Lubricating oil process optimization method and system for wear resistance and noise reduction performance

Abstract

The invention provides a lubricating oil process optimization method and system for wear resistance and noise reduction performance, and the method comprises the steps: constructing a fitness function according to a to-be-optimized lubricating oil process parameter vector, a feasible region and a plurality of target parameters; converting the particle position vector into a process parameter vector through a mapping function; on the basis of a particle swarm historical optimal position set and a dynamic weight adjustment mechanism of expert priori knowledge, automatically adjusting the weight of each target in a fitness function according to an optimization process; calculating a feasible solution according to the historical experiment database and the process parameter vector, and introducing a local neighborhood optimal guide source for each particle through the feasible solution; estimating a feasible domain center by adopting a self-adaptive layered penalty term and combining a historical feasible data set, and constructing a double-constraint processing mechanism; constructing a speed updating equation; calculating a comprehensive premature index according to the position distribution diversity index and the fitness distribution concentration index; and the intelligence and automation level of formula design can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent optimization, in particular to a lubricating oil process optimization method and system for anti-wear and noise reduction performance. BACKGROUND

[0002] In the modern development process of lubricating oil formula, how to achieve the optimal balance between anti-wear performance, noise reduction effect and cost control has been an important challenge faced by the industry and scientific research. The process parameters of lubricating oil not only involve continuous variables (such as base oil viscosity, concentration of various additives), but also include discrete variables (such as additive types), and are strictly limited by multiple physical constraints and safety limits. With the increasing complexity of material systems and the highly heterogeneous formula space, traditional experience-based design methods are inefficient and difficult to meet the fine optimization needs under multi-objective and strong constraint conditions.

[0003] Conventional optimization methods such as standard particle swarm optimization (PSO), weighted sum method or non-dominated sorting genetic algorithm II (NSGA-II) have significant shortcomings when dealing with similar problems: on the one hand, they have limited ability to balance multi-objective conflicts, usually using fixed weights, which cannot dynamically adapt to changes in the importance of optimization; on the other hand, these algorithms generally lack support for mixed parameter types, especially when dealing with discrete variables, there is semantic distortion, which may produce intermediate solutions without physical meaning. In addition, the existing methods have relatively rough processing means for process constraints, mostly using static penalty mechanisms, which lack search guidance and are prone to fall into local optima, while lacking mechanisms to dynamically detect premature convergence of algorithms and maintain population diversity, resulting in poor stability and low practicality of the final results. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a lubricating oil process optimization method and system for anti-wear and noise reduction performance, which integrates multi-source guidance information, dynamic weight adjustment, mixed parameter processing and premature detection mechanisms, establishes a closed-loop optimization framework for anti-wear and noise reduction performance of lubricating oil process suitable for actual engineering needs, and improves the intelligent and automated level of formula design.

[0005] In a first aspect, the embodiments of the present application provide a lubricating oil process optimization method for anti-wear and noise reduction performance, which comprises: defining a lubricating oil process parameter vector to be optimized and a feasible region, the feasible region being used to represent the physically feasible range of process parameters; constructing a fitness function according to the lubricating oil process parameter vector to be optimized, the feasible region and multiple target parameters; defining a hybrid encoded particle position vector, and converting the particle position vector into a process parameter vector through a mapping function; adjusting the weight of each target in the fitness function according to an optimization process based on a dynamic weight adjustment mechanism of a historical optimal position set of a particle swarm and expert prior knowledge; calculating a feasible solution according to a historical experiment database and the process parameter vector, and introducing a local neighborhood optimal guide source for each particle through the feasible solution; adopting an adaptive hierarchical penalty term, and combining a historical feasible data set to estimate a feasible region center, and constructing a dual constraint processing mechanism; constructing a velocity update equation containing multi-source guide information according to the particle position vector, a local neighborhood optimal guide term, a feasible region center guide term and an adaptive disturbance term; calculating a position distribution diversity index and a fitness distribution concentration index; calculating a comprehensive precocity index according to the position distribution diversity index and the fitness distribution concentration index.

[0006] In a second aspect, an embodiment of the present application provides a lubricating oil process optimization system for anti-wear and noise reduction performance, which comprises: a definition module configured to define a lubricating oil process parameter vector to be optimized and a feasible region, the feasible region being used to represent a physically feasible range of process parameters; a fitness function construction module configured to construct a fitness function according to the lubricating oil process parameter vector to be optimized, the feasible region and multiple target parameters; a conversion module configured to define a hybrid encoded particle position vector, and convert the particle position vector into a process parameter vector through a mapping function; an adjustment module configured to adjust the weight of each target in the fitness function according to an optimization process based on a dynamic weight adjustment mechanism of a historical optimal position set of a particle swarm and expert prior knowledge; a feasible solution calculation module configured to calculate a feasible solution according to a historical experiment database and the process parameter vector, and introduce a local neighborhood optimal guide source for each particle through the feasible solution; a dual constraint processing mechanism construction module configured to adopt an adaptive hierarchical penalty term, and combine a historical feasible data set to estimate a feasible region center, and construct a dual constraint processing mechanism; a velocity update equation construction module configured to construct a velocity update equation containing multi-source guide information according to the particle position vector, a local neighborhood optimal guide term, a feasible region center guide term and an adaptive disturbance term; A plurality of index calculation modules are configured to calculate a position distribution diversity index and a fitness distribution concentration index. A comprehensive precocity index calculation module is configured to calculate a comprehensive precocity index according to the position distribution diversity index and the fitness distribution concentration index.

[0007] The embodiment of the present application provides a lubricating oil process optimization method and system for anti-wear and noise reduction performance, comprising: defining a to-be-optimized lubricating oil process parameter vector and a feasible region, the feasible region being used to represent a physically feasible range of the process parameters; constructing a fitness function according to the to-be-optimized lubricating oil process parameter vector, the feasible region and a plurality of target parameters; defining a hybrid coded particle position vector, and converting the particle position vector into the process parameter vector through a mapping function; automatically adjusting the weight of each target in the fitness function according to the optimization process based on a dynamic weight adjustment mechanism of a historical optimal position set of the particle swarm and expert prior knowledge; calculating a feasible solution according to a historical experiment database and the process parameter vector, and introducing a local neighborhood optimal guide source for each particle through the feasible solution; constructing a dual constraint processing mechanism by using an adaptive layered penalty term and combining a historical feasible data set to estimate the center of the feasible region; constructing a velocity update equation containing multi-source guide information according to the particle position vector, the local neighborhood optimal guide term, the center guide term of the feasible region and an adaptive disturbance term; calculating a position distribution diversity index and a fitness distribution concentration index; calculating a comprehensive precocity index according to the position distribution diversity index and the fitness distribution concentration index; fusing multi-source guide information, dynamic weight adjustment, hybrid parameter processing and precocity detection mechanism, establishing a lubricating oil process closed-loop optimization framework suitable for actual engineering requirements and oriented to anti-wear and noise reduction performance, and improving the intelligentization and automation level of formula design.

