially arranged Pareto fronts, and each Pareto front includes at least one individual in the initial population. The non-dominated sorting genetic algorithm generates a new initial population according to the Pareto front sequence, the crowdedness of each individual, the target crossover rate and the target mutation rate.

[0109] After the new initial population is generated, it is necessary to determine whether the convergence condition is established to determine whether the new initial population is the optimal solution.

[0110] If the convergence condition is not established, the target crossover rate is taken as a new initial crossover rate, the target mutation rate is taken as a new target mutation rate, the above steps are repeated until the convergence condition is established, and the target suspension system parameters are selected from the new initial population.

[0111] The state data of the initial population is calculated according to the initial population and the non-dominated sorting genetic algorithm, and the calculation manner is as follows:

[0112] The Euclidean distance between each two individuals in the initial population is calculated to obtain a first Euclidean distance data set. The variance of the first Euclidean distance data set is taken as a population diversity index.

[0113] The initial population is subjected to non-dominated sorting processing and calculation of crowding degree processing by using the non-dominated sorting genetic algorithm and combining the objective function of the multi-objective optimization model, and the crowding degree of each individual in the initial population is obtained. The domination relationship distribution index is calculated according to the crowding degree of each individual in the initial population.

[0114] If the current cycle is the first cycle, the convergence index is determined as a preset value.

[0115] It should be noted that the preset value can be 1, 0.9, 0.8, etc. The preset value is not limited in the embodiments of the present application, and can be determined according to actual conditions.

[0116] If the current cycle is not the first cycle, the convergence index is calculated according to the initial population of the current cycle and the initial population of the last cycle of the current cycle.

[0117] The population diversity index, the domination relationship distribution index and the convergence index constitute the state data.

[0118] When the current cycle is not the first cycle, the convergence index is calculated in the following manner:

[0119] The initial population of the current cycle is subjected to non-dominated sorting processing by using the non-dominated sorting genetic algorithm and combining the objective function of the multi-objective optimization model to obtain a first Pareto front sequence.

[0120] The initial population of the last cycle of the current cycle is subjected to non-dominated sorting processing by using the non-dominated sorting genetic algorithm and combining the objective function of the multi-objective optimization model to obtain a second Pareto front sequence.

[0121] The first hyper-volume of the first Pareto front in the first Pareto front sequence and the second hyper-volume of the first Pareto front in the second Pareto front sequence are calculated.

[0122] The difference between the first hyper-volume and the second hyper-volume is divided by the second hyper-volume to obtain the convergence index.

[0123] The domination relationship distribution index is calculated in the following manner:

[0124] The domination relationship distribution index is calculated by using the formula , wherein , , represents the domination relationship distribution index, and are preset weights, represents the number of individuals in the first Pareto front in the first Pareto front sequence, and N represents the number of individuals in the initial population, represents the crowding degree of the i-th individual in the initial population, represents the maximum value of the crowding degrees of all individuals in the initial population.

[0125] It should be noted that and sum to 1, and may be 0.1, 0.2, 0.4, 0.6, 0.8, 0.9, etc. and are not limited, and can be determined according to actual conditions.

[0126] The convergence condition is that the current loop is not the first loop, and the updated convergence cumulative number is equal to the preset number.

[0127] The manner of determining the updated convergence cumulative number is as follows:

[0128] The non-dominated sorting genetic algorithm is used in combination with the objective function of the multi-objective optimization model to perform non-dominated sorting processing on the new population, to obtain the fifth Pareto front sequence. The fifth hyper volume of the first Pareto front in the fifth Pareto front sequence is calculated. The absolute value of the difference between the fifth hyper volume and the first hyper volume is divided by the first hyper volume to determine the convergence parameter. If the convergence parameter is less than the preset convergence threshold, the convergence cumulative number is incremented by one to obtain the updated convergence cumulative number. If the convergence parameter is greater than or equal to the preset convergence threshold, the convergence cumulative number is updated to 0 to obtain the updated convergence cumulative number.

[0129] It should be noted that the preset convergence threshold can be 0.001, 0.01, 0.1, 1, etc. The preset number can be 3, 5, 7, etc. The preset convergence threshold and the preset number are not limited in the embodiments of the present application, and can be determined according to actual conditions.

[0130] In an implementation manner, the convergence condition can also be that the loop number is equal to the preset loop number threshold.

