Parameter optimization method and device and electronic equipment

Abstract

A parameter optimization method and apparatus, and an electronic device, the method comprising: obtaining a first set of values of a first performance index, the first set of values of the first performance index being generated when a first service runs within a first time period based on a first value of a parameter to be adjusted, the first performance index being associated with the performance of the first service, the first service runs on the first equipment; according to the first group of numerical values of the first performance index and the second group of numerical values of the first performance index, determining that the numerical value change of the first performance index is caused by the change of the to-be-adjusted parameter from the second numerical value to the first numerical value, the second group of numerical values of the first performance index are generated when the first service runs in a second time period based on the second numerical value of the to-be-adjusted parameter; and optimizing the to-be-adjusted parameter according to the first numerical value. The scheme can be applied to the end-to-end performance optimization of the automatic driving service of the intelligent vehicle, and the parameter optimization efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving, and more specifically, to a parameter optimization method, device and electronic device. Background Art

[0002] The end-to-end performance of autonomous driving services refers to the network performance from the time the sensor sends collected data to the time the controller sends control commands.

[0003] The factors that affect the end-to-end performance of autonomous driving services mainly include hardware computing power, memory bandwidth, operating system (OS) scheduling capabilities, service triggering methods, etc. Among them, memory bandwidth and OS scheduling capabilities can be configured by adjusting parameters related to service performance, thereby optimizing the end-to-end performance of autonomous driving services. Since the mutual influence between various services in the system is relatively complex and the adjustment effects of different parameters are coupled with each other, adjusting related parameters through parameter optimization algorithms has become the current mainstream parameter optimization solution.

[0004] When using a parameter optimization algorithm for parameter optimization, a parameter is usually selected as a reference for optimization (hereinafter referred to as the parameter as an indicator), and based on the indicator, it is determined whether to use the current value of the parameter to be adjusted to continue adjusting the parameter to be adjusted. Since there are many factors that affect the end-to-end performance of the autonomous driving business, the indicator itself is unstable and has large jitters, so that the indicator optimization effect after the adjustment of the parameter to be adjusted may be submerged in the noise, and may even cause a deviation in the optimization direction, causing the performance of the parameter optimization algorithm to degenerate to an effect close to that of random search, and the optimization effect of the parameter optimization algorithm cannot be exerted.

[0005] In view of this, an efficient and reliable parameter optimization scheme needs to be developed urgently. Summary of the invention

[0006] The present application provides a parameter optimization method, device and electronic device, which can screen the numerical values ​​of the parameters to be adjusted corresponding to the indicator changes caused by non-indicator jitter as the input of the next iteration of the parameter optimization algorithm, which helps to improve the convergence of the parameter optimization algorithm, and then improve the efficiency of parameter optimization and the reliability of the optimization results.

[0007] In a first aspect, a parameter optimization method is provided, the method comprising: obtaining a first set of values ​​of a first performance indicator, the first set of values ​​of the first performance indicator being generated when a first service is running within a first time period based on a first value of a parameter to be adjusted, the first performance indicator being associated with performance of the first service, and the first service running on a first device; determining, based on the first set of values ​​of the first performance indicator and a second set of values ​​of the first performance indicator, that a value change of the first performance indicator is caused by a change of the parameter to be adjusted from a second value to a first value, the second set of values ​​of the first performance indicator being generated when the first service is running within a second time period based on the second value of the parameter to be adjusted; and optimizing the parameter to be adjusted according to the first value.

[0008] It should be noted that optimizing the parameter to be adjusted can be understood as: determining a certain value or a certain value range of the parameter to be adjusted to achieve the purpose of optimizing the first performance indicator.

[0009] Exemplarily, the first set of values ​​and the second set of values ​​of the first performance indicator may respectively include multiple data, and the number of data in the first set of values ​​of the first performance indicator may be the same as or different from the number of data in the second set of values ​​of the first performance indicator.

[0010] Exemplarily, the first time period and the second time period may be set by the user or by the system by default, and the lengths of the first time period and the second time period may be the same, or the lengths of the first time period and the second time period may be different. For example, the first time period may be 3 seconds, or 5 seconds, or other values; the second time period may be 3 seconds, or 5 seconds, or other values.

[0011] For example, the second value of the parameter to be adjusted may be the value before the parameter adjustment is performed, and the first value of the parameter to be adjusted may be the value of the parameter to be adjusted after adjustment; or, the second value of the parameter to be adjusted may be the legal value output by the last iteration of the parameter optimization algorithm, and the first value of the parameter to be adjusted may be the value output by the current iteration of the parameter optimization algorithm. A legal value may be understood as a change in the first performance indicator caused by the value that is not caused by indicator jitter.

