An optimization method for performance model parameter simulation, electronic equipment and storage medium
By combining local weighted regression and nonlinear fitting models with particle swarm optimization algorithm, the performance model parameters are optimized, solving the problem of insufficient model accuracy caused by abnormal data interference, and achieving more efficient performance indicator prediction.
Patent Information
- Application Number
- CN202511640561.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-02-27
- Estimated Expiration
- 2045-11-11
AI Technical Summary
Existing performance models suffer from insufficient accuracy during parameter optimization due to interference from outlier data and a single fitting method, making it difficult to accurately predict performance indicators.
Abnormal samples are identified and corrected by local weighted regression to generate an effective sample set. A nonlinear fitting model is established by combining polynomial spline function and radial basis function. An improved particle swarm optimization algorithm is used to find the optimal model parameters based on the correlation matrix of performance parameters and the gradient of fitting residuals.
This improved data quality and fitting accuracy, reduced redundant variables, enhanced the model's optimization efficiency and parameter reliability, and ensured the effectiveness and reliability of the optimized performance model parameters.
Smart Images
Figure CN121094654B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of simulation optimization, and more particularly, to a performance model parameter simulation optimization method, an electronic device and a storage medium. BACKGROUND
[0002] In actual production or engineering practice, it is often necessary to establish a performance model to predict and analyze key performance indicators, thereby providing data support for decision-making.
[0003] The existing performance model usually relies on the statistical relationship between historical parameter data and performance indicators to select parameters and establish a model, and usually uses a simple linear model or a single form of nonlinear model for parameter fitting. When optimizing the performance model parameters, the problem of insufficient model accuracy caused by abnormal data interference and single fitting method cannot be effectively solved, making it difficult to accurately predict the performance indicators using the optimized performance model parameters. SUMMARY
[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a performance model parameter simulation optimization method, an electronic device and a storage medium to solve the problems raised in the background art.
[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:
[0006] A performance model parameter simulation optimization method, comprising the following steps:
[0007] S1: Collecting historical parameters and corresponding performance indicator data of the performance model, identifying and correcting abnormal samples through local weighted regression, and generating an effective sample set after correction of abnormal values;
[0008] S2: Performing parameter correlation analysis based on the effective sample set after correction of abnormal values, calculating the contribution of performance parameters to the change of performance indicators, and generating a key performance parameter set;
[0009] S3: Based on the key performance parameter set, combining a polynomial spline function and a radial basis function to establish a nonlinear fitting model between performance parameters and performance indicators;
[0010] S4: Based on the key performance parameter set, calculating the function partial derivative between key performance parameters of the nonlinear fitting model to generate a performance parameter correlation matrix;
[0011] S5: Based on the performance parameter correlation matrix, taking the fitting residual gradient of the nonlinear fitting model as a particle velocity correction term, using an improved particle swarm optimization algorithm to optimize the performance parameters, and outputting the optimized performance model parameters;
[0012] S6: Based on the sensitivity of the fitting error and the performance parameter of the nonlinear fitting model, the effectiveness and reliability of the optimized performance model parameter are evaluated, and it is determined whether the optimized performance model parameter is reliable.
[0013] In a preferred embodiment, S1, specifically:
[0014] The performance model input parameters and the performance indicators corresponding to each group of performance model input parameters under the historical running state of the performance model are collected.
[0015] The performance model input parameters and the performance indicators are subjected to weighted residual analysis to determine whether each group of performance model input parameters and performance indicators is abnormal.
[0016] The abnormal performance model input parameters and performance indicators are subjected to local weighted regression calculation to obtain corrected performance model input parameters and performance indicators, and replace the original abnormal performance model input parameters and performance indicators, to obtain an effective sample set after correction of outliers.
[0017] In a preferred embodiment, S2, specifically:
[0018] Based on the effective sample set after correction of outliers, the Pearson correlation coefficient between each performance model input parameter and the corresponding performance indicator is calculated.
[0019] According to the absolute value of the calculated Pearson correlation coefficient, the importance of the performance model input parameters is sorted, and the performance model input parameters with an absolute value of the Pearson correlation coefficient greater than a preset threshold are selected.
[0020] The selected performance model input parameters are determined as a set of key performance parameters.
[0021] In a preferred embodiment, S3, specifically:
[0022] A polynomial spline function fitting model and a radial basis function fitting model are constructed with the set of key performance parameters as input data and the corresponding performance indicators as output data.
[0023] The order of the polynomial spline function fitting model is selected, and the best order of the polynomial spline function fitting model is determined according to the Bayesian information criterion.
[0024] The base function parameter of the radial basis function fitting model is determined, and the optimal radial basis function parameter of the radial basis function fitting model is determined according to the Bayesian information criterion.
[0025] The polynomial spline function fitting model with the optimal order determined is fused with the radial basis function fitting model with the optimal radial basis function parameter determined to construct a nonlinear fitting model between the key performance parameters and the performance indicators.
[0026] In a preferred embodiment, S4, specifically,
[0027] Based on the set of key performance parameters, a first-order partial derivative of each key performance parameter with respect to the performance indicator in the nonlinear fitting model is calculated to determine a sensitivity coefficient of each key performance parameter.
[0028] A second-order mixed partial derivative corresponding to any two different key performance parameters in the nonlinear fitting model is calculated to determine an interaction influence coefficient between the two key performance parameters.
[0029] Based on the sensitivity coefficient of each key performance parameter and the interaction influence coefficient between any two key performance parameters, a performance parameter correlation matrix is constructed.