[0131] It should be noted that the preset cycle threshold can be 100, 1000, 10000, etc., and the embodiment does not limit the preset cycle threshold, which can be determined according to actual conditions.

[0132] The way of selecting the target suspension system parameter from the new initial population is: selecting the target suspension system parameter from the first Pareto front in the fifth Pareto front sequence.

[0133] The method for determining the suspension system parameter provided in the embodiment, after the initial crossover rate, the initial mutation rate and the multi-objective optimization model of the suspension system are obtained, the initial population is determined according to the constraint condition of the multi-objective optimization model, and then the multi-objective optimization model is solved according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate, to obtain the target suspension system parameter. The algorithm parameter dynamic optimization model is a pre-selected deep Q network model for determining the crossover rate increment and the mutation rate increment according to the state data of the population. In the solving process of the present scheme, the crossover rate and the mutation rate of the non-dominated sorting genetic algorithm are dynamically adjusted according to the state data of the population, which can improve the solving efficiency and thus improve the efficiency of determining the suspension system parameter.

[0134] In addition, by dynamically adjusting the crossover rate and the mutation rate of the non-dominated sorting genetic algorithm, the local optimal solution caused by premature convergence can be avoided. By dynamically adjusting the crossover rate and the mutation rate through the state data, the crossover rate and the mutation rate are more matched with the population state. The present scheme is applicable to different multi-objective optimization models and has stronger applicability.

[0135] Figure 2 The flowchart of the second embodiment of the method for determining the suspension system parameter provided in the present application is described based on the above-mentioned embodiments. As shown in Figure 2 The method for determining the suspension system parameter specifically includes the following steps:

[0136] S201: generating a training data set according to the non-dominated sorting genetic algorithm and the obtained training population, the initial deep Q network model, the training crossover rate and the training mutation rate.

[0137] In this step, in order to generate the algorithm parameter dynamic optimization model, the training data set needs to be generated first. The training data set can be generated according to the non-dominated sorting genetic algorithm and the obtained training population, the initial deep Q network model, the training crossover rate and the training mutation rate.

[0138] Each training data in the training data set comprises the first state data, the action data, the reward data and the second state data, and the action data comprises the crossover rate increment and the mutation rate increment.

[0139] Specifically, the first state data is calculated according to the training population and the non-dominated sorting genetic algorithm.

[0140] Further, the first state data is input into the initial deep Q network model to obtain the action data.

[0141] According to the action data, the training crossover rate and the training mutation rate, a new training crossover rate and a new training mutation rate are calculated. That is, the training crossover rate is added to the crossover rate increment to obtain the new training crossover rate, and the training mutation rate is added to the mutation rate increment to obtain the new training mutation rate.

[0142] According to the new training crossover rate, the new training mutation rate, the training population and the non-dominated sorting genetic algorithm, the target population is generated.

[0143] It should be noted that the way of generating the target population is similar to the way of generating the new initial population in step S103 of Embodiment One, which will not be repeated here.

[0144] According to the training population and the target population, the reward data is calculated.

[0145] According to the target population and the non-dominated sorting genetic algorithm, the second state data is calculated.

[0146] It should be noted that the first state data and the second state data comprise population diversity indicators, convergence indicators and dominance relationship distribution indicators. The way of calculating the first state data and the second state data is similar to the way of calculating the state data of the initial population in step S103 of Embodiment One, which will not be repeated here.

[0147] The first state data, the action data, the reward data and the second state data are combined to form a training data.

[0148] The iteration number is updated.

[0149] If the updated iteration number is less than the preset iteration number threshold, the target population is taken as a new training population, and the above steps are repeated until the updated iteration number is equal to the preset iteration number threshold, and a preset training number of data is selected from all the training data to form a training data set.

[0150] It should be noted that the preset training quantity is less than the preset iteration number, and the preset training quantity can be 500, 1000, 10000, etc., and the preset iteration number can be 800, 1500, 20000, etc. The preset training quantity and the preset iteration number are not limited by the embodiments of the application, and can be determined according to actual conditions.

[0151] The method for calculating the reward data according to the training population and the target population is as follows:

[0152] The Euclidean distance between each two individuals in the target population is calculated to obtain a second Euclidean distance data set, and the variance of the second Euclidean distance data set is taken as the diversity parameter.