[0012] Exemplarily, the numerical change of the first performance indicator may include: the mean value change of the first performance indicator before and after the numerical adjustment of the parameter to be adjusted.

[0013] Exemplarily, the first device may be an intelligent driving device such as a vehicle, or the first device may be a computer loaded with a virtual autonomous driving system or an autonomous driving system simulator.

[0014] In the above technical scheme, by comparing two groups of values ​​of the first performance indicator corresponding to before and after the adjustment of the parameter to be adjusted, it can be determined whether the value change of the first performance indicator after the adjustment of the parameter to be adjusted is caused by indicator jitter. When the value change of the first performance indicator is not caused by indicator jitter but is the result of the adjustment of the parameter to be adjusted, the parameter to be adjusted is further optimized based on the value of the adjusted parameter to be adjusted, which helps to improve the convergence ability (or convergence speed) of the parameter optimization algorithm, and thus improve the efficiency of parameter optimization and the reliability of the optimization results.

[0015] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: obtaining a first set of values ​​of a second performance indicator within a first time period, the second performance indicator being associated with the performance of a second service, and the second service running on the first device; determining a value change of the second performance indicator when the parameter to be adjusted changes from the second value to the first value based on the second set of values ​​of the second performance indicator and the first set of values ​​of the second performance indicator, the second set of values ​​of the second performance indicator being generated when the second service runs within the second time period based on the second value of the parameter to be adjusted; optimizing the parameter to be adjusted based on the first value, including: optimizing the parameter to be adjusted based on the first value when the value change of the second performance indicator does not exceed the deterioration limit threshold.

[0016] Exemplarily, the second service and the first service may be the same service or different services. When the two are different services, the second service and the first service both run on the first device and share resources of the first device, such as memory bandwidth, OS scheduling capability and other resources.

[0017] Exemplarily, the first performance indicator is the utilization rate of a central processing unit (CPU) associated with the first business, and the second performance indicator is the utilization rate of a CPU associated with the second business; or, the first performance indicator is the utilization rate of a CPU associated with the first business, and the second performance indicator is the frame rate of a graphics processing unit (GPU) associated with the first business.

[0018] Exemplarily, the numerical change of the second performance indicator may include: the mean change of the second performance indicator before and after the numerical adjustment of the parameter to be adjusted. Specifically, the numerical change of the second performance indicator may be the difference between the first group numerical mean of the second performance indicator and the second group numerical mean of the second performance indicator.

[0019] Exemplarily, the degradation limit threshold may be determined according to an average value of the second group of values ​​of the second performance indicator, or the degradation limit threshold may also be other values.

[0020] In one example, the larger the value of the second performance indicator is, the worse the performance of the second service is, and the deterioration limit threshold may be 10% or 5% of the second group of numerical average values ​​of the second performance indicator. Furthermore, the numerical change of the second performance indicator does not exceed the deterioration limit threshold, which may include: the difference between the first group of numerical average values ​​of the second performance indicator and the second group of numerical average values ​​of the second performance indicator is less than or equal to the deterioration limit threshold. For example, if the second performance indicator is CPU usage, and the second group of numerical average values ​​of the second performance indicator is 50%, the deterioration limit threshold may be 5%; if the first group of numerical average values ​​of the second performance indicator is 55%, the numerical change of the second performance indicator does not exceed the deterioration limit threshold; if the first group of numerical average values ​​of the second performance indicator is 56%, the numerical change of the second performance indicator exceeds the deterioration limit threshold.

[0021] In another example, the smaller the value of the second performance indicator, the worse the performance of the second service, and the deterioration limit threshold may be -10% or -5% of the average value of the second group of values ​​of the second performance indicator. Furthermore, the change in the value of the second performance indicator does not exceed the deterioration limit threshold, which may include: the difference between the average value of the first group of values ​​of the second performance indicator and the average value of the second group of values ​​of the second performance indicator is greater than or equal to the deterioration limit threshold. For example, the second performance indicator is the GPU frame rate, and the average value of the second group of values ​​of the second performance indicator is 50 frames per second (fps), then the deterioration limit threshold may be -5fps; if the average value of the first group of values ​​of the second performance indicator is 46fps, then the change in the value of the second performance indicator does not exceed the deterioration limit threshold, and if the average value of the first group of values ​​of the second performance indicator is 44fps, then the change in the value of the second performance indicator does not exceed the deterioration limit threshold.