[0030] In a preferred embodiment, S5, specifically,
[0031] Based on the performance parameter correlation matrix, a fitting residual between the performance indicator predicted by the nonlinear fitting model and the actual performance indicator in the effective sample set after the abnormal values are corrected and a gradient of the fitting residual with respect to each key performance parameter are calculated to obtain a fitting residual gradient.
[0032] The fitting residual gradient is used as a particle velocity correction term in a particle swarm optimization algorithm, and the particle swarm optimization algorithm is used to update the position and velocity of each particle; the position of the particle corresponds to a set of key performance parameters to be optimized.
[0033] The particle position and velocity of the particle swarm optimization algorithm are iterated continuously until a preset termination condition is met, and the optimized performance model parameters are obtained.
[0034] In a preferred embodiment, S6, specifically,
[0035] Based on the nonlinear fitting model, the key performance parameters in the effective sample set after the abnormal values are corrected are used to predict the performance indicators using the optimized performance model parameters, and a fitting error between the predicted performance indicators and the corresponding actual performance indicators in the effective sample set after the abnormal values are corrected is calculated.
[0036] The sensitivity coefficient of each key performance parameter is determined based on the first-order partial derivative of each key performance parameter in the performance parameter correlation matrix.
[0037] According to the fitting error and the sensitivity coefficient, the effectiveness and reliability of the optimized performance model parameters are evaluated, and it is determined whether the optimized performance model parameters are reliable.
[0038] In another aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the optimization method of performance model parameter simulation.
[0039] In another aspect, the present application provides a storage medium comprising computer executable instructions for implementing the optimization method of performance model parameter simulation when executed by a computer processor.
[0040] The technical effects and advantages of the optimization method of performance model parameter simulation, the electronic device and the storage medium of the present application are as follows:
[0041] By identifying and correcting the abnormal values in the historical parameters and the corresponding performance indicator data, the interference of noise on model fitting is eliminated, the data quality and sample representativeness are improved; by parameter correlation analysis to extract key parameters, redundant variables are reduced; by using the fusion of polynomial spline and radial basis function to construct a nonlinear fitting model, the fitting accuracy can be improved; by calculating the function partial derivative between the key performance parameters of the nonlinear fitting model, the parameter coupling effect is quantified; the fitting residual gradient is introduced into the particle swarm optimization algorithm to accelerate the convergence and enhance the optimization accuracy, so that the optimal performance parameter is quickly located; finally, based on the fitting error of the nonlinear fitting model and the sensitivity of the performance parameters, the effectiveness and reliability of the optimized performance model parameters are evaluated to ensure the effectiveness and reliability of the optimized parameters and improve the performance model parameter simulation and prediction effect. BRIEF DESCRIPTION OF DRAWINGS
[0042] Figure 1 FIG. 1 is a schematic diagram of the optimization method of performance model parameter simulation of the present application. DETAILED DESCRIPTION
[0043] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0044] Embodiment: Figure 1 The optimization method of performance model parameter simulation of the present application is given, which comprises the following steps:
[0045] S1: Collect the historical parameters of the performance model and the corresponding performance indicator data, identify and correct the abnormal samples by local weighted regression, and generate an effective sample set after correction of abnormal values;
[0046] S2: Perform parameter correlation analysis based on the effective sample set after abnormal value correction, generate a key performance parameter set by calculating the contribution of performance parameters to the change of performance indicators;
[0047] S3: Based on the key performance parameter set, establish a nonlinear fitting model between performance parameters and performance indicators by combining polynomial spline function and radial basis function;
[0048] S4: Based on the key performance parameter set, generate a performance parameter correlation matrix by calculating the function partial derivative between key performance parameters in the nonlinear fitting model;
[0049] S5: Based on the performance parameter correlation matrix, use the fitting residual gradient of the nonlinear fitting model as the particle velocity correction term, and use the improved particle swarm optimization algorithm to optimize the performance parameters, and output the optimized performance model parameters;
[0050] S6: Based on the fitting error of the nonlinear fitting model and the sensitivity of the performance parameters, evaluate the effectiveness and reliability of the optimized performance model parameters, and determine whether the optimized performance model parameters are reliable.
[0051] S1: Collect historical performance model parameters and corresponding performance indicator data, identify and correct abnormal samples by local weighted regression, and generate an effective sample set after abnormal value correction, including:
[0052] Collect performance model input parameters under historical running state and performance indicators corresponding to each group of performance model input parameters;
[0053] Specifically, the input parameters of the performance model refer to a series of input variables used for inputting the model and capable of causing changes in the output of the model in the actual production and operation or engineering application process. The input parameters of the performance model directly affect the accuracy of the model output. For example, in the performance analysis of the production system of an enterprise, the input parameters of the performance model can include the working temperature of equipment, the running time of equipment, the purity of raw materials, the operation skill level of production personnel, etc. The input parameters are usually obtained through an enterprise internal data acquisition system, such as a temperature sensor, a database of a production management system or a quality detection device, etc. under the historical running state. Each time an input parameter of the performance model is collected, the performance indicators corresponding to the input parameters of the performance model are also ensured to be collected at the same time. The performance indicators are output variables of the performance model, i.e. actual results produced under the action of specific input parameters. For example, in the production process of an enterprise, when the input parameters of the performance model, such as the temperature of equipment, the running time, the purity of raw materials and the skill level of operators, are collected, the corresponding performance indicators, such as the qualified rate of products, the unit energy consumption level, the product delivery cycle or the production cost indicators, can be recorded at the same time. The collection of the performance indicators also uses enterprise internal information systems or field sensing devices to record, ensuring that the input parameters and the output performance indicators correspond to each other one by one and ensuring that each group of data has complete correlation. Through the corresponding grouping, the historical data samples of the performance model are finally obtained. Each group of sample data is composed of the input parameters of the performance model and the corresponding performance indicators.