[0153] The non-dominated sorting genetic algorithm is used to perform non-dominated sorting processing on the training population in combination with the objective function of the multi-objective optimization model to obtain a third Pareto front sequence. The non-dominated sorting genetic algorithm is used to perform non-dominated sorting processing on the target population in combination with the objective function of the multi-objective optimization model to obtain a fourth Pareto front sequence.

[0154] The third hyper-volume of the first Pareto front in the third Pareto front sequence and the fourth hyper-volume of the first Pareto front in the fourth Pareto front sequence are calculated. The difference between the fourth hyper-volume and the third hyper-volume is taken as the convergence quality parameter.

[0155] The reward data is calculated according to the diversity parameter and the convergence quality parameter. That is, the product of the diversity parameter and a preset diversity weight is added to the product of the convergence quality parameter and a preset convergence weight to obtain the reward data.

[0156] It should be noted that the preset diversity weight and the preset convergence weight can be 0.1, 0.3, 0.4, 0.6, 0.7, 0.9, etc. The preset diversity weight and the preset convergence weight are not limited by the embodiments of the application, and can be determined according to actual conditions.

[0157] S202: Model training is performed on the initial deep Q network model according to the training data set to obtain an algorithm parameter dynamic optimization model.

[0158] In this step, after the computer obtains the training data set, model training is performed on the initial deep Q network model according to the training data in the training data set to obtain an algorithm parameter dynamic optimization model.

[0159] It should be noted that the training method used in this step is the training method of the deep Q network model in the prior art, which will not be described here.

[0160] It should be noted that in the determination of the target suspension system parameters, the computer can determine the first state data, the action data, the reward data and the second state data in step S201 in each cycle, and store them. After the target suspension system parameters are determined, the stored data is used to perform incremental training on the algorithm parameter dynamic optimization model, so as to improve the accuracy of the algorithm parameter dynamic optimization model.

[0161] The method for determining suspension system parameters provided in this embodiment can generate a training data set according to the non-dominated sorting genetic algorithm and the initial deep Q network model, and then perform model training on the initial deep Q network model according to the training data set to obtain an algorithm parameter dynamic optimization model, so as to improve the accuracy of the algorithm parameter dynamic optimization model. The reward data is calculated by the diversity parameter and the convergence quality parameter, so that the algorithm parameter dynamic optimization model can dynamically balance global search and local development according to the evolution stage.

[0162] The following is an apparatus embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the apparatus embodiments of the present application, please refer to the method embodiments of the present application.

[0163] Figure 3 The structure diagram of the suspension system parameter determination apparatus embodiment provided in the present application is shown in FIG. 3. As shown in FIG. 3, the suspension system parameter determination apparatus 30 comprises: Figure 3

[0164] The acquisition module 31 is configured to acquire an initial crossover rate, an initial mutation rate and a multi-objective optimization model of a suspension system.

[0165] The processing module 32 is configured to:

[0166] determine an initial population according to the constraint condition of the multi-objective optimization model;

[0167] solve the multi-objective optimization model according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate, to obtain target suspension system parameters;

[0168] The algorithm parameter dynamic optimization model is a pre-selected deep Q network model for determining a crossover rate increment and a mutation rate increment according to state data of a population, and the state data of the population includes a population diversity index, a convergence index and a dominance relationship distribution index.

[0169] Further, the processing module 32 is specifically configured to:

[0170] select a preset number of individuals from all individuals satisfying the constraint condition of the multi-objective optimization model to form the initial population.

[0171] Further, the processing module 32 is specifically configured to:​

[0172] According to the initial population and the non-dominated sorting genetic algorithm, state data of the initial population is calculated;

[0173] According to the state data of the initial population and the algorithm parameter dynamic optimization model, a crossover rate increment and a mutation rate increment are determined;

[0174] According to the crossover rate increment, the mutation rate increment, an initial crossover rate and an initial mutation rate, a target crossover rate and a target mutation rate are calculated;

[0175] According to the target crossover rate, the target mutation rate, an objective function of the multi-objective optimization model and the non-dominated sorting genetic algorithm, a new initial population is generated;

[0176] It is judged whether a convergence condition is established or not;

[0177] If the convergence condition is not established, the target crossover rate is taken as a new initial crossover rate, the target mutation rate is taken as a new target mutation rate, and the above steps are repeated until the convergence condition is established, and a target suspension system parameter is selected from the new initial population.