[0022] In the specific implementation process, after the value of the parameter to be adjusted changes, the second performance indicator may be optimized or degraded. For example, when the performance indicator is the CPU usage rate, if the value change of the parameter to be adjusted causes the value of the performance indicator to increase, that is, the CPU usage rate increases, then the performance indicator is considered to be degraded; if the value change of the parameter to be adjusted causes the value of the performance indicator to decrease, that is, the CPU usage rate decreases, then the performance indicator is considered to be optimized. For another example, when the performance indicator is the GPU frame rate, if the value change of the parameter to be adjusted causes the value of the performance indicator to increase, that is, the GPU frame rate increases, then the performance indicator is considered to be optimized; if the value change of the parameter to be adjusted causes the value of the performance indicator to decrease, that is, the GPU frame rate decreases, then the performance indicator is considered to be degraded. In the present application, the optimization or degradation of a performance indicator may indicate that the performance of the business associated with the performance indicator has been optimized or degraded.

[0023] That is to say, the numerical change of the second performance indicator can evaluate the optimization degree or degradation degree of the second performance indicator, that is, the optimization degree or degradation degree of the second performance indicator is evaluated according to the numerical change amplitude of the second performance indicator; similarly, the numerical change of the first performance indicator can evaluate the optimization degree or degradation degree of the first performance indicator.

[0024] In the above technical solution, when the numerical change of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value, and the numerical change of the second performance indicator does not exceed the deterioration limit threshold, the parameter to be adjusted is optimized according to the first value. When the parameter to be adjusted affects multiple performance indicators, it helps to ensure that the degree of degradation of other performance indicators is small when optimizing one performance indicator, and even achieve simultaneous optimization of multiple performance indicators.

[0025] In combination with the first aspect, in certain implementations of the first aspect, determining whether the numerical change of the first performance indicator is caused by the parameter to be adjusted changing from the second numerical value to the first numerical value includes: performing a hypothesis test on the average value of a first group of numerical values ​​of the first performance indicator and the average value of the second group of numerical values ​​of the first performance indicator according to the first hypothesis; when the hypothesis test passes, determining that the numerical change of the first performance indicator is caused by the parameter to be adjusted changing from the second numerical value to the first numerical value.

[0026] Exemplarily, a hypothesis test can be performed on the average value of the first group of values ​​of the first performance indicator and the average value of the second group of values ​​of the first performance indicator through a T test or a Z test. The first hypothesis can be the null hypothesis in the T test or the Z test. When the null hypothesis is established, it is determined that the hypothesis test passes.

[0027] Generally speaking, indicator jitter conforms to the normal distribution. In the above technical solution, whether the numerical change of the first performance indicator is caused by indicator jitter can be determined by hypothesis testing. The solution is simple and easy to deploy.

[0028] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: determining whether the variance change of the first performance indicator is within a preset range based on a first set of values ​​of the first performance indicator and a second set of values ​​of the first performance indicator; optimizing the parameter to be adjusted according to the first value, including: when the variance change of the first performance indicator is within the preset range, optimizing the parameter to be adjusted according to the first value.

[0029] Exemplarily, the jitter of the first performance indicator may increase due to the change in the value of the parameter to be adjusted. In this case, limiting the variance change of the first performance indicator to a preset range helps to reduce the jitter of the first performance indicator.

[0030] In some possible implementations, when the variance change of the first performance indicator is within a preset range and the value change of the second performance indicator does not exceed the deterioration limit threshold, the parameter to be adjusted is optimized according to the first value.

[0031] In the above technical solution, the first performance indicator can be optimized by reducing the jitter of the first performance indicator. Moreover, through the pre-set conditions, when reducing the jitter of the first performance indicator, it can be ensured that the degradation degree of the second performance indicator will not be too large, and even the second performance indicator can be optimized.

[0032] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: performing a hypothesis test on the variance of a first set of values ​​of the first performance indicator and the variance of a second set of values ​​of the first performance indicator based on a second hypothesis; when the hypothesis test passes, determining that the variance change of the first performance indicator is within a preset range.

[0033] Exemplarily, a chi-square test or an F test may be used to perform a hypothesis test on the variance of the first set of values ​​of the first performance indicator and the variance of the second set of values ​​of the first performance indicator. Exemplarily, the second hypothesis may be the null hypothesis of the chi-square test or the F test, and when the null hypothesis is established, it is determined that the hypothesis test passes.

[0034] In the above technical solution, the jitter of the first performance indicator can be reduced by means of hypothesis testing, and the solution is simple and easy to deploy.

[0035] In combination with the first aspect, in some implementations of the first aspect, the method further includes: when the value change of the first performance indicator is not caused by the parameter to be adjusted changing from the second value to the first value, re-obtaining the value of the parameter to be adjusted.

[0036] In the above technical solution, when the value change of the first performance indicator is not caused by the change of the parameter to be adjusted from the second value to the first value, it means that the value change of the first performance indicator may be caused by noise such as indicator jitter. If the parameter to be adjusted is optimized according to the first value at this time, the parameter optimization algorithm may not converge, thereby affecting the parameter optimization efficiency. Therefore, when the value change of the first performance indicator is not caused by the change of the parameter to be adjusted from the second value to the first value, re-obtaining the value of the parameter to be adjusted is helpful to improve the parameter optimization efficiency.