[0054] The weighted residual error analysis is performed on the input parameters of the performance model and the performance indicators to determine whether each group of input parameters of the performance model and performance indicators is abnormal.
[0055] Specifically, for each set of historical data samples collected, i.e., each set of performance model input parameters and corresponding performance indicators, the residual between the theoretical predicted value and the actual measured value is calculated; all collected data is taken as the initial sample to establish a preliminary fitting analysis model, specifically, the theoretical function form obtained by fitting the historical samples; the performance model input parameters in each set of historical data samples are substituted into the preliminary fitting analysis model to obtain the performance indicator predicted value predicted by the model, and then the performance indicator predicted value is compared with the actual performance indicator true value collected, and the difference between the performance indicator predicted value and the actual performance indicator true value collected is the residual; considering that different samples have different importance, therefore, a weight is introduced when calculating the residual, and the weight of each set of data is set according to the actual working condition when the data is collected, the reliability of the data collection source or the fluctuation degree of the sample data; for example, samples with greater volatility or samples with greater uncertainty in the measurement process will be given a smaller weight, and stable and reliable sample data will be given a larger weight; based on the weight setting, the residual is weighted to form a weighted residual value; for each set of historical data, by comparing the weighted residual with the pre-set threshold, it is determined whether each set of historical data sample is abnormal data; if the weighted residual of a set of historical data sample is greater than the set threshold, it means that the data sample has a large deviation, and therefore the set of historical data is judged to be abnormal data; if the weighted residual value is less than or equal to the set threshold, it is determined to be normal data.
[0056] The abnormal performance model input parameters and performance indicators are subjected to local weighted regression calculation to obtain corrected performance model input parameters and performance indicators, and the original abnormal performance model input parameters and performance indicators are replaced to obtain an effective sample set with abnormal values corrected.
[0057] Specifically, after the abnormal data samples are identified through weighted residual analysis, the abnormal data samples are corrected through local weighted regression calculation; for each set of data samples marked as abnormal, a local regression model is constructed based on the normal sample data around the abnormal sample; the weight of each adjacent sample in the local regression model is assigned according to the distance or similarity between the samples, the closer the normal sample to the abnormal data, the greater the weight value, and the farther the normal sample, the smaller the weight value. The normal sample data with assigned weights are used to predict the correction values of the performance model input parameters and performance indicators in the abnormal data samples through local regression calculation, so as to obtain the corrected performance model input parameters and performance indicators; the local regression calculation includes determining the weight through a distance function, and then performing weighted least squares regression analysis on the samples with weights to obtain the best correction value; after the local regression calculation is completed, the corrected performance model input parameters and performance indicators are substituted into the original abnormal sample data to form a corrected historical data set, i.e., an effective sample set with abnormal values corrected.
[0058] S2: Perform parameter correlation analysis based on the effective sample set after the correction of outliers, generate a set of key performance parameters by calculating the contribution of performance parameters to the change of performance indicators, including:
[0059] Based on the effective sample set after the correction of outliers, calculate the Pearson correlation coefficient between each performance model input parameter and the corresponding performance indicator;
[0060] Specifically, the Pearson correlation coefficient is a statistical indicator used to measure the degree of linear correlation between variables, reflecting the strength and direction of the linear correlation between variables; for example, in the performance analysis of an enterprise production system, the performance model input parameters include the working temperature of the equipment, the running time of the equipment, the purity of raw materials, the operation skill level of production personnel, etc., and the corresponding performance indicators may be the product qualification rate or the unit energy consumption level; in order to calculate the Pearson correlation coefficient between the working temperature of the equipment and the product qualification rate, specifically: extract the working temperature of the equipment and the corresponding product qualification rate from each group of data in the effective sample set after the correction of outliers; calculate the mean of the equipment working temperature data and the product qualification rate data; take the difference between each equipment working temperature and the average equipment working temperature as the first data difference, and take the difference between each product qualification rate and the average product qualification rate as the second data difference; then multiply each pair of data differences to obtain a series of products; add the series of products to obtain the cumulative sum; at the same time, calculate the sum of the squares of the differences between the equipment working temperature data and the corresponding average value, i.e. square the difference between each equipment working temperature and the average equipment working temperature and add them up to obtain the sum of the squares of the differences of the equipment working temperature; square the difference between each product qualification rate and the average product qualification rate and add them up to obtain the sum of the squares of the differences of the product qualification rate; take the square root of the sum of the squares of the differences of the equipment working temperature and the sum of the squares of the differences of the product qualification rate respectively, and multiply them to obtain the denominator; finally, divide the cumulative sum by the denominator to obtain the Pearson correlation coefficient between the equipment working temperature and the product qualification rate; similarly, calculate the correlation between all performance model input parameters and corresponding performance indicators, such as the correlation between equipment running time and product qualification rate, the correlation between raw material purity and product qualification rate, the correlation between production personnel operation skill level and product qualification rate, etc.