[0178] Further, the processing module 32 is specifically configured to:

[0179] Euclidean distances between each two individuals in the initial population are calculated to obtain a first Euclidean distance data set;

[0180] A variance of the first Euclidean distance data set is taken as a population diversity index;

[0181] The initial population is subjected to non-dominated sorting processing and crowdedness processing by using the non-dominated sorting genetic algorithm to obtain crowdedness of each individual in the initial population;

[0182] According to the crowdedness of each individual in the initial population, a dominance relationship distribution index is calculated;

[0183] If the current cycle is the first cycle, the convergence index is determined as a preset value;

[0184] If the current cycle is not the first cycle, the convergence index is calculated according to the initial population of the current cycle and the initial population of the last cycle of the current cycle.

[0185] Further, the processing module 32 is specifically configured to:

[0186] The initial population of the current cycle is subjected to non-dominated sorting processing by using the non-dominated sorting genetic algorithm to obtain a first Pareto front sequence;

[0187] The initial population of the last cycle of the current cycle is subjected to non-dominated sorting processing by using the non-dominated sorting genetic algorithm to obtain a second Pareto front sequence;

[0188] calculate a first hyper-volume of a first Pareto front in the first sequence of Pareto fronts, and a second hyper-volume of a first Pareto front in the second sequence of Pareto fronts;

[0189] divide the difference between the first hyper-volume and the second hyper-volume by the second hyper-volume to obtain a convergence indicator.

[0190] Further, before obtaining the initial crossover rate, the initial mutation rate and the system data of the pending system, the processing module 32 is further configured to:

[0191] generate, according to the non-dominated sorting genetic algorithm and the obtained training population, the initial deep Q network model, the training crossover rate and the training mutation rate, a training data set, each training data in the training data set comprising first state data, action data, reward data and second state data, the action data comprising a crossover rate increment and a mutation rate increment;

[0192] perform model training on the initial deep Q network model according to the training data set to obtain an algorithm parameter dynamic optimization model.

[0193] Further, the processing module 32 is specifically further configured to:

[0194] calculate the first state data according to the training population and the non-dominated sorting genetic algorithm;

[0195] input the first state data into the initial deep Q network model to obtain the action data;

[0196] calculate a new training crossover rate and a new training mutation rate according to the action data, the training crossover rate and the training mutation rate;

[0197] generate a target population according to the new training crossover rate, the new training mutation rate, the training population and the non-dominated sorting genetic algorithm;

[0198] calculate the reward data according to the training population and the target population;

[0199] calculate the second state data according to the target population and the non-dominated sorting genetic algorithm;

[0200] combine the first state data, the action data, the reward data and the second state data to form a training data;

[0201] update the iteration number;

[0202] if the updated iteration number is less than a preset iteration number threshold, take the target population as a new training population, repeat the above steps until the updated iteration number is equal to the preset iteration number threshold, and select a preset training number of data from all the training data to form a training data set.

[0203] Further, the processing module 32 is further configured to:

[0204] calculate the Euclidean distance of each two individuals in the target population to obtain a second Euclidean distance data set;

[0205] take the variance of the second Euclidean distance data set as the diversity parameter;

[0206] perform non-dominated sorting on the training population by using the non-dominated sorting genetic algorithm to obtain a third Pareto front sequence;

[0207] perform non-dominated sorting on the target population by using the non-dominated sorting genetic algorithm to obtain a fourth Pareto front sequence;

[0208] calculate a third hyper volume of the first Pareto front in the third Pareto front sequence and a fourth hyper volume of the first Pareto front in the fourth Pareto front sequence;

[0209] take the difference between the fourth hyper volume and the third hyper volume as the convergence quality parameter;

[0210] calculate the reward data according to the diversity parameter and the convergence quality parameter.

[0211] The suspension system parameter determination apparatus provided in this embodiment is used to execute the technical solutions in any of the method embodiments, and has similar implementation principles and technical effects, which will not be described herein again.

[0212] Figure 4 A structural schematic diagram of an electronic device is provided in this application. As shown in the figure, the electronic device 40 includes: Figure 4

[0213] a processor 41, a memory 42, and a communication interface 43;

[0214] The memory 42 is configured to store executable instructions of the processor 41.

[0215] The processor 41 is configured to execute the technical solutions in any of the method embodiments by executing the executable instructions.

[0216] Optionally, the memory 42 can be independent or integrated with the processor 41.