[0037] In combination with the first aspect, in some implementations of the first aspect, optimizing the parameter to be adjusted according to the first value includes: adjusting the first value to a third value of the parameter to be adjusted so that the operating result of the first device meets the set condition.

[0038] Exemplarily, the operation result of the first device includes the operation result of the first service and / or the operation result of the second service.

[0039] Exemplarily, the operating result of the first device satisfies the set conditions, which may include that the value of the first performance indicator reaches a set range; or, it may include that the value of the second performance indicator is within a set range; or, it may include that the jitter of the first performance indicator is within a set range.

[0040] Exemplarily, when the parameter to be adjusted is a third value, the parameter optimization algorithm reaches convergence.

[0041] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: when optimizing the parameter to be adjusted according to the first numerical value, recording an optimization log, the optimization log includes at least one of the following: the reason why the first numerical value of the parameter to be adjusted is modified, the numerical value determined for the parameter to be adjusted that exceeds the set range, the numerical value of the parameter to be adjusted when the first performance indicator fails the hypothesis test, and the numerical value of the parameter to be adjusted when the first performance indicator passes the hypothesis test but the second performance indicator fails the hypothesis test.

[0042] Exemplarily, the reasons why the first value of the parameter to be adjusted is modified may include: the first performance parameter corresponding to the first value does not reach the set range or the first value does not make the parameter optimization algorithm converge; when the parameter to be adjusted becomes the first value, the change of the first performance indicator is caused by jitter; when the parameter to be adjusted becomes the first value, the jitter of the first performance indicator exceeds the preset range; when the parameter to be adjusted becomes the first value, the value change of the second performance indicator deteriorates beyond the limit threshold.

[0043] Exemplarily, the failure of the first performance indicator to pass the hypothesis test may include: when the parameter to be adjusted becomes the first value, the change in the first performance indicator is caused by indicator jitter; or, when the first performance indicator is required to be optimized, the degree of degradation of the first performance indicator is high; or, the jitter of the first performance indicator exceeds the set range.

[0044] Exemplarily, the first performance indicator passes the hypothesis test but the second performance indicator fails the hypothesis test, which may include: a value change of the second performance indicator exceeds a deterioration limit threshold.

[0045] In the above technical solution, during the parameter optimization process, recording the abnormal values ​​of the parameters to be adjusted and / or the causes of the abnormalities helps to improve the problem location capability during the parameter optimization process, helps to determine the causes of parameter optimization failure, and modifies the relevant configurations of the parameter optimization algorithm, thereby improving the parameter optimization efficiency.

[0046] Exemplarily, the first business and / or the second business are related to autonomous driving.

[0047] In a second aspect, a parameter optimization device is provided, which includes a first acquisition unit and a processing unit, wherein the first acquisition unit is used to: acquire a first set of values ​​of a first performance indicator, the first set of values ​​of the first performance indicator is generated when the first service runs within a first time period based on the first value of the parameter to be adjusted, the first performance indicator is associated with the performance of the first service, and the first service runs on a first device; the processing unit is used to: determine, based on the first set of values ​​of the first performance indicator and the second set of values ​​of the first performance indicator, whether the value change of the first performance indicator is caused by the parameter to be adjusted changing from the second value to the first value, the second set of values ​​of the first performance indicator is generated when the first service runs within a second time period based on the second value of the parameter to be adjusted; and optimize the parameter to be adjusted according to the first value.

[0048] In combination with the second aspect, in certain implementations of the second aspect, the first acquisition unit is also used to: obtain a first set of values ​​of a second performance indicator within a first time period, the second performance indicator is associated with the performance of a second service, and the second service runs on the first device; the processing unit is also used to: determine, based on the second set of values ​​of the second performance indicator and the first set of values ​​of the second performance indicator, a value change of the second performance indicator when the parameter to be adjusted changes from the second value to the first value, the second set of values ​​of the second performance indicator is generated when the second service runs during the second time period based on the second value of the parameter to be adjusted; when the value change of the second performance indicator does not exceed the deterioration limit threshold, the parameter to be adjusted is optimized according to the first value.

[0049] In combination with the second aspect, in certain implementations of the second aspect, the processing unit is used to: perform a hypothesis test on the average value of a first group of values ​​of a first performance indicator and the average value of a second group of values ​​of the first performance indicator based on a first hypothesis; when the hypothesis test passes, determine that the numerical change of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value.