[0008] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and achieved by the structure particularly pointed out in the description, claims and drawings.

[0009] In order to make the above-mentioned objects, characteristics and advantages of the present application more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creating any creative labor.

[0011] Figure 1 The flow chart of the lubricating oil process optimization method for anti-wear and noise reduction performance provided for the first embodiment of the application is shown in the figure. Figure 2 The dynamic weight self-adaptive adjustment process schematic diagram provided for the first embodiment of the application is shown in the figure. Figure 3 The different optimization algorithm convergence curve comparison schematic diagram provided for the first embodiment of the application is shown in the figure. Figure 4 The multi-objective performance comparison schematic diagram of different algorithm optimization results provided for the first embodiment of the application is shown in the figure. Figure 5 The constraint violation situation comparison schematic diagram of different algorithms provided for the first embodiment of the application is shown in the figure. Figure 6 The improved PS0 exploration ability in the parameter space schematic diagram provided for the first embodiment of the application is shown in the figure. Figure 7 The lubricating oil process optimization system schematic diagram provided for the second embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0012] To make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme of the application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are some embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0013] The application proposes a lubricating oil process optimization method and system for anti-wear and noise reduction performance, defines a complete closed-loop iterative optimization process, and constructs a full-process optimization framework from initialization to final output, as follows: 1) Initialize the particle swarm and parameters Perform initial settings for algorithm running, including: preset all hyperparameters; load the historical experiment database , calculate the feasible region center estimation vector ; randomly initialize the particle position, which should meet the range of the feasible region after being mapped by the mapping function; initialize the particle velocity vector as a 0 vector, and randomly initialize the individual historical optimal position and the global historical optimal position.

[0014] 2) Execute the iterative optimization loop Loop the optimization process within the maximum number of iterations, and the iterative optimization loop process is as follows according to the standard particle swarm optimization algorithm: Map the particle position to the actual process parameters, and calculate the fitness function for fitness evaluation; update the individual historical optimal position and the global historical optimal position; update the dynamic weight coefficient; find the local neighborhood optimal guide source; calculate the constraint violation penalty value for the next iteration of the fitness function calculation; update the particle velocity vector and position vector; perform premature detection and diversity enhancement processing, and judge whether the stop condition is met; the iteration number is incremented by 1.

[0015] 3) Output the optimization result after meeting the stop condition Output the process parameter vector corresponding to the global historical optimal position.

[0016] In order to facilitate the understanding of the present embodiment, the following will introduce the embodiment of the present application in detail.

[0017] Embodiment one: Figure 1 The process flow chart of the lubricating oil process optimization method for anti-wear and noise reduction performance provided by the first embodiment of the present application.

[0018] Referring to Figure 1 The method comprises the following steps: Step S101, defining the lubricating oil process parameter vector to be optimized and the feasible region, the feasible region being used to represent the physically feasible range of the process parameters; Step S102, constructing a fitness function according to the lubricating oil process parameter vector to be optimized, the feasible region and a plurality of target parameters; The lubricating oil process parameter optimization is faced with challenges such as multi-objective conflict, large difference in index dimension, mixed parameter types and strong constraints, and specifically involves multiple optimization objectives such as wear resistance, noise reduction and cost, the units of the indexes such as wear amount, noise and cost are different, the process parameters include both continuous variables such as viscosity and discrete variables such as additive types, and are subject to strict constraints such as viscosity range and additive safety limit.

[0019] When dealing with such problems, the conventional particle swarm optimization algorithm usually adopts a single objective weighted sum or linear normalization method to deal with multiple objectives, which is difficult to effectively balance the conflicting objectives, for example, it is often difficult to balance wear resistance and cost, in addition, the conventional method has weak processing ability for mixed parameter types and poor processing effect for strong constraints, which can easily lead the algorithm to fall into a local optimal solution or produce invalid schemes that violate the process constraints.

[0020] The application defines a multi-objective dynamic weighted fitness function as an optimization target, and clearly defines the lubricating oil process parameter vector to be optimized, while eliminating the differences caused by different dimensions of each target value through normalization processing. Specifically, the lubricating oil process parameter vector to be optimized is defined, including continuous variables and discrete variables, and the value of the vector is limited in the feasible region determined by the process constraints, the dimension and type of the parameter vector are defined, and the physically feasible value range and safety limit value are set to ensure that the formula generated by the optimization process meets the engineering practical constraints. Referring to formula (1): (1) Wherein, is the lubricating oil process parameter vector to be optimized, the dimension is ; is the total number of process parameters to be optimized.

[0021] In one embodiment, , that is, 6 types of process parameters are optimized, respectively: is the base oil viscosity, continuous type, unit: cSt, such as the value range is [40, 100]; is the anti-wear additive A concentration, continuous type, unit: wt%, such as the value range is [0.1, 2.0]; is the anti-wear additive B concentration, continuous type, unit: wt%, the value range is [0.05, 1.5]; is the noise reduction additive type, discrete type, and the integer value represents different types (such as 1, 2, 3); is the noise reduction additive concentration, continuous type, unit: wt%, such as the value range is [0.2, 3.0]; is the extreme pressure additive concentration, continuous type, unit: wt%, such as the value range is [0.1, 1.0].

[0022] Wherein, the feasible region of the process parameter is defined as , which is the feasible region defined by the process constraints, such as , , the feasible region defines the physically feasible range of the process parameter, such as the viscosity range and the safety limit value of the additive, to ensure that the optimization result meets the engineering practical constraints and avoids generating invalid or dangerous formula.

[0023] Step S103, define the mixed encoding particle position vector, and convert the particle position vector into the process parameter vector through the mapping function; Specifically, the lubricant process parameter vector to be optimized contains both continuous and discrete variables, and is therefore heterogeneous. Conventional particle swarm optimization algorithms typically use a single real-number encoding scheme and cannot directly process discrete parameters. Forcibly using real numbers to optimize discrete parameters can produce meaningless intermediate values. For example, the presence of decimals in the additive type can disrupt the algorithm's convergence performance.

[0024] The present invention adopts a segmented hybrid coding strategy to solve the discrete parameter processing problem, so that the algorithm can optimize both continuous and discrete parameters at the same time, as follows: 1) Define the hybrid coded particle position vector The particle position vector is divided into continuous parameter sub-vectors and discrete parameter sub-vectors. During the update process of the particle swarm optimization algorithm, the continuous parameter sub-vector is directly processed as a real number, and the discrete parameter sub-vector is also temporarily processed as a real number, allowing the algorithm to process continuous and discrete process parameters at the same time, referring to formula (2): (2) in, For the The particle in The position vector of the iteration, dimension is ; For the The particle in The continuous parameter subvectors of the iterations, i.e. ,Include Continuous process parameters, such as viscosity, concentration, etc.; For the The particle in The discrete parameter subvector of the iteration, that is, ,Include discrete process parameters, such as additive type; 、 、 、 、 、 Both The form, definition is the i-th particle in the The position component of the j-th dimension of the iteration, ; is the particle index, ; is the particle swarm size; is the number of iterations; is the number of consecutive parameters.