[0217] Optionally, when the memory 42 is independent of the processor 41, the electronic device 40 can further include:

[0218] a bus 44, the memory 42 and the communication interface 43 are connected with the processor 41 through the bus 44 and complete communication with each other, and the communication interface 43 is configured to communicate with other devices.

[0219] ​Optionally, the communication interface 43 can be implemented by a transceiver. The communication interface is used to realize the communication between the database access device and other devices (for example, a client, a read-write library and a read-only library). The memory can include a random access memory (RAM) and can also include a non-volatile memory, for example, at least one disk memory.

[0220] The bus 44 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0221] The processor described above can be a general-purpose processor, including a central processing unit CPU, a network processor NP, etc.; can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0222] The electronic device is used to execute the technical solutions in any of the preceding method embodiments, and has similar implementation principles and technical effects, which will not be described here.

[0223] The embodiments of the present application also provide a readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the technical solutions provided by any of the preceding method embodiments.

[0224] The embodiments of the present application also provide a computer program product including a computer program, and the computer program is executed by a processor to implement the technical solutions provided by any of the preceding method embodiments.

[0225] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The above-mentioned program can be stored in a computer readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the above-mentioned storage medium includes a ROM, a RAM, a magnetic disk or an optical disk and various storage media that can store program codes.

[0226] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method of determining parameters of a suspension system, characterized by The method comprises the following steps: acquiring an initial crossover rate, an initial mutation rate and a multi-objective optimization model of a suspension system; determining an initial population according to a constraint condition of the multi-objective optimization model; solving the multi-objective optimization model according to a non-dominated sorting genetic algorithm, an algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate to obtain target suspension system parameters; wherein the algorithm parameter dynamic optimization model is a preselected deep Q network model for determining a crossover rate increment and a mutation rate increment according to state data of the population, and the state data of the population comprises a population diversity index, a convergence index and a dominance relationship distribution index; the solving of the multi-objective optimization model according to the non-dominated sorting genetic algorithm, the algorithm parameter dynamic optimization model, the initial population, the initial crossover rate and the initial mutation rate to obtain the target suspension system parameters comprises: calculating the state data of the initial population according to the initial population and the non-dominated sorting genetic algorithm; determining the crossover rate increment and the mutation rate increment according to the state data of the initial population and the algorithm parameter dynamic optimization model; calculating a target crossover rate and a target mutation rate according to the crossover rate increment, the mutation rate increment, the initial crossover rate and the initial mutation rate; generating a new initial population according to the target crossover rate, the target mutation rate, an objective function of the multi-objective optimization model and the non-dominated sorting genetic algorithm; judging whether a convergence condition is established; if the convergence condition is not established, taking the target crossover rate as a new initial crossover rate, taking the target mutation rate as a new target mutation rate, repeating the above steps until the convergence condition is established, and selecting the target suspension system parameters from the new initial population.

2. The method of claim 1, wherein, the determining of the initial population according to the constraint condition of the multi-objective optimization model comprises: selecting a preset number of individuals from all individuals satisfying the constraint condition of the multi-objective optimization model to form the initial population.

3. The method of claim 1, wherein, the calculating of the state data of the initial population according to the initial population and the non-dominated sorting genetic algorithm comprises: calculating the Euclidean distance between each two individuals in the initial population to obtain a first Euclidean distance data set; taking the variance of the first Euclidean distance data set as the population diversity index; performing non-dominated sorting processing and computing crowding degree processing on the initial population by using the non-dominated sorting genetic algorithm to obtain the crowding degree of each individual in the initial population; calculating the dominance relationship distribution index according to the crowding degree of each individual in the initial population; if the current cycle is the first cycle, determining the convergence index as a preset value; if the current cycle is not the first cycle, calculating the convergence index according to the initial population of the current cycle and the initial population of the last cycle of the current cycle.

4. The method of claim 3, wherein, the calculating of the convergence index according to the initial population of the current cycle and the initial population of the last cycle of the current cycle comprises: performing non-dominated sorting processing on the initial population of the current cycle by using the non-dominated sorting genetic algorithm to obtain a first Pareto front sequence; The non-dominated sorting genetic algorithm is used to perform non-dominated sorting on the initial population of the previous cycle of the current cycle to obtain a second Pareto front sequence; A first hyper volume of a first Pareto front in the first Pareto front sequence and a second hyper volume of a first Pareto front in the second Pareto front sequence are calculated; The difference between the first hyper volume and the second hyper volume is divided by the second hyper volume to obtain the convergence index.