[0061] According to the absolute value of the Pearson correlation coefficient obtained by calculation, the importance of the performance model input parameters is sorted, and the performance model input parameters with an absolute value of the Pearson correlation coefficient greater than a preset threshold are selected;
[0062] Specifically, the Pearson correlation coefficient is in the range of -1 to 1, a negative number indicates that the performance model input parameter and the performance indicator have a negative correlation, that is, when the performance model input parameter increases, the performance indicator decreases, and a positive number indicates that the two have a positive correlation, that is, when the performance model input parameter increases, the performance indicator also increases accordingly; in order to comprehensively reflect the degree of influence of the performance model input parameter on the performance indicator, the absolute value of all Pearson correlation coefficients is taken to eliminate the influence of negative signs, and only the numerical value is considered; the absolute values of the Pearson correlation coefficients corresponding to each performance model input parameter are arranged in descending order according to the numerical value, so as to determine the order of the influence degree of each performance model input parameter on the performance indicator; for example, assuming that the absolute value of the Pearson correlation coefficient between the equipment working temperature and the product qualified rate is large, and the absolute value of the Pearson correlation coefficient between the equipment running time and the product qualified rate is relatively small, then the sorting priority of the equipment working temperature will be higher than that of the equipment running time in the sorting process, which means that the priority of the equipment working temperature as a performance model input parameter will be higher; in actual operation, a threshold value of the absolute value of the Pearson correlation coefficient is set according to the enterprise management or technical requirements, that is, a numerical limit is determined; all performance model input parameters with an absolute value of the Pearson correlation coefficient greater than the limit will be retained.
[0063] S2: Select the performance model input parameter as the key performance parameter set;
[0064] Specifically, through the sorting and threshold selection of the performance model input parameter, all performance model input parameters with an absolute value of the Pearson correlation coefficient greater than the set threshold value will be uniformly determined as the key performance parameter.
[0065] S3: Based on the key performance parameter set, a nonlinear fitting model between the performance parameter and the performance indicator is established by combining the polynomial spline function and the radial basis function, including:
[0066] Taking the key performance parameter set as the input data and the corresponding performance indicator as the output data, a polynomial spline function fitting model and a radial basis function fitting model are constructed;
[0067] Specifically, the polynomial spline function fitting model is a segmented function form composed of polynomial function segments, the overall model uses segmented polynomial functions to represent the relationship between input data and output data, each function segment is connected through a specific junction point to maintain the smoothness and continuity of the overall model at the data junction point; for example, in the performance analysis of an enterprise production system, when a polynomial spline function model is constructed with equipment operating temperature as input data and product pass rate as output data, the data will be divided into multiple intervals according to the equipment operating temperature range, a polynomial function segment is fitted in each interval, and the function value and derivative value at the junction point of each function segment need to be continuous and smooth; the overall polynomial spline function model is obtained through the above method; the radial basis function fitting model is a nonlinear fitting method based on radial basis functions, which uses the spatial distance relationship based on data points to construct mathematical functions, uses radial distance as a variable, and establishes a nonlinear relationship between input data and output data; the construction process of the radial basis function model is based on the spatial distribution relationship between the value of each key performance parameter and the corresponding performance indicator, the spatial distance between each group of key performance parameter values and other sample data is calculated, the spatial distance is used as an input variable to calculate the radial basis function, thereby realizing the nonlinear fitting relationship between the input parameters and the performance indicators; for example, taking the purity of raw materials and the skill level of production personnel as input variables, the radial distance between each group of input data and other sample data is calculated, and a radial basis function fitting model is constructed with radial distance as the independent variable to obtain the relationship model between input and output.
[0068] The order of the polynomial spline function fitting model is selected, and the best order of the polynomial spline function fitting model is determined according to the Bayesian information criterion;
[0069] Specifically, the accuracy and adaptability of the polynomial spline function fitting model are directly affected by the order of the function, which represents the highest power of the polynomial function. If the order is too low, the relationship between the input data and the output data may not be fully expressed. If the order is too high, the fitting may be too complex, causing data overfitting. Therefore, the order of the polynomial spline function fitting model needs to be optimized. The order is selected using the Bayesian information criterion. The Bayesian information criterion is a statistical standard for evaluating the fitting effect of the model. By considering the accuracy of the model fitting and the complexity of the model itself, the best model order that takes into account the fitting accuracy and the simplicity of the model is selected. The Bayesian information criterion is calculated as follows: a plurality of polynomial spline function fitting models are constructed according to different orders, and the residual sum of squares and the number of model parameters are calculated for each model. The residual sum of squares is calculated by squaring the difference between the predicted performance indicator and the actual performance indicator of each data point and then summing. The number of model parameters is determined by the order of the model. The higher the order, the more the number of model parameters increases. The residual sum of squares and the number of model parameters are jointly substituted into the Bayesian information criterion calculation formula for calculation. The calculation formula is calculated by combining the natural logarithm of the residual sum of squares and the number of model parameters with the natural logarithm of the number of sample data. The Bayesian information criterion values of the models corresponding to each order are calculated. The smaller the value, the better the balance between the fitting effect and the complexity of the model. Therefore, the order corresponding to the smallest Bayesian information criterion value is selected as the best order of the polynomial spline function fitting model.
[0070] The basis function parameters of the radial basis function fitting model are determined, and the optimal radial basis function parameter of the radial basis function fitting model is determined according to the Bayesian information criterion.
[0071] Specifically, the accuracy of the radial basis function fitting model depends on the selection of the basis function parameters, for example, the width parameter of the radial basis function determines the spatial distribution range of the function, and different parameter settings will affect the fitting effect of the model; Therefore, it is necessary to determine the optimal basis function parameters of the radial basis function fitting model; The way to determine the basis function parameters is the same as the way to select the order of the polynomial spline function model, and the Bayesian information criterion is used for parameter selection, specifically: according to different radial basis function parameters, a plurality of radial basis function fitting models are constructed, and the residual sum of squares between the predicted performance indicators and the actual performance indicators and the number of parameters of each model are calculated respectively; The calculation process is similar to that of the polynomial spline function model; The residual sum of squares, the number of model parameters and the number of sample data are jointly substituted into the Bayesian information criterion calculation formula, and the Bayesian information criterion value corresponding to each radial basis function parameter is calculated respectively; The lower the Bayesian information criterion value, the better the balance between the fitting effect and the complexity of the corresponding model, so the basis function parameter corresponding to the model with the lowest Bayesian information criterion value is selected as the optimal parameter of the radial basis function model.