[0025] The particle position vector is converted into a physically meaningful actual process parameter vector by a mapping function. In the mapping process, the continuous parameter component remains unchanged, and the discrete parameter component is converted into a corresponding integer value by rounding operation, ensuring that the output process parameters meet the physical constraints of discrete variables, as shown in equation (3): (3) wherein, is the actual process parameter vector used for fitness evaluation of the i-th particle in the k-th iteration; is a mapping function that converts the internal position to physically achievable process parameters; is a rounding function that converts the discrete parameter component into an integer value, such as additive type 1 / 2 / 3; denotes the transpose symbol.

[0026] It should be noted that the segmented mixed coding strategy separates the continuous parameters and discrete parameters through the particle position vector The continuous parameters directly participate in real number operations, preserving the gradient search capability of the particle swarm optimization algorithm. The discrete parameters are temporarily treated as real numbers in the algorithm, but are converted into physical integers by the mapping function. The discrete parameters retain the decimal part during the algorithm update process, allowing the particles to smoothly transition between different types. For example, if the optimal additive type switches from 1 to 2, the particle position component can be updated from 1.2 to 1.8, and finally mapped to type 2, thereby maintaining the continuous search characteristics of the particle swarm optimization algorithm while ensuring that the output solution meets the discrete constraints.

[0027] Step S104, based on the dynamic weight adjustment mechanism of the particle swarm historical optimal position set and expert prior knowledge, automatically adjusts the weight of each target in the fitness function according to the optimization process; Step S105, calculates the feasible solution based on the historical experimental database and the process parameter vector, and introduces a local neighborhood optimal guide source for each particle through the feasible solution; Step S106, adopts an adaptive hierarchical penalty term and estimates the feasible region center combined with the historical feasible data set to construct a dual constraint processing mechanism; Step S107, constructs a velocity update equation containing multi-source guide information based on the particle position vector, local neighborhood optimal guide term, feasible region center guide term, and adaptive disturbance term; Step S108, calculates the position distribution diversity index and fitness distribution concentration index; Step S109, calculates the comprehensive premature index based on the position distribution diversity index and fitness distribution concentration index.

[0028] ​Further, according to the lubricating oil process parameter vector to be optimized, the feasible region and multiple targets, a fitness function is constructed, including: The fitness function is calculated according to formula (4): (4) Wherein, is the fitness function, is the lubricating oil process parameter vector to be optimized The smaller the value of the corresponding individual fitness function is, the better the comprehensive performance of the lubricating oil is, that is, strong wear resistance, low noise, low cost and meeting the constraints; is the wear amount predicted by the pre-trained neural network model, unit: μm, representing the wear resistance of the lubricating oil; is the running noise value predicted by the pre-trained neural network model, unit: dB, representing the noise reduction performance of the lubricating oil; is the unit cost, which is calculated according to the formula and process, unit: yuan; is the minimum value of the wear amount in the historical data or the feasible region; is the maximum value of the wear amount in the historical data or the feasible region; is the minimum value of the noise in the historical data or the feasible region; is the maximum value of the noise in the historical data or the feasible region; is the minimum value of the cost in the historical data or the feasible region; is the maximum value of the cost in the historical data or the feasible region, are all in the form of is the dynamic weight coefficient of the kth target, satisfying and , with the iteration number dynamically adjusted to balance the target conflict; is the target index, , corresponding to the wear amount, corresponding to the running noise value, corresponding to the unit cost; is the lubricating oil process parameter vector to be optimized is the constraint violation penalty value of the lubricating oil process parameter vector to be optimized, representing the constraint violation penalty term, when the process parameter violates the constraint, , otherwise 0.

[0029] Specifically, a fitness function is defined to integrate the three objectives of wear resistance, noise reduction, and cost, and through normalization processing, the dimensional differences of the objective values are eliminated, and the conflicts between the objectives are balanced by combining dynamic weight coefficients, and the process constraints are processed by combining constraint penalty terms. The smaller the fitness function value is, the better the comprehensive performance of the lubricating oil is. Referring to formula (4).

[0030] It should be noted that the fitness function integrates the normalization processing of the three objectives, dynamic weights, and constraint penalties: in terms of normalization, the dimensional differences are eliminated by , so that the objective values are dimensionless and proportional; in terms of dynamic weights, corresponds to the wear weight, corresponds to the noise weight, corresponds to the cost weight, and satisfies , the weight is adaptively adjusted with iterations, encouraging exploration at the beginning and focusing on key objectives at the end; in terms of penalty terms, the constraint violation penalty term applies a penalty to the solution that violates the constraint, forcing the search direction to be biased towards the feasible region.

[0031] It should also be noted that in the particle swarm optimization algorithm, "individual" refers to a candidate solution, i.e., a particle representing a combination of process parameters.

[0032] Further, a dynamic weight adjustment mechanism based on the historical optimal position set of the particle swarm and expert prior knowledge automatically adjusts the weight of each objective in the fitness function according to the optimization process, including the following steps: Step S201, analyze the distribution and dispersion of the historical optimal position set of the particle swarm individual on each objective function value, and calculate the coefficient of variation of each objective; Step S202, compare the coefficient of variation of each objective with the preset objective coefficient of variation expectation reference value, and calculate the distribution similarity score by an exponential decay function; wherein the distribution similarity score is used to measure the closeness between the actual distribution state and the expected distribution state; Step S203, calculate the normalized dynamic weight coefficient according to the static preference weight of each objective and the distribution similarity score; Step S204, adjust the relative importance of each objective in the fitness function by the normalized dynamic weight coefficient.

[0033] Specifically, the importance of the lubricating oil optimization objective is not fixed and will change dynamically with the optimization process, and the optimal trade-off relationship between the objectives is unknown. The conventional particle swarm optimization algorithm often uses fixed weight linear weighting or multi-objective processing method, which cannot dynamically adjust the importance of each objective according to the distribution state of the solution set during the search process, resulting in rigid search direction and difficulty in obtaining high-quality optimal trade-off solution.

[0034] This application uses a dynamic weight adjustment mechanism based on the historical optimal position distribution of the particle swarm and expert prior knowledge to automatically adjust the weight of each objective in the fitness function according to the optimization process. The details are as follows: Analyze the distribution dispersion of the individual historical optimal position set of the particle swarm on each objective function value, and quantify the distribution state of the objective on the approximate Pareto frontier by calculating the coefficient of variation of each objective, referring to formula (5): (5) in, For the The goal is The coefficient of variation of the iteration reflects the degree of dispersion of the target value on the approximate Pareto frontier; For the The particle in The historical best position of the iteration; For the The objective function, represents the wear amount prediction function, represents the noise prediction function, represents the cost calculation function; To distinguish The particle index, ; is the standard deviation calculation function; is the mean calculation function.