5. The method according to any one of claims 1 to 4, characterized in that, Before the initial crossover rate, the initial mutation rate and the system data of the suspension system are obtained, the method further comprises: According to the non-dominated sorting genetic algorithm, and the obtained training population, initial deep Q network model, training crossover rate and training mutation rate, a training data set is generated, each training data in the training data set comprises first state data, action data, reward data and second state data, and the action data comprises a crossover rate increment and a mutation rate increment; According to the training data set, model training is performed on the initial deep Q network model to obtain the algorithm parameter dynamic optimization model.

6. The method of claim 5, wherein, According to the non-dominated sorting genetic algorithm, and the obtained training population, initial deep Q network model, training crossover rate and training mutation rate, a training data set is generated, which comprises: According to the training population and the non-dominated sorting genetic algorithm, first state data is calculated; The first state data is input into the initial deep Q network model to obtain action data; According to the action data, the training crossover rate and the training mutation rate, a new training crossover rate and a new training mutation rate are calculated; According to the new training crossover rate, the new training mutation rate, the training population and the non-dominated sorting genetic algorithm, a target population is generated; According to the training population and the target population, reward data is calculated; According to the target population and the non-dominated sorting genetic algorithm, second state data is calculated; The first state data, the action data, the reward data and the second state data are combined to form a training data; The iteration number is updated; If the updated iteration number is less than a preset iteration number threshold, the target population is taken as a new training population, and the above steps are repeated until the updated iteration number is equal to the preset iteration number threshold, a preset training number of data is selected from all training data to form the training data set.

7. The method of claim 6, wherein, The calculation of the reward data according to the training population and the target population comprises: The Euclidean distance between each two individuals in the target population is calculated to obtain a second Euclidean distance data set; The variance of the second Euclidean distance data set is taken as a diversity parameter; The non-dominated sorting genetic algorithm is used to perform non-dominated sorting on the training population to obtain a third Pareto front sequence; The non-dominated sorting genetic algorithm is used to perform non-dominated sorting on the target population to obtain a fourth Pareto front sequence; A third hyper volume of a first Pareto front in the third Pareto front sequence and a fourth hyper volume of a first Pareto front in the fourth Pareto front sequence are calculated; The difference between the fourth hyper volume and the third hyper volume is taken as a convergence quality parameter; According to the diversity parameter and the convergence quality parameter, the reward data is calculated.

8. A device for determining parameters of a suspension system, characterized in that The method comprises the steps of: An acquisition module is configured to acquire an initial crossover rate, an initial mutation rate, and a multi-objective optimization model of a suspension system. A processing module is configured to: determine an initial population according to a constraint condition of the multi-objective optimization model; solve the multi-objective optimization model according to a non-dominated sorting genetic algorithm, an algorithm parameter dynamic optimization model, the initial population, the initial crossover rate, and the initial mutation rate to obtain target suspension system parameters; wherein the algorithm parameter dynamic optimization model is a preselected deep Q network model configured to determine a crossover rate increment and a mutation rate increment according to state data of a population, and the state data of the population comprises a population diversity index, a convergence index, and a dominance relationship distribution index; the processing module is specifically configured to calculate the state data of the initial population according to the initial population and the non-dominated sorting genetic algorithm; determine the crossover rate increment and the mutation rate increment according to the state data of the initial population and the algorithm parameter dynamic optimization model; calculate a target crossover rate and a target mutation rate according to the crossover rate increment, the mutation rate increment, the initial crossover rate, and the initial mutation rate; generate a new initial population according to the target crossover rate, the target mutation rate, an objective function of the multi-objective optimization model, and the non-dominated sorting genetic algorithm; determine whether a convergence condition is met; if the convergence condition is not met, repeat the above steps by taking the target crossover rate as a new initial crossover rate and the target mutation rate as a new target mutation rate until the convergence condition is met, and select target suspension system parameters from the new initial population.

9. An electronic device, comprising: The method comprises the steps of: a processor, a memory, and a communication interface; the memory is configured to store executable instructions of the processor; wherein the processor is configured to execute the executable instructions to perform the method for determining suspension system parameters according to any one of claims 1 to 7.

10. A readable storage medium, having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method for determining suspension system parameters according to any one of claims 1 to 7.

11. A computer program product, characterised in that, The computer program is executed by the processor to implement the method for determining suspension system parameters according to any one of claims 1 to 7.

Images