[0072] The polynomial spline function fitting model with the best order and the radial basis function fitting model with the optimal radial basis function parameter are fused to construct a nonlinear fitting model between the key performance parameters and the performance indicators.
[0073] Specifically, the polynomial spline function model with the best order and the radial basis function model with the optimal parameter are fused, that is, the predicted results of the performance indicators of the polynomial spline function model with the best order and the radial basis function model with the optimal parameter are respectively combined and calculated through weighted combination; The fusion method includes assigning different weight coefficients to the predicted values of the two models, and the determination of the weight coefficients can be determined according to the fitting effects of the two models, for example, a larger weight is assigned to the model with lower prediction error, and a smaller weight is assigned to the model with larger error; Through the weighted calculation of the weight coefficients, the prediction results of the two models are fused, so as to obtain the overall nonlinear fitting model between the key performance parameters and the performance indicators, and realize more accurate and stable prediction effect on the performance indicators.
[0074] S4: Based on the set of key performance parameters, the function partial derivative between the key performance parameters in the nonlinear fitting model is calculated to generate a performance parameter correlation matrix, including:
[0075] Based on the set of key performance parameters, the first-order partial derivative of each key performance parameter with respect to the performance indicators in the nonlinear fitting model is calculated to determine the sensitivity coefficient of each key performance parameter;
[0076] Specifically, in order to determine the influence degree of each key performance parameter on the performance indicator, the first-order partial derivative of each key performance parameter on the performance indicator in the nonlinear fitting model needs to be calculated; the actual meaning of the first-order partial derivative is: under the condition of keeping the values of other key performance parameters unchanged, the sensitivity degree of the performance indicator change caused by the change of a certain key performance parameter; for example, taking the equipment working temperature as the key performance parameter and the product qualified rate as the performance indicator, the first-order partial derivative of the equipment working temperature represents the change amount of the product qualified rate caused by the unit tiny change of the equipment working temperature under the condition that other parameters such as the purity of raw materials and the skill level of operators remain unchanged, and the calculation method is: determining the value interval of each key performance parameter in the nonlinear fitting model, and gradually taking points in the value interval; at each value point, the value of the key performance parameter is changed slightly, that is, it is slightly increased or decreased by a very small amount; the changed key performance parameter is substituted into the nonlinear fitting model for calculation to obtain the change of the corresponding performance indicator prediction value; the ratio between the change amount of the performance indicator prediction value and the tiny change amount of the key performance parameter is the first-order partial derivative; the above calculation is performed on each key performance parameter in the entire value interval to obtain the first-order partial derivative of each key performance parameter at each value point; then the absolute values of the first-order partial derivatives of each key performance parameter calculated in the value interval are taken and the average value is calculated to obtain the average influence degree of each key performance parameter on the performance indicator, and then the sensitivity coefficient of each key performance parameter is determined; the greater the sensitivity coefficient, the more significant the influence degree of the corresponding key performance parameter on the performance indicator.
[0077] The second-order mixed partial derivative corresponding to any two different key performance parameters in the nonlinear fitting model is calculated to determine the interaction influence coefficient between any two key performance parameters;
[0078] Specifically, after determining the individual influence of each key performance parameter on the performance indicator, it is also necessary to determine the influence of the interaction between different key performance parameters on the performance indicator, i.e. the interaction effect; for example, there may be an interaction between the equipment operating temperature and the purity of raw materials, i.e. when the equipment operating temperature and the purity of raw materials change at the same time, the change in product pass rate is not simply the superposition of the individual effects of the equipment operating temperature and the purity of raw materials, but may also have a different trend due to the combined effect; therefore, it is necessary to calculate the second-order mixed partial derivative of any two different key performance parameters; the second-order mixed partial derivative represents the proportional relationship between the change amplitude of the performance indicator and the product of the small changes in the two parameters; the calculation of the second-order mixed partial derivative is as follows: within the value range of the key performance parameters, a plurality of data points are selected, each data point including the values of two key performance parameters; at each data point, the two key performance parameters are simultaneously changed slightly, and the new values of the two key performance parameters after the slight changes are obtained; the new values of the two key performance parameters are substituted into the nonlinear fitting model to calculate the predicted value change of the performance indicator; the predicted value change of the performance indicator is divided by the product of the small changes in the two key performance parameters to obtain the second-order mixed partial derivative; the above calculation is performed for each pair of key performance parameters, and the second-order mixed partial derivative is calculated at a plurality of data points in the value range; the absolute values of the second-order mixed partial derivatives calculated for each pair of key performance parameters are averaged to obtain the interaction coefficient between any two key performance parameters; the larger the interaction coefficient, the more significant the interaction between the two key performance parameters on the performance indicator.
[0079] Based on the sensitivity coefficients of each key performance parameter and the interaction coefficients between any two key performance parameters, a performance parameter correlation matrix is constructed.
[0080] Specifically, after determining the sensitivity coefficients of each key performance parameter and the interaction coefficients between any two key performance parameters, a performance parameter correlation matrix is constructed to represent the importance of the key performance parameters on the performance indicator and the correlation between the parameters; the rows and columns of the performance parameter correlation matrix represent the key performance parameters; the main diagonal positions of the performance parameter correlation matrix correspond to the sensitivity coefficients of each key performance parameter, representing the importance of each key performance parameter when it changes alone; the non-diagonal positions of the matrix represent the interaction coefficients between any two key performance parameters, representing the combined influence of any two key performance parameters on the performance indicator when they change simultaneously.