[0035] The calculated coefficient of variation of each target is compared with the preset expected reference value of the target coefficient of variation, and the distribution similarity score is calculated using the exponential decay function to measure the closeness between the actual distribution state and the expected distribution state, referring to formula (6): (6) in, For the The goal is The distribution similarity score of the iteration is ; For the preset Expected reference value of the coefficient of variation of each target; is the scaling factor, which is a hyperparameter that controls the sensitivity to deviations from the reference value, such as ; is the natural exponential function; It is the absolute value operation.

[0036] It should be noted that the expected reference value of the coefficient of variation By artificial preset, key objectives such as wear resistance are required to be concentrated in the Pareto front distribution, with a small coefficient of variation, and set to be small. , secondary objectives such as cost are allowed to be distributed dispersedly, and a larger , coefficient of variation Reflects the degree of dispersion of the target value. The smaller the value, the more concentrated the distribution. For example, if wear resistance is the core target, the preset , to ensure that the difference in wear is small, the cost is relatively minor, and the preset , allowing the cost value to fluctuate more in the solution set.

[0037] Combining the static preference weight of each target and the calculated distribution similarity score, the normalized dynamic weight coefficient is calculated to adjust the relative importance of each target in the fitness function, referring to formula (7): (7) in, is the dynamic weight coefficient of the kth target; For the preset The static preference weight of each target is set according to the process requirements; for example, the emphasis on wear resistance can be set , must meet and ; For the The goal is The distribution similarity score of the iteration; is the target index different from k.

[0038] It should be noted that if near , indicating that the target distribution is in line with expectations. , weight By static preference Leading; if Significant deviation If the wear distribution is too dispersed, Reduce, reduce the weight of the target, force the algorithm to strengthen the optimization, and then automatically identify the conflict intensity between the targets. For example, when cost and wear resistance are strongly negatively correlated, the coefficient of variation of the two Will increase synchronously, and then dynamically reduce , giving priority to ensuring anti-wear performance and avoiding the degradation of the compromise solution quality caused by fixed weights in conventional methods.

[0039] Furthermore, a feasible solution is calculated based on the historical experimental database and the process parameter vector, and a local neighborhood optimal guidance source is introduced for each particle through the feasible solution, including the following steps: Step S301, constructing a similarity function between two process parameter vectors; Step S302, calculating the mixing distance between two process parameter vectors; Step S303, retrieving a feasible solution from the historical experiment database, which has a similarity to the current particle actual process parameter vector exceeding a set threshold; Step S304, selecting a process parameter vector with the optimal value of the fitness function from the feasible solutions as the local optimal guidance source of the current particle.

[0040] Specifically, the lubricating oil formula space is huge and complex. The conventional particle swarm optimization algorithm only relies on the individual historical optimal position of the particle and the global historical optimal position of the population to guide the search, which is easy to fall into local optimum and slow to converge in high-dimensional mixed parameter space.

[0041] The application adopts the construction of process similarity measurement by using the historical experiment database, and introduces the guidance of local optimal neighborhood for each particle, specifically as follows: The similarity function between two process parameter vectors is constructed, and the similarity is calculated based on the mixed parameter distance, and the distance is converted into a similarity value between 0 and 1 through an exponential decay function. The greater the value, the more similar the process is. Refer to formula (8): (8) Wherein, is the process parameter vector ; is the similarity of , the value range is , the greater the value, the more similar; is the process parameter vector to be compared ; is the process parameter vector to be compared ; is the mixed parameter distance of the process parameter vector and , the smaller the value, the more similar the process is; is the distance scaling coefficient, which controls the rate of similarity decay with distance, for example, .

[0042] The mixed distance between two process parameter vectors is calculated, which integrates the Euclidean distance component of the normalized continuous parameters and the Hamming distance component of the discrete parameters. The smaller the value, the more similar the two process schemes are. Refer to formula (9): (9) Wherein, is the mixed parameter distance, the smaller the value, the more similar the process is; is the value of the th continuous parameter in the vector ; is the value of the th continuous parameter in the vector ; For the The value range of a continuous parameter is used for normalization processing; is a vector Middle The value of a discrete parameter; is a vector Middle The value of a discrete parameter; is the discrete parameter difference function, which is defined as shown in formula (10): (10) Retrieve feasible solutions whose similarity with the actual process parameter vector of the current particle exceeds the set threshold in the historical experiment database, and select the process parameter vector with the best fitness function value from the feasible solutions as the optimal guidance source in the local neighborhood of the particle, as shown in formula (11): (11) in, For the The optimal process parameter vector of the local neighborhood of the particle at the tth iteration; It is a historical experimental database or a high-confidence proxy model prediction set, containing feasible process parameter vectors And the fitness function of the feasible process parameter vector ; is the similarity threshold, only solutions with similarity higher than this value are considered, for example, ; For the The process parameter vector of the particle at the tth iteration; Indicates that the return condition is met And the smallest target .

[0043] It should be noted that the hybrid parameter distance is used to calculate the heterogeneous characteristics of continuous parameters such as concentration and viscosity and discrete parameters such as additive type in lubricant formulations. The Euclidean distance of the normalized continuous parameter is fused with the Hamming distance of the discrete parameter; for the continuous parameter, The term is used to normalize the dimensions and eliminate the influence of the difference in the magnitude of the concentration and viscosity. For discrete parameters, the The term directly quantifies the type difference, avoiding the semantic distortion caused by forcibly converting discrete values ​​into numerical ones, and then constructing a recipe similarity measure with clear physical meaning to identify potential high-quality solutions.

[0044] Furthermore, an adaptive hierarchical penalty term is adopted and the feasible region center is estimated in combination with the historical feasible data set to construct a dual constraint processing mechanism, which includes the following steps: Step S401, for each constraint condition, calculate the normalized violation distance, and construct an adaptive hierarchical penalty term through a nonlinear function and a time-varying penalty intensity coefficient; Step S402, construct a penalty intensity coefficient, which is used to adjust the overall intensity of the hierarchical penalty term; Step S403, based on the historical feasible data set, calculate the mean value of each process parameter component; Step S404, according to the mean value of each process parameter component, obtain a feasible region center estimation vector; wherein the feasible region center estimation vector is used to represent the average position of the historical feasible solution in the parameter space; Step S405, take the feasible region center estimation vector as a reference point for guiding the particle to the center of the feasible region.

[0045] Specifically, the lubricating oil process has strict constraint conditions, resulting in a large number of infeasible regions in the parameter space. The conventional penalty function method often uses a fixed constant or a penalty based on linear violation distance, which is difficult to effectively guide the particle to return to the feasible region, and may destroy the gradient information and reduce the search efficiency.