[0050] In combination with the second aspect, in certain implementations of the second aspect, the processing unit is also used to: determine whether the variance change of the first performance indicator is within a preset range based on a first set of values ​​of the first performance indicator and a second set of values ​​of the first performance indicator; when the variance change of the first performance indicator is within the preset range, optimize the parameter to be adjusted according to the first value.

[0051] In combination with the second aspect, in certain implementations of the second aspect, the processing unit is also used to: perform a hypothesis test on the variance of the first group of values ​​of the first performance indicator and the variance of the second group of values ​​of the first performance indicator based on the second hypothesis; when the hypothesis test passes, determine that the variance change of the first performance indicator is within a preset range.

[0052] In combination with the second aspect, in some implementations of the second aspect, the device also includes a second acquisition unit, which is used to: when the value change of the first performance indicator is not caused by the parameter to be adjusted changing from the second value to the first value, re-acquire the value of the parameter to be adjusted.

[0053] In combination with the second aspect, in some implementations of the second aspect, the processing unit is used to: adjust the first value to a third value of the parameter to be adjusted so that the operation result of the first device meets the set condition.

[0054] In combination with the second aspect, in certain implementations of the second aspect, the device also includes a recording unit, which is used to: when the parameter to be adjusted is optimized according to the first numerical value, record an optimization log, and the optimization log includes at least one of the following items: the reason why the first numerical value of the parameter to be adjusted is modified, the numerical value determined for the parameter to be adjusted that exceeds the set range, the numerical value of the parameter to be adjusted when the first performance indicator fails the hypothesis test, and the numerical value of the parameter to be adjusted when the first performance indicator passes the hypothesis test but the second performance indicator fails the hypothesis test.

[0055] Exemplarily, the first business and / or the second business are related to autonomous driving.

[0056] In a third aspect, a parameter optimization device is provided, which includes: a memory for storing a computer program; and a processor for executing the computer program stored in the memory, so that the device executes the method in any possible implementation of the first aspect.

[0057] In a fourth aspect, an electronic device is provided, which includes a device as in any possible implementation of the second aspect or the third aspect.

[0058] In combination with the fourth aspect, in certain implementations of the fourth aspect, the electronic device is an intelligent driving device, such as a vehicle.

[0059] In a fifth aspect, a computer program product is provided, the computer program product comprising: a computer program code, when the computer program code is run on a computer, the computer executes the method in any possible implementation of the first aspect.

[0060] It should be noted that the above-mentioned computer program code may be stored in whole or in part on a first storage medium, wherein the first storage medium may be packaged together with the processor or may be packaged separately from the processor.

[0061] In a sixth aspect, a computer-readable medium is provided, wherein the computer-readable medium stores instructions, and when the instructions are executed by a processor, the processor implements the method in any possible implementation manner of the first aspect.

[0062] In a seventh aspect, a chip is provided, the chip comprising a circuit, the circuit being used to execute the method in any possible implementation manner of the first aspect above. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a schematic diagram of an autonomous driving system;

[0064] Figure 2 It is a structural schematic diagram of a parameter optimization system according to an embodiment of the present application;

[0065] Figure 3 is a schematic flow chart of a parameter optimization method provided in an embodiment of the present application;

[0066] Figure 4 is a schematic block diagram of a parameter optimization device provided in an embodiment of the present application;

[0067] Figure 5 It is another schematic block diagram of the parameter optimization device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0068] In order to facilitate understanding of the solutions of the embodiments of the present application, the concepts involved in the present application are first introduced:

[0069] 1. Autonomous driving operating system (hereinafter referred to as autonomous driving system): for example, an operating system based on the portable operating system interface (POSIX) standard, suitable for high-performance computing and high-bandwidth communication required for autonomous driving, providing vehicle environmental perception, sensor fusion, path planning and other functions.

[0070] 2. Autonomous driving services: In an autonomous driving system, services are usually a collection of processes and / or threads encapsulated with communication middleware and having specific autonomous driving functions. The communication middleware can be a robot operating system (ROS), an automotive open system architecture communication management module (AUTOSAR CM), etc. The autonomous driving function can include vehicle environmental perception, sensor fusion, and path planning.

[0071] 3. Performance indicator (or indicator): a parameter used as a reference in adjusting parameters that affect the performance of autonomous driving services.

[0072] 4. Index jitter: When no parameter adjustment / optimization is performed, performance indicators are inherently unstable due to the influence of hardware computing power, memory bandwidth, OS scheduling capabilities, business triggering methods, etc.

[0073] 5. Hypothesis testing: It is a statistical inference method used to determine whether the differences between samples or between samples and the population are caused by sampling errors or essential differences. Hypothesis testing includes but is not limited to the following:

[0074] (1) Z test: also known as U test, is a method used to test the difference of sample means. It uses the theory of standard normal distribution (or Z distribution) to infer the probability of difference, so as to compare whether the difference between two means is significant. When the population standard deviation is known, the Z test is used to verify whether the sample mean is equal to a certain expected value.