[0081] S5: Based on the performance parameter correlation matrix, the fitting residual gradient of the nonlinear fitting model is used as a particle velocity correction term, and an improved particle swarm optimization algorithm is used to optimize the performance parameters, and the optimized performance model parameters are output, including:
[0082] Based on the performance parameter correlation matrix, the fitting residual between the performance index predicted by the nonlinear fitting model and the actual performance index in the effective sample set after the abnormal value correction and the gradient of the fitting residual relative to each key performance parameter are calculated to obtain the fitting residual gradient.
[0083] Specifically, to optimize the performance model parameters, the fitting residual between the performance index predicted by the nonlinear fitting model and the actual performance index in the effective sample set after the abnormal value correction is calculated based on the performance parameter correlation matrix, and the gradient of the fitting residual relative to each key performance parameter is calculated. The nonlinear fitting model is a model that fuses a polynomial spline function and a radial basis function, and the predicted performance index is the result calculated by the nonlinear fitting model under the condition of a given set of key performance parameter values. The actual performance index is the real measured value recorded in the effective sample set after the abnormal value correction. When calculating the fitting residual, the difference between the performance index predicted by the nonlinear fitting model and the actual performance index at the corresponding position in the effective sample set after the abnormal value correction is determined as the fitting residual. The size of the fitting residual reflects the prediction accuracy of the nonlinear fitting model. For example, when the key performance parameters are the working temperature of the equipment, the purity of raw materials, and the operation skill level of production personnel, and the performance index is the product pass rate, the key performance parameters are substituted into the nonlinear fitting model for prediction to obtain the predicted product pass rate. Then, the actual product pass rate recorded in a data sample with the same or closest values of the key performance parameters is found from the effective sample set after the abnormal value correction. The numerical difference between the predicted product pass rate and the actual product pass rate is calculated, and the numerical difference is determined as the fitting residual of the data sample. To determine the relationship between the fitting residual and the key performance parameters, the gradient of the fitting residual relative to each key performance parameter also needs to be calculated. The gradient represents the trend and change speed of the fitting residual with the key performance parameters. The gradient calculation method is as follows: a very small change is made to each key performance parameter individually while keeping the other key performance parameters unchanged, the nonlinear fitting model is substituted again, and the corresponding performance index prediction value change is calculated. The newly calculated performance index prediction value is compared with the actual performance index to determine the new fitting residual. The difference between the new fitting residual and the original fitting residual is calculated and divided by the small change amount of the key performance parameter to obtain the gradient of the fitting residual relative to the key performance parameter. The fitting residual gradient represents the sensitive direction and change trend of the key performance parameter value in the nonlinear model fitting.
[0084] The fitting residual gradient is used as a particle speed correction term in the particle swarm optimization algorithm, and the particle swarm optimization algorithm is used to update the position and speed of each particle. The position of the particle corresponds to a set of key performance parameters to be optimized.
[0085] Specifically, the particle swarm optimization algorithm searches for a position with better performance by simulating the cooperation and information sharing behavior of each particle in the swarm; the position of a particle represents a combination of values of a set of key performance parameters to be optimized, such as the working temperature of equipment, the purity of raw materials, and the operation skill level of production personnel; the speed of a particle represents the moving trend and moving step of the particle in the parameter space; in order to improve the optimization accuracy and efficiency of the particle swarm optimization algorithm, the calculated fitting residual gradient is used as a correction term for the speed of the particle, specifically: initialize the particle swarm, and randomly assign an initial position and speed to each particle in the parameter space; for each particle, calculate the predicted value of the performance indicator corresponding to the position of the particle according to the nonlinear fitting model, and then compare it with the actual performance indicator of the effective sample set after the abnormal value is corrected, calculate the fitting residual and calculate the fitting residual gradient; when calculating the new speed of the particle, the fitting residual gradient is added as an additional correction term to the particle speed update formula, specifically: the original speed update formula is a weighted combination of the differences between the current position of the particle, the historical individual optimal position, and the group optimal position; after introducing the fitting residual gradient, the speed update formula additionally adds a term combining the fitting residual gradient and the speed of the particle to adjust the moving direction of the particle; the particle is guided to move in the direction in which the fitting residual decreases rapidly, thereby effectively improving the optimization efficiency of the particle swarm optimization algorithm in the parameter space; the updated speed is then used to update the position of the particle, that is, the new position of the particle is determined by adding the current position and the updated speed; the new position represents a new combination of key performance parameters to be optimized, and the above process is repeated until a position with the best performance or close to the best performance is searched.
[0086] By continuously iterating the particle position and speed of the particle swarm optimization algorithm until the preset termination condition is met, the optimized performance model parameters are obtained.
[0087] Specifically, the particle swarm optimization algorithm is an iterative process, and the position and velocity of the particle are updated in each iteration to obtain a better solution; the particle position corresponds to the key performance parameter combination to be optimized, and the particle velocity determines the direction and distance of the next movement of the particle position; the termination condition is set as: setting the maximum number of iterations or the minimum threshold of fitting residual; in the iteration process of the particle swarm optimization algorithm, first, it is judged whether the current number of iterations reaches the maximum number of iterations, if the maximum number of iterations is reached, the particle swarm optimization algorithm stops; if the maximum number of iterations is not reached, it is judged whether the fitting residual corresponding to the optimal position found by the current particle swarm is less than the preset minimum fitting residual threshold; if the fitting residual is less than the preset minimum fitting residual threshold, the particle swarm optimization algorithm stops; if the fitting residual is greater than or equal to the preset minimum fitting residual threshold, continue iteration; through the judgment, the particle swarm optimization algorithm continuously updates the particle position and velocity, and the particle position is continuously optimized, so as to obtain a key performance parameter combination with higher fitting accuracy; when the particle swarm optimization algorithm meets the termination condition and stops, the current global optimal particle position represents the key performance parameter combination with the optimal performance, that is, the optimized performance model parameter; through the combination parameter, the fitting residual between the performance index prediction and the actual performance index is minimized in the nonlinear fitting model, ensuring the effectiveness and reliability of the model parameter optimization.