[0046] The present application adopts a hierarchical adaptive penalty term and estimates the feasible region center combined with historical feasible data to construct a dual constraint processing mechanism, specifically as follows: Calculate the penalty value of the process parameter vector violating the constraint, for each constraint condition, calculate the normalized violation distance, and construct an adaptive hierarchical penalty term through a nonlinear function and a time-varying penalty intensity coefficient, as shown in formula (12): (12) Wherein, is the constraint violation penalty value of the process parameter vector ; is the process parameter vector to be evaluated; is the total number of constraint conditions; is the actual value of the process parameter vector on the first constraint condition; is the midpoint value of the first constraint interval , and the calculation method is represented as ; for example, the midpoint of the viscosity range is 75; is the lower limit of the first constraint interval; is the upper limit of the first constraint interval; is the half-width of the first constraint interval, for example, the viscosity range half-width is 10; For the The reference range of the constraints is used for normalization, for example, ; is the penalty exponent, which makes the penalty grow nonlinearly with the degree of violation, for example, ; represents the maximum value function; is the penalty intensity coefficient for the tth iteration.

[0047] A penalty intensity coefficient that increases with the number of iterations is constructed to adjust the overall intensity of the penalty term. In the early stages of optimization, moderate exploration of the area outside the constraint boundary is allowed, while in the later stages, the penalty is strengthened to drive the particles back to the feasible region, as shown in formula (13): (13) in, is the base penalty intensity, e.g. ; is the penalty growth coefficient, for example, ; is the maximum number of iterations.

[0048] Based on the historical feasible data set, the mean of each process parameter component is calculated to obtain the feasible region center estimation vector, which represents the average position of the historical feasible solution in the parameter space and serves as a reference point to guide the particles toward the center of the feasible region, as shown in formulas (14) and (15): (14) (15) in, is the estimated vector of the feasible region center; For historical feasible data sets, Screen the process parameter vector that meets all constraints ; For historical feasible data sets The size of represents the number of samples; For historical feasible data sets The process parameter vector in ; It is a historical experiment database.

[0049] It should be noted that from the historical experimental database The similarity between the selected particles and the current particles exceeds the threshold The feasible solution of , when the particle falls into the local optimum, It can provide a springboard direction for particles with similar physical properties but better performance, for example, if the performance of the current particle is stagnant due to the wrong type of additive, It can be guided to switch to a more effective additive type at a similar viscosity in history, achieving a targeted crossing of discrete parameters.

[0050] Further, according to the particle position vector, the local neighborhood optimal guiding term, the feasible region center guiding term and the adaptive disturbance term, a speed update equation containing multi-source guiding information is constructed, including the following steps: Step S501, on the basis of the conventional particle swarm optimization relying on individual historical optimum and global historical optimum, a multi-source guiding speed update equation is constructed through a local neighborhood optimal guiding term, a feasible region center guiding term and an adaptive disturbance term; Step S502, according to the position of the fitness value of the particle relative to the current population fitness range, the disturbance intensity of each dimension parameter is dynamically calculated; Step S503, the particle position vector is updated through the multi-source guiding speed update equation.

[0051] Specifically, the conventional particle swarm optimization algorithm only relies on the individual historical optimal position and the global historical optimal position to update the particle speed, which has insufficient exploration ability in high-dimensional, multi-objective and strongly constrained lubricating oil process optimization problems, is prone to premature convergence, and fails to effectively utilize the field knowledge guiding source and the feasible region center information.

[0052] The present application fuses hybrid coding, local neighborhood guidance, feasible region center gravity and adaptive disturbance mechanism, and constructs a speed update equation containing multi-source guiding information, specifically as follows: The multi-source guiding speed update equation is constructed, on the basis of the conventional particle swarm optimization relying on individual historical optimum and global historical optimum, a local neighborhood optimal guiding term, a feasible region center guiding term and an adaptive disturbance term are added to construct a particle speed update equation, forming a multi-source guiding mechanism, as shown in formula (16): (16) Wherein, is the velocity vector of the i-th particle in the t-th iteration; is the velocity vector of the i-th particle in the t-th iteration; is the inertia weight, controlling the retention proportion of historical speed, for example, ; is the acceleration constant, respectively controlling the guiding intensity of each term, for example, , , , , , is the acceleration constant, respectively controlling the guiding intensity of each term, for example, , , ​、 、 ; 、 、 、 、 for interval uniform distribution of random numbers, introducing randomness; for the th particle position vector at the tth iteration; for the th particle individual historical optimal position; for the global historical optimal position of the particle swarm; for the process parameter vector mapping back to the internal position of the algorithm, continuous parameters directly take values, discrete parameters take integer values corresponding to real numbers; for mapping back to the internal position of the algorithm; for the th particle perturbation intensity diagonal matrix at the tth iteration, dimension ; for dimensional random vector, elements are uniformly distributed in .

[0053] It should be noted that in the setting of the preferred value of the acceleration constant, and follow the classical PSO standard value , to ensure basic convergence; and are set to lower values and , to avoid the excessive dominance of domain knowledge and feasible region gravity , minimize the influence of random disturbance, and only activate when necessary.

[0054] According to the position of the fitness value of the particle relative to the current population fitness range, the perturbation intensity of each dimension parameter is dynamically calculated. The particle with poor fitness obtains larger perturbation intensity to enhance exploration, and the particle with good fitness has smaller perturbation intensity to maintain development, as shown in formulas (17) and (18): (17) (18) where, is the th particle at the The perturbation intensity of the dimensional parameter; is a diagonal matrix constructor; is the baseline perturbation intensity, for example, ; is the disturbance attenuation coefficient, for example, ; For the The fitness value of a particle in the tth iteration; is the minimum fitness value of all particles in the tth iteration; is the maximum fitness of all particles in the tth iteration.

[0055] According to the updated velocity vector, the particle position vector is updated in the standard form, as shown in formula (19): (19) in, For the The particle in The position vector of the iteration; For the The position vector of a particle at the tth iteration; For the The particle in The velocity vector for the iteration.

[0056] It should be noted that Represents the midpoint of the interval, Characterize half-width, Characterize the reference range and jointly construct the buffer zone of constraints, When the item is established, it indicates a slight crossing of the boundary, the penalty is 0, and boundary exploration is allowed. After exceeding the half width, the penalty is The nonlinear growth of the term and the sharp increase in penalties when the boundary is seriously exceeded; for example, under the viscosity constraint In, if Indicates slight out-of-bounds, mild punishment, and retains gradient information; if To represent serious out-of-bounds situations, exponential penalties are used to force particles to return to the feasible region.