[0075] (2) T test: Use t distribution theory to infer the probability of a difference, and thus compare whether the difference between two means is significant. When the population standard deviation is unknown, use the T test to verify whether the sample mean is equal to a certain expected value.

[0076] It should be noted that the two parameters of the normal distribution, μ (sample mean) and σ (population standard deviation), determine the position and form of the normal distribution. The general normal variable X is often transformed into a standard normal variable Z (or Z statistic) through Z transformation, that is, Z = [(X-μ) / σ], so that the original normal distribution of various forms is converted into a standard normal distribution with μ = 0 and σ = 1. According to the central limit theorem, sampling simulation experiments show that when a number of samples are drawn from a normal distribution population with a fixed number of n, the distribution of the sample mean still obeys the normal distribution, that is, N (μ, σ). Therefore, the distribution of the sample mean can also be transformed into the standard normal distribution N (0, 1) by Z transformation. In actual use, the population standard deviation of the sample is usually unknown. Therefore, when transforming the normal variable X, S (sample variance) is often used as the estimated value of σ, that is, t = [(X-μ) / S]. In order to distinguish it from the Z transformation, the above transformation is called t transformation, and the distribution of the t statistic is called t distribution.

[0077] It can be understood that, according to the properties of the standard normal distribution and the t-distribution, the z-value and the t-value include positive values ​​and negative values.

[0078] (3) Chi-square test: also known as x 2 The chi-square test is used to verify whether there is a significant difference between a certain ratio of two samples. The chi-square test is a hypothesis testing method based on the chi-square distribution. The chi-square distribution means that if n independent random variables all obey the standard normal distribution, then the sum of the squares of these n random variables that obey the standard normal distribution constitutes a new random variable, and the distribution law of the new random variable is called the chi-square distribution.

[0079] (4) F test: also known as joint hypothesis test, variance ratio test or variance homogeneity test, it is a significance test used to determine whether there is a difference in variance between two or more groups of samples. The F test is a hypothesis testing method based on the F distribution. The F distribution means: suppose x and y are two independent random variables, x follows a distribution with n degrees of freedom, and y follows a distribution with m degrees of freedom. The distribution law of the ratio of these two independent chi-square distributions divided by their respective degrees of freedom is called the F distribution.

[0080] 6. Test statistics: random variables calculated based on sample data, including Z statistics, t statistics, F statistics, chi-square statistics, etc.

[0081] 7. Null hypothesis: also known as the hypothesis to be tested, the null hypothesis, or the eliminated hypothesis, generally denoted as H0. In hypothesis testing, the hypothesis of a proposition about the overall distribution to be tested is called the null hypothesis.

[0082] 8. Alternative hypothesis: Also known as the alternative hypothesis, alternative hypothesis, generally denoted as H1. In hypothesis testing, the alternative hypothesis contains all propositions about the population distribution that make the null hypothesis invalid.

[0083] 9. Confidence: also known as significance level, significance stage, trust coefficient, etc., refers to the probability of making an error when estimating that the population parameter falls within a certain interval, generally represented by α (0<α<1).

[0084] 10. Rejection region: refers to the range of values ​​of the statistic based on which the null hypothesis is rejected in the hypothesis test. In the hypothesis test, according to the distribution of the test statistic, the interval for rejecting the null hypothesis H0 determined by the given probability α (0<α<1) as the significance level is called the rejection region.

[0085] Under the current technical background, in order to improve the efficiency of parameter optimization, parameter optimization algorithms are generally used for parameter optimization. Commonly used parameter optimization algorithms include heuristic search algorithms and reinforcement learning algorithms. Among them, heuristic search algorithms include particle swarm optimization algorithms and simulated annealing algorithms, and reinforcement learning algorithms include proximal policy optimization (PPO) algorithms and deep deterministic policy gradient (DDPG) algorithms. The goal of the heuristic search algorithm is to find the global optimal solution through the principle of heuristics; the focus of the reinforcement learning algorithm is to improve the convergence ability of the reinforcement learning process itself, and achieve convergence to obtain the optimal solution of the parameters to be adjusted. However, in the actual implementation process, not only the parameter adjustment during the optimization process may cause changes in the indicators, but also the indicators themselves have jitters, that is, the indicators without parameter adjustment may also change at all times. In the above-mentioned parameter optimization algorithms, the situation of indicator jitter is not considered, resulting in the inability to determine whether the difference between the indicators before and after the parameter adjustment is the result of the parameter adjustment or the indicator jitter. This may lead to deviations in the direction of parameter optimization and even cause the parameter optimization algorithm to fail to converge.