[0088] S6: Based on the fitting error of the nonlinear fitting model and the sensitivity of the performance parameter, the effectiveness and reliability of the optimized performance model parameter are evaluated, and whether the optimized performance model parameter is reliable is judged, including:
[0089] Based on the nonlinear fitting model, the key performance parameters in the effective sample set after the abnormal value correction are used to predict the performance index by using the optimized performance model parameter, and the fitting error between the predicted performance index and the corresponding actual performance index in the effective sample set after the abnormal value correction is calculated;
[0090] Specifically, in order to verify the reliability and effectiveness of the optimized performance model parameter combination, the optimized performance model parameters are re-substituted into the nonlinear fitting model, and then the key performance parameters recorded in the effective sample set after the correction of outliers are used to predict the performance indicators, so as to ensure the prediction accuracy of the nonlinear fitting model under the condition of the optimized parameters: each group of key performance parameter combinations is extracted from the effective sample set after the correction of outliers, and the optimized performance model parameters are used to obtain the predicted performance indicators corresponding to each group of key performance parameter combinations by combining the nonlinear fitting model; for example, taking the product pass rate as the performance indicator, obtaining the equipment working temperature, raw material purity and production personnel operation skill level in the production process as the key performance parameters, combining the optimized equipment working temperature, raw material purity and production personnel operation skill level with the key performance parameters, and then bringing them into the nonlinear fitting model to calculate the corresponding predicted product pass rate; after the performance indicator prediction is completed, the fitting error between the predicted performance indicator and the corresponding actual performance indicator in the effective sample set also needs to be predicted; the fitting error is the difference between the predicted performance indicator and the actual performance indicator, which reflects the applicability and effectiveness of the optimized performance model parameter combination in the real production environment or engineering environment; all key performance parameter combinations in the effective sample set after the correction of outliers are predicted and the fitting error is calculated, so as to obtain the fitting error of all data; in order to avoid the influence of error signs, the fitting error absolute value is usually used for description; the average value of the fitting error absolute value of all data is calculated to evaluate the prediction accuracy of the optimized performance model parameters as a whole; the smaller the average value of the fitting error absolute value, the better the optimization effect of the model parameters, that is, the higher the accuracy of the nonlinear fitting model in predicting the performance indicators under the condition of the optimized key performance parameters, that is, it is considered that the optimized performance model parameters have good effectiveness.
[0091] The sensitivity coefficient of each key performance parameter is determined based on the first-order partial derivative of each key performance parameter in the performance parameter correlation matrix;
[0092] Specifically, the first-order partial derivative is obtained by changing a certain key performance parameter in the nonlinear fitting model alone while keeping other key performance parameters unchanged, and calculating the rate of change of the performance index with the slight change of the key performance parameter; for example, taking the product pass rate as the performance index and the device working temperature as the key performance parameter, when the device working temperature changes in a very small range, the ratio of the change of the product pass rate to the change of the temperature is the first-order partial derivative of the device working temperature, which is the change amount of the device working temperature corresponding to each unit of the slight change of the product pass rate; in order to determine the sensitivity coefficient, the first-order partial derivative needs to be calculated at multiple data points in the value range of each key performance parameter; the absolute values of the first-order partial derivatives of each key performance parameter in the respective value ranges are averaged to obtain the average sensitivity coefficient of each key performance parameter to the performance index; the higher the sensitivity coefficient, the stronger the influence of the key performance parameter on the performance index.
[0093] According to the fitting error and the sensitivity coefficient, the effectiveness and reliability of the optimized performance model parameters are evaluated, and whether the optimized performance model parameters are reliable is judged.
[0094] Specifically, the smaller the average value of the absolute value of the fitting error, the higher the prediction accuracy of the optimized performance model parameters in the actual application scenario; the higher the sensitivity coefficient, the more important the influence of the key performance parameter on the performance index.
[0095] Set the comprehensive evaluation index, and assign corresponding weights to the average value of the absolute value of the fitting error and the sensitivity coefficient of each key performance parameter, respectively; the smaller the average value of the absolute value of the fitting error, the higher the weight, and the greater the sensitivity coefficient, the higher the weight; multiply the average value of the absolute value of the fitting error of each key performance parameter by the corresponding fitting error weight to obtain the fitting error weighted value; then multiply each sensitivity coefficient by the corresponding sensitivity weight to obtain the sensitivity weighted value; sum the fitting error weighted values and the sensitivity weighted values of all key performance parameters to obtain the comprehensive evaluation index; the lower the comprehensive evaluation index, the higher the effectiveness and reliability of the optimized performance model parameters; if the comprehensive evaluation index exceeds the set standard, it indicates that the reliability of the optimized performance model parameters is insufficient, and optimization needs to be performed again.
[0096] The application discloses a kind of performance model parameter simulation optimization equipment, the equipment includes: memory, processor and the optimization program of a kind of performance model parameter simulation stored on the memory and can be run on the processor, the optimization program of a kind of performance model parameter simulation is configured to realize the steps of a kind of performance model parameter simulation optimization method as described above.
[0097] The application discloses a storage medium, characterized in that an optimization program based on a performance model parameter simulation is stored on the storage medium, and the optimization program based on the performance model parameter simulation, when executed by a processor, implements the steps of the optimization method of the performance model parameter simulation.
[0098] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the flow or function described in the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center through a wired (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.