[0057] It should also be noted that The calculation allows moderate out-of-bounds exploration in the early iterations and strict punishment in the later iterations, and the center of the feasible region By estimating the mean of historical feasible solutions, a gravitational anchor point of the feasible domain is provided. In the early stage of optimization, particles can explore potential optimal solutions outside the constraint boundary, such as new formulas with additive concentrations slightly exceeding the safety limit. and enhanced Pull back to the feasible region.

[0058] Furthermore, the position distribution diversity index and the fitness distribution concentration index are calculated, including the following steps: Step S601, the position distribution diversity index is obtained by calculating the ratio of the standard deviation of each dimension parameter component to the dynamic range and averaging; Step S602, the individual optimal fitness mean is calculated; Step S603, the difference between the individual optimal fitness mean and the optimal value is calculated; Step S604, the fitness distribution concentration index is calculated according to the ratio of the difference to the fitness range.

[0059] Specifically, in the later stage of lubricating oil optimization search, the particle swarm may gather in the local optimal area, losing the ability to explore new areas, resulting in missing the global optimal solution. The conventional particle swarm optimization algorithm lacks an effective mechanism to dynamically detect this premature convergence phenomenon, and also lacks a strategy to actively inject diversity to help the particle swarm jump out of the local optimal trap.

[0060] The present application designs a comprehensive premature detection index based on the position distribution and fitness change of the particle swarm, and triggers a diversity enhancement operation based on historical knowledge when premature convergence is detected, as follows: The position distribution diversity index is obtained by calculating the ratio of the standard deviation of each dimension parameter component to the dynamic range and averaging, and the smaller the value, the higher the particle aggregation degree, thereby quantifying the spatial distribution dispersion degree of the particle swarm in the parameter space, as shown in formula (20): (20) Wherein, is the position distribution diversity index of the particle swarm in the tthiteration, and the value range is , and the smaller the value, the more the particles are aggregated; is the position component of the nthparticle in the tthiteration and the mthdimension parameter; is the dynamic range of the mthdimension parameter. is the dynamic range of the mthdimension parameter. The fitness distribution concentration index is obtained by calculating the ratio of the difference between the individual optimal fitness mean and the optimal value to the fitness range, and the smaller the value, the closer the solution quality, thereby quantifying the distribution concentration degree of the individual historical optimal fitness value of the particle swarm, as shown in formula (21):

[0061] (21) Wherein, is the fitness distribution concentration index in the tthiteration, and the value range is , and the smaller the value, the closer the solution quality; is the individual historical optimal fitness value of the nthparticle in the tthiteration; is the individual historical optimal fitness value of the nthparticle in the tthiteration; ​​The minimum value of the individual optimal fitness of all particles in the tth iteration; The maximum value of the individual optimal fitness of all particles in the tth iteration; The individual historical optimal process parameter vector of the tth particle. The individual historical optimal process parameter vector of the tth particle.

[0062] The fusion of the position distribution diversity index and the fitness distribution concentration index forms a comprehensive premature convergence index. If the index is continuously lower than a set threshold value for multiple generations, it is determined that premature convergence occurs, and a diversity enhancement operation is triggered, that is, the positions of the worst particles in fitness are reset to the feasible solutions randomly extracted from the historical experimental database, and their velocities are reset. For reference, formula (22) can be used.

[0063] Further, according to the position distribution diversity index and the fitness distribution concentration index, a comprehensive premature convergence index is calculated, including: calculating the comprehensive premature convergence index according to formula (22): (22) Wherein, The comprehensive premature convergence index, the value range is ; The position diversity weight coefficient, the value range is .

[0064] For example, ; The position distribution diversity index of the particle swarm in the tth iteration, The fitness distribution concentration index of the tth iteration.

[0065] The premature convergence threshold value, for example, ; The continuous detection generation threshold value, for example, ; The number of particles to be reset; The reset particle proportion coefficient, for example, ; The upward rounding function; The random process parameter vector is mapped back to the internal position of the algorithm. The continuous parameters are unchanged, and the integer values of the discrete parameters correspond to real numbers; The feasible process scheme randomly extracted from the historical experimental database The zero vector; The fitness value of the random scheme.

[0066] If continuously generations, select the worst particles in fitness and reset the positions​​ , and reset the speed , and then update the individual historical optimal, expressed as , .

[0067] It should be noted that by integrating position distribution diversity and fitness distribution concentration Constructing comprehensive indicators ,in, Quantify the geometric dispersion of particles in the normalized parameter space, the discrete parameters are affected by the Hamming distance, It reflects the similarity of solution quality, couples the geometric distribution of parameter space with the fitness distribution, and avoids semantic distortion caused by forced numericalization of discrete parameters.

[0068] In one embodiment, the dynamic weight adaptive adjustment process analysis tracks the changing trends of the three objective weights (anti-wear performance, noise reduction, and cost) during the optimization iteration process and analyzes how the weights are automatically adjusted according to the optimization process. Figure 2 In the figure, the horizontal axis represents the number of iterations (dimensionless), and the vertical axis represents the weight coefficient (dimensionless). Experimental results show that in the early stages of optimization (approximately the first 30 iterations), the weights of the three objectives are relatively balanced, corresponding to the need to comprehensively consider all objectives during the algorithm's exploration phase. As optimization progresses (iterations 30-70), the weight of the cost increases significantly, reflecting the focus on key objectives during the algorithm development phase. In the later stages of optimization (after 70 iterations), the weights tend to stabilize but still undergo subtle adjustments, reflecting the algorithm's meticulous search for the optimal solution. This adaptive adjustment mechanism avoids the drawback of conventional fixed-weight methods, which are unable to adapt to changes in the optimization process. Figure 2 The grey dotted lines in the figure mark the dividing points of different optimization stages, which intuitively shows the corresponding relationship between the weight adjustment strategy and the optimization stage.

[0069] In one embodiment, the convergence curves are analyzed to evaluate the convergence performance of different optimization algorithms, and the iterative optimization efficiency of the improved particle swarm optimization algorithm of the present invention is compared with that of the standard particle swarm optimization algorithm, the weighted sum method, and the second generation non-dominated sorting genetic algorithm. Figure 3In the figure, the horizontal axis is the iteration number (unitless), and the vertical axis is the fitness function value (unitless). The experimental data shows that the improved particle swarm optimization algorithm (improved PS0) exhibits significant convergence acceleration characteristics in the early iterations, with the maximum curve slope indicating the fastest rate of decline in the target function value. In the middle of the iteration (about 40 iterations), it has approached a stable value, while other algorithms are still in the slow decline stage. In addition, the curve of the improved algorithm has reasonable fluctuations, reflecting the active escape ability of the dynamic weight adjustment mechanism from local extrema. Finally, the improved algorithm achieves the lowest fitness value, which is significantly better than the standard particle swarm optimization algorithm and the weighted sum method, indicating that the combined effect of the multi-source guidance mechanism and the adaptive perturbation term effectively avoids the premature convergence problem caused by parameter heterogeneity in conventional algorithms.