[0086] In view of this, the embodiments of the present application provide a parameter optimization method, device, and electronic device, which can determine whether the difference in the index before and after the parameter adjustment is caused by the index jitter, and select the parameter value corresponding to the index change caused by non-index jitter as the input of the next iteration of the parameter optimization algorithm, which helps to improve the convergence of the parameter optimization algorithm. When adjusting the parameters, taking multiple other parameters as performance indicators can achieve simultaneous optimization of multiple objectives, avoiding the significant degradation of other performance indicators while optimizing one performance indicator.

[0087] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0088] Figure 1A schematic block diagram of an automatic driving system in an intelligent driving device is shown, and the automatic driving system 100 may include an automatic driving software system 110 and an automatic driving hardware platform 120. The automatic driving software system 110 includes an automatic driving service 111, a communication middleware 112, and an automatic driving operating system 113. Among them, the automatic driving service 111 may include one or more services, each of which is used to implement one or more functions; the communication middleware 112 is responsible for docking with the automatic driving operating system 113, providing an operating and development environment for the automatic driving service 111, for example, providing communication, resource management and other services for the automatic driving service 111; the automatic driving operating system 113 runs on the automatic driving hardware platform 120, and can perform resource scheduling according to the actual needs of the automatic driving service 111 during operation. The automatic driving hardware platform 120 can provide basic hardware resources for the automatic driving software system 110, for example, the automatic driving service 111 obtains the computing resources, memory resources, communication resources, general storage resources and perception data resources required for operation from the automatic driving hardware platform 120 through the automatic driving operating system 113.

[0089] The intelligent driving devices involved in the embodiments of the present application may include road vehicles, water vehicles, air vehicles, industrial equipment, agricultural equipment, or entertainment equipment, etc. For example, the intelligent driving device may be a vehicle, which is a vehicle in a broad sense, and may be a vehicle (such as a commercial vehicle, a passenger car, a motorcycle, a flying car, a train, etc.), an industrial vehicle (such as a forklift, a trailer, a tractor, etc.), an engineering vehicle (such as an excavator, a bulldozer, a crane, etc.), agricultural equipment (such as a lawn mower, a harvester, etc.), amusement equipment, a toy vehicle, etc. The embodiments of the present application do not specifically limit the type of vehicle.

[0090] For example, Figure 1The automatic driving system 100 shown can be set in the computing platform of the intelligent driving device. Taking the intelligent driving device as a vehicle as an example, the computing platform may include any one of the following: vehicle domain controller (VDC), automatic driving domain controller (ADC), cockpit domain controller (CDC). For another example, the computing platform may also include but is not limited to: in-car application-server (ICAS) controller, body domain controller (BDC), special equipment system (SAS), media graphics unit (MGU), body super core (BSC), advanced driving assistant system super core (ADAS super core). Among them, ICAS may include at least one of the following: vehicle control server ICAS1, intelligent driving server ICAS2, intelligent cockpit server ICAS3, infotainment server ICAS4.

[0091] Figure 2 FIG. 1 shows a schematic block diagram of a parameter optimization system provided in an embodiment of the present application. Figure 2As shown, the parameter optimization system 200 includes a middleware / OS scheduling module 210, a business to be tuned 220, a test module 230 and an optimization algorithm module 240. Among them, the test module 230 includes a variety of hypothesis testing methods, such as Z test, T test, F test, chi-square test, each hypothesis testing method is related to the parameter to be adjusted and the optimization target; the optimization algorithm module 240 may include a variety of optimization algorithms, such as particle swarm optimization algorithm, simulated annealing algorithm, PPO algorithm, DDPG algorithm, etc. When the parameter optimization system 200 is running, it is necessary to obtain the pre-configured items associated with the business to be tuned 220, including the index to be optimized and the parameter to be adjusted. In addition, during the parameter optimization process of the parameter optimization system 200, it is necessary to record and optimize the relevant log information to monitor the optimization process. When the business to be tuned 220 is running, the value of the parameter to be adjusted and the value of the index to be optimized are output to the test module 230. The inspection module 230 performs a hypothesis test based on the value of the parameter to be adjusted and the value of the indicator to be optimized to determine whether the indicator to be optimized passes the hypothesis test. When the indicator to be optimized passes the hypothesis test, it means that the change in the value of the indicator to be optimized is not caused by indicator jitter. Then the inspection module 230 inputs the value of the parameter to be adjusted at this time into the optimization algorithm module 240, so that the optimization algorithm module 240 performs optimization calculation based on the value of the parameter to be adjusted. Further, the optimization algorithm module 240 configures the calculated optimized value of the parameter to be adjusted to the middleware / OS scheduling module 210. After the configuration is successful, the numerical collection of the indicator to be optimized of the business to be adjusted 220 continues, and the above steps are repeated until the optimization algorithm module 240 outputs the final optimized value of the parameter to be adjusted corresponding to the convergence of the optimization algorithm. When the final optimized value is deployed and operated on a real vehicle (such as in the communication middleware of the real vehicle), the end-to-end performance of the vehicle's autonomous driving business can be improved.