[0099] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.
[0100] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device, and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.
[0101] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the division of the above-described device embodiment is merely a logical function division, and there can be another division manner for the actual implementation, for example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different modules can be indirect couplings or communication connections through some interfaces, devices or modules, and can be electrical, mechanical or other forms.
[0102] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules, and can be located in one place, or can be distributed on multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.
[0103] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can be physically present alone, or two or more modules can be integrated into one module.
[0104] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program codes that can be stored in the medium.
[0105] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, and all should be covered within 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.
[0106] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.
Claims
1. An optimization method for simulating performance model parameters, characterized in that, Includes the following steps: S1: Collect historical parameters and corresponding performance indicator data of the performance model, identify and correct outlier samples through local weighted regression, and generate a valid sample set after outlier correction; S2: Perform parameter correlation analysis based on the effective sample set after outlier correction, and generate a set of key performance parameters by calculating the contribution of performance parameters to changes in performance indicators; S3: Based on the set of key performance parameters, a nonlinear fitting model between performance parameters and performance indicators is established by combining polynomial spline functions and radial basis functions. Specifically: Using the set of key performance parameters as input data and the corresponding performance indicators as output data, we construct a polynomial spline function fitting model and a radial basis function fitting model. The order of the polynomial spline function fitting model is selected, and the optimal order of the polynomial spline function fitting model is determined according to the Bayesian information criterion. The basis function parameters of the radial basis function fitting model are determined, and the optimal radial basis function parameters of the radial basis function fitting model are determined according to the Bayesian information criterion. By fusing the polynomial spline function fitting model with the optimal order and the radial basis function fitting model with the optimal radial basis function parameters, a nonlinear fitting model between key performance parameters and performance indicators is constructed. S4: Based on the set of key performance parameters, a correlation matrix of performance parameters is generated by calculating the partial derivatives of the functions among the key performance parameters in the nonlinear fitting model. Specifically: Based on the set of key performance parameters, the first-order partial derivative of each key performance parameter with respect to the performance index is calculated in the nonlinear fitting model, and the sensitivity coefficient of each key performance parameter is determined. Calculate the second-order mixed partial derivatives for any two different key performance parameters in the nonlinear fitting model, and determine the interaction coefficient between any two key performance parameters. Based on the sensitivity coefficient of each key performance parameter and the interaction coefficient between any two key performance parameters, a correlation matrix of performance parameters is constructed. S5: Based on the correlation matrix of performance parameters, using the gradient of the fitting residual of the nonlinear fitting model as the particle velocity correction term, an improved particle swarm optimization algorithm is used to optimize the performance parameters, and the optimized performance model parameters are output, specifically: Based on the performance parameter correlation matrix, the fitting residuals between the performance indicators predicted by the nonlinear fitting model and the actual performance indicators in the effective sample set after outlier correction, as well as the gradient of the fitting residuals with respect to each key performance parameter, are calculated to obtain the fitting residual gradient. The fitted residual gradient is used as the particle velocity correction term in the particle swarm optimization algorithm, and the position and velocity of each particle are updated using the particle swarm optimization algorithm; the position of the particle corresponds to a set of key performance parameters to be optimized. By iteratively adjusting the particle position and velocity of the particle swarm optimization algorithm until the preset termination condition is met, the optimized performance model parameters are obtained. S6: Based on the fitting error of the nonlinear fitting model and the sensitivity of the performance parameters, evaluate the effectiveness and reliability of the optimized performance model parameters, and determine whether the optimized performance model parameters are reliable.
2. The optimization method for performance model parameter simulation according to claim 1, characterized in that, S1, specifically: Collect the performance model input parameters and the performance indicators corresponding to each set of performance model input parameters under historical operating conditions; Weighted residual analysis was performed on the input parameters and performance indicators of the performance model to determine whether there were any anomalies in each set of input parameters and performance indicators of the performance model. Local weighted regression calculations are performed on the abnormal performance model input parameters and performance indicators to obtain the corrected performance model input parameters and performance indicators. The original abnormal performance model input parameters and performance indicators are then replaced to obtain the effective sample set after outlier correction.
3. The optimization method for performance model parameter simulation according to claim 2, characterized in that, S2, specifically: Based on the effective sample set after outlier correction, the Pearson correlation coefficient between each performance model input parameter and the corresponding performance indicator is calculated. The importance of the performance model input parameters is ranked according to the absolute value of the calculated Pearson correlation coefficient, and the performance model input parameters with an absolute value of Pearson correlation coefficient greater than a preset threshold are selected. The selected performance model input parameters are defined as the set of key performance parameters.
4. The optimization method for performance model parameter simulation according to claim 3, characterized in that, S6, specifically: Based on a nonlinear fitting model, the optimized performance model parameters are used to predict the performance indicators of key performance parameters in the effective sample set after outlier correction, and the fitting error between the predicted performance indicators and the corresponding actual performance indicators in the effective sample set after outlier correction is calculated. The sensitivity coefficient of each key performance parameter is determined based on the first-order partial derivative of each key performance parameter in the performance parameter correlation matrix; Based on the fitting error and sensitivity coefficient, the effectiveness and reliability of the optimized performance model parameters are evaluated, and it is determined whether the optimized performance model parameters are reliable.
5. An electronic device, characterized in that, The electronic device includes a processor and a memory; the processor executes an optimization method for simulating performance model parameters as described in any one of claims 1 to 4 by calling programs or instructions stored in the memory.
6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program or instructions that cause a computer to perform an optimization method for simulating performance model parameters as described in any one of claims 1 to 4.
Citation Information
Patent Citations
Enterprise digital performance quantitative management platform based on machine learning
CN120106599A