[0070] In one embodiment, a multi-objective performance comparison is conducted, focusing on three core indicators of lubricating oil: wear amount (unit: microns), noise (unit: decibels), and cost (unit: yuan), to compare the actual engineering performance of the optimization results of different algorithms. Figure 4 In the figure, the vertical axis is the wear amount, and the improved particle swarm optimization algorithm achieves the lowest wear amount, lower than the standard particle swarm optimization algorithm. In terms of noise reduction, the vertical axis is the noise, and the noise value of the improved algorithm is about 48 decibels, significantly lower than the weighted sum method. In terms of cost control, the vertical axis is the cost, and although the improved algorithm is slightly higher than the weighted sum method, the standard deviation is the smallest, indicating the best result stability. The experimental results show that the dynamic weight mechanism gives the wear amount a higher weight in the early iterations, and then focuses on cost optimization. Through the coefficient of variation, the target priority is adjusted in real time, overcoming the performance imbalance caused by the fixed weight of the weighted sum method.

[0071] In one embodiment, a constraint violation comparison is conducted to verify the algorithm's ability to handle process constraints. In Figure 5 In the figure, the horizontal axis is the algorithm grouping (unitless), and the vertical axis is the constraint violation degree (normalized value). The scatter plot shows that the violation value of the improved particle swarm optimization algorithm (improved PS0) is densely distributed in the 0.00-0.10 interval, while the other algorithm, the standard particle swarm optimization algorithm, is widely distributed in the 0.10-0.35 interval. The box plot further quantifies the difference, with the lower edge of the improved algorithm's box approaching 0.03 (25th percentile), the upper edge being 0.07 (75th percentile), and the median being 0.05. The other algorithm's box spans 0.18-0.26, with a median of 0.22. This shows that the improved algorithm has a lower constraint violation probability, and when the parameters are slightly out of bounds, the penalty term can effectively avoid the gradient distortion problem of conventional linear penalties.

[0072] In one embodiment, the parameter space exploration ability diagram is analyzed. Referring to Figure 6, the algorithm exploration behavior is visualized by a two-dimensional parameter space (base oil viscosity: 40-100 centistokes; anti-wear agent A concentration: 0.1-2.0 weight percent). Blue regions in the heat map correspond to low fitness values (<0.4, superior performance), while red regions correspond to high fitness values (>0.8, poor performance). The optimization path (white dots connected by lines) shows that the algorithm starts from the initial point (45 centistokes, 0.3 weight percent), first explores in the direction of increasing viscosity to 88 centistokes (avoiding the red trap region in the upper left corner), then reduces the viscosity and increases the concentration to the optimal interval (70-80 centistokes, 0.5-1.0 weight percent). Notably, the path has a concentration fine-tuning (1.0 0.7 0.5 weight percent) at the medium viscosity region (75 centistokes), reflecting the role of the hybrid coding strategy, where discrete parameters are smoothly transitioned through real position components, maintaining the continuity of gradient search. Finally, the algorithm stabilizes in the deep blue region (fitness ≈ 0.3), verifying the effectiveness of the feasible region center attractive term in guiding the complex constraint space.

[0073] The present application has the following effects: 1) Multi-objective dynamic weighting mechanism A dynamic weight adjustment method based on the coefficient of variation and expert knowledge is constructed to balance the conflict between anti-wear, noise reduction, and cost.

[0074] 2) Hybrid continuous and discrete coding and mapping mechanism are used to ensure the physical feasibility and smoothness of the search during optimization.

[0075] 3) A similarity function is constructed using a historical experiment database, combined with local optimal guide sources, to improve the ability to jump out of local extrema.

[0076] 4) A multi-source guided velocity update equation is constructed, combined with position and fitness distribution detection to trigger diversity enhancement operations, improving global search capability.

[0077] Example Two: Figure 7 The lubricating oil process optimization system for anti-wear and noise reduction performance provided in Example Two of the present application is shown in the schematic diagram.

[0078] Referring to Figure 7 , the system comprises: A definition module for defining the lubricating oil process parameter vector to be optimized and the feasible region, which represents the physically feasible range of process parameters; An adaptive function construction module for constructing an adaptive function according to the lubricating oil process parameter vector to be optimized, the feasible region, and multiple target parameters; A conversion module is configured to define a hybrid coded particle position vector and convert the particle position vector into a process parameter vector through a mapping function; An adjustment module is configured to automatically adjust the weight of each target in the fitness function according to an optimization process based on a dynamic weight adjustment mechanism of a particle swarm historical optimal position set and expert prior knowledge; A feasible solution calculation module is configured to calculate a feasible solution according to a historical experiment database and the process parameter vector, and introduce a local neighborhood optimal guide source for each particle through the feasible solution; A dual constraint processing mechanism construction module is configured to construct a dual constraint processing mechanism by using an adaptive hierarchical penalty term and combining a historical feasible data set to estimate a feasible region center; A velocity update equation construction module is configured to construct a velocity update equation containing multi-source guide information according to a particle position vector, a local neighborhood optimal guide term, a feasible region center guide term and an adaptive disturbance term; A plurality of index calculation modules are configured to calculate a position distribution diversity index and a fitness distribution concentration index; A comprehensive prematurity index calculation module is configured to calculate a comprehensive prematurity index according to the position distribution diversity index and the fitness distribution concentration index.

[0079] The embodiment of the present application also provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to realize the steps of the anti-wear and noise reduction performance oriented lubricating oil process optimization method provided by the above embodiment.

[0080] The embodiment of the present application also provides a computer readable medium with non-volatile program code executable by a processor, and the computer readable medium stores a computer program, and the computer program is executed by the processor to execute the steps of the anti-wear and noise reduction performance oriented lubricating oil process optimization method of the above embodiment.

[0081] Finally, it should be noted that: the above-described embodiments are merely specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, but not to limit them, and the protection scope of the present application is not limited thereto, although the above-mentioned embodiments of the present application have been described in detail, those skilled in the art should understand that any skilled person familiar with the technical field can modify or easily think of changes to the technical solutions recorded in the above-mentioned embodiments within the technical range disclosed by the present application, or make equivalent replacement to part of the technical features; and these modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and all should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A lubricating oil process optimization method for anti-wear and noise reduction performance, characterized in that: The method comprises: defining a lubricating oil process parameter vector to be optimized and a feasible domain, wherein the feasible domain is used to characterize a physically feasible range of the process parameters; Constructing a fitness function according to the lubricating oil process parameter vector to be optimized, the feasible region and a plurality of target parameters; defining a hybrid-coded particle position vector, and converting the particle position vector into a process parameter vector through a mapping function; A dynamic weight adjustment mechanism based on the particle swarm's historical optimal position set and expert prior knowledge automatically adjusts the weight of each objective in the fitness function according to the optimization process; Calculating a feasible solution based on a historical experiment database and the process parameter vector, and introducing a local neighborhood optimal guidance source for each particle through the feasible solution; Adopting adaptive hierarchical penalty terms and estimating the center of the feasible region in combination with historical feasible data sets, a dual constraint processing mechanism is constructed. Constructing a velocity update equation containing multi-source guidance information according to the particle position vector, the local neighborhood optimal guidance term, the feasible region center guidance term, and the adaptive perturbation term; Calculate the position distribution diversity index and fitness distribution concentration index; A comprehensive precocity index is calculated based on the position distribution diversity index and the fitness distribution concentration index.

2. The lubricating oil process optimization method for anti-wear and noise reduction performance according to claim 1, characterized in that: Constructing a fitness function according to the lubricating oil process parameter vector to be optimized, the feasible region, and multiple objectives, including: The fitness function is calculated according to the following formula: in, is the fitness function; is the lubricating oil process parameter vector to be optimized The fitness function of the corresponding individual; is the wear amount predicted by the pre-trained neural network model; is the running noise value predicted by the pre-trained neural network model; is the unit cost; is the minimum value of wear in historical data or the feasible domain; is the maximum value of the wear amount in the historical data or the feasible domain; is the minimum value of the noise in the historical data or the feasible domain; is the maximum value of the noise in the historical data or the feasible domain; is the minimum value of the cost in the historical data or the feasible domain; is the maximum value of the cost in the historical data or the feasible domain; Both The form, definition is the dynamic weight coefficient of the kth target, satisfying and , with the number of iterations Dynamically adjust to balance conflicting goals; is the target index, is the lubricating oil process parameter vector to be optimized Constraint violation penalty value.

3. The lubricating oil process optimization method for anti-wear and noise reduction performance according to claim 1, characterized in that: A dynamic weight adjustment mechanism based on the particle swarm's historical optimal position set and expert prior knowledge automatically adjusts the weight of each objective in the fitness function according to the optimization process, including: Analyze the distribution and dispersion of the individual historical optimal position set of the particle swarm on each objective function value, so as to calculate the coefficient of variation of each objective; Comparing the coefficient of variation of each target with a preset target coefficient of variation expected reference value, and calculating a distribution similarity score using an exponential decay function; wherein the distribution similarity score is used to measure the degree of closeness between the actual distribution state and the expected distribution state; Calculating a normalized dynamic weight coefficient based on the static preference weight of each target and the distribution similarity score; The relative importance of each objective in the fitness function is adjusted by the normalized dynamic weight coefficient.

4. The lubricating oil process optimization method for anti-wear and noise reduction performance according to claim 1, characterized in that: A feasible solution is calculated based on the historical experimental database and the process parameter vector, and a local neighborhood optimal guidance source is introduced for each particle through the feasible solution, including: Constructing a similarity function between two process parameter vectors; Calculating a mixing distance between two of the process parameter vectors; Retrieving the feasible solution whose similarity with the actual process parameter vector of the current particle exceeds a set threshold from the historical experiment database; A process parameter vector with an optimal value of the fitness function is selected from the feasible solutions as an optimal guidance source in the local neighborhood of the current particle.

5. The lubricating oil process optimization method for anti-wear and noise reduction performance according to claim 1, characterized in that: Adopting an adaptive hierarchical penalty term and combining the historical feasible data set to estimate the center of the feasible region, a dual constraint processing mechanism is constructed, including: For each constraint condition, a normalized violation distance is calculated, and the adaptive hierarchical penalty term is constructed through a nonlinear function and a time-varying penalty intensity coefficient; Constructing a penalty intensity coefficient, wherein the penalty intensity coefficient is used to adjust the overall intensity of the hierarchical penalty term; Calculating the mean of each process parameter component based on the historical feasible data set; Obtaining a feasible region center estimation vector based on the mean values ​​of the process parameter components; wherein the feasible region center estimation vector is used to represent the average position of historical feasible solutions in the parameter space; The feasible region center estimation vector is used as a reference point to guide the particle toward the feasible region center.

6. The lubricating oil process optimization method for anti-wear and noise reduction performance according to claim 1, characterized in that: According to the particle position vector, the local neighborhood optimal guidance term, the feasible region center guidance term and the adaptive perturbation term, a velocity update equation containing multi-source guidance information is constructed, including: On the basis of conventional particle swarm optimization relying on individual historical optimum and global historical optimum, a multi-source guided velocity update equation is constructed through the local neighborhood optimal guided term, the feasible region center guided term and the adaptive perturbation term; Dynamically calculate the perturbation intensity of each dimension parameter based on the position of the particle's fitness value relative to the current population fitness range; The particle position vector is updated using the multi-source guided velocity update equation.

7. The lubricating oil process optimization method for anti-wear and noise reduction performance according to claim 1, characterized in that: Calculate the position distribution diversity index and fitness distribution concentration index, including: The position distribution diversity index is obtained by calculating the ratio of the standard deviation of the parameter components of each dimension to the dynamic range and averaging them; Calculate the mean of individual optimal fitness; Calculating the difference between the mean value and the optimal value of the individual optimal fitness; The fitness distribution concentration index is calculated according to the ratio of the difference to the fitness range.

8. A lubricating oil process optimization system for anti-wear and noise reduction performance, characterized in that: The system comprises: A definition module, used to define a lubricating oil process parameter vector to be optimized and a feasible domain, wherein the feasible domain is used to characterize a physically feasible range of the process parameters; a fitness function construction module, configured to construct a fitness function according to the lubricating oil process parameter vector to be optimized, the feasible region, and a plurality of target parameters; a conversion module, configured to define a hybrid-coded particle position vector and convert the particle position vector into a process parameter vector through a mapping function; An adjustment module is used to automatically adjust the weight of each objective in the fitness function according to the optimization process based on a dynamic weight adjustment mechanism based on the particle swarm's historical optimal position set and expert prior knowledge; A feasible solution calculation module is used to calculate a feasible solution based on a historical experiment database and the process parameter vector, and introduce a local neighborhood optimal guidance source for each particle through the feasible solution; A dual-constraint processing mechanism construction module is used to adopt an adaptive hierarchical penalty term and estimate the center of the feasible region in combination with the historical feasible data set to construct a dual-constraint processing mechanism; A velocity update equation construction module is used to construct a velocity update equation containing multi-source guidance information based on the particle position vector, the local neighborhood optimal guidance term, the feasible region center guidance term and the adaptive perturbation term; Multiple indicator calculation modules are used to calculate the position distribution diversity index and fitness distribution concentration index; The comprehensive precociousness index calculation module is used to calculate the comprehensive precociousness index according to the position distribution diversity index and the fitness distribution concentration index.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable medium having a non-volatile program code executable by a processor, characterized in that The program code causes the processor to execute the method according to any one of claims 1 to 7.

Images