[0092] Exemplarily, the pre-configuration items may include but are not limited to:

[0093] 1. Scenario configuration items: indicators to be optimized, deterioration limit indicators, deterioration limit thresholds, single iteration duration, and statistical time windows. Among them, the deterioration limit indicator is an indicator that the user expects not to deteriorate or deteriorate within a certain range during the optimization of the parameters to be adjusted; the deterioration limit threshold indicates the value that the deterioration of a certain performance indicator (such as the deterioration limit indicator or the indicator to be optimized) cannot exceed during the optimization of the parameters to be adjusted; the single iteration duration is the duration of one iteration of the parameter optimization algorithm; and the statistical time window indicates the time period for obtaining the indicators to be optimized.

[0094] The deterioration limit threshold may be ±10% or ±5% of the deterioration limit indicator value before the parameter to be adjusted is adjusted, or may be other values.

[0095] In one example, the larger the value of a certain performance indicator is, the worse the performance of the business associated with the performance indicator is, and the deterioration limit threshold can be 10% or 5% of the value of the performance indicator before the parameter to be adjusted is adjusted. Furthermore, the change in the value of the performance indicator does not exceed the deterioration limit threshold, which may include: the difference between the value of the performance indicator after the parameter to be adjusted is adjusted and the value of the performance indicator before the parameter to be adjusted is less than or equal to the deterioration limit threshold. For example, if the performance indicator is CPU usage, and the value of the performance indicator before the parameter to be adjusted is 50%, the deterioration limit threshold can be 5%; if the average value of the first group of values ​​of the performance indicator is 55%, the change in the value of the performance indicator does not exceed the deterioration limit threshold, and if the average value of the first group of values ​​of the performance indicator is 56%, the change in the value of the performance indicator exceeds the deterioration limit threshold.

[0096] In another example, the smaller the value of a certain performance indicator, the worse the performance of the business associated with the performance indicator, and the deterioration limit threshold may be -10% or -5% of the average value of the second group of values ​​of the performance indicator. Furthermore, the change in the value of the performance indicator does not exceed the deterioration limit threshold, which may include: the difference between the average value of the first group of values ​​of the performance indicator and the average value of the second group of values ​​of the performance indicator is greater than or equal to the deterioration limit threshold. For example, if the performance indicator is the GPU frame rate, and the average value of the second group of values ​​of the performance indicator is 50 frames per second (fps), the deterioration limit threshold may be -5fps; if the average value of the first group of values ​​of the performance indicator is 46fps, the change in the value of the performance indicator does not exceed the deterioration limit threshold; if the average value of the first group of values ​​of the performance indicator is 44fps, the change in the value of the performance indicator does not exceed the deterioration limit threshold.

[0097] It should be noted that the value of the above-mentioned deterioration limit index may be an average value of multiple values ​​of the deterioration limit index collected within a period of time.

[0098] 2. Hypothesis test configuration items: hypothesis test method, such as one or more selected from the above-mentioned hypothesis test methods; confidence level, such as 0.95, 0.99, etc.

[0099] 3. Optimization algorithm configuration items: the parameter optimization algorithm used, such as one of the particle swarm optimization algorithm, simulated annealing algorithm, PPO algorithm, and DDPG algorithm; algorithm parameter configuration, such as the parameter search range and search step size of the optimization algorithm.

[0100] 4. Configuration items of parameters to be adjusted: search range configuration, search step configuration, discrete / continuous configuration, etc. It should be noted that some parameters to be adjusted are variables that can take any value within a certain range, that is, two adjacent values ​​can be infinitely divided, for example, a parameter to be adjusted can take any value within the range of 0 to 5; some parameters to be adjusted are variables whose values ​​can be listed one by one in a certain order, for example, a parameter to be adjusted can take any value among 0, 1, 2, 3, 4, and 5. Discrete / continuous configuration is used to configure whether to search in continuous space or in discrete space when searching for the value of the parameter to be adjusted.

[0101] Exemplarily, during the operation of the parameter optimization system 200, optimization-related log information is recorded to monitor the optimization process, so as to facilitate problem location and feedback on optimization effects. Exemplarily, the optimization-related log information may include at least one of the following:

[0102] 1. Records of the optimization algorithm search process, including: the values ​​of the selected parameters to be adjusted; records of optimization effects, such as changes in the values ​​of the indicators to be optimized, changes in the values ​​of the deterioration limit indicators, etc.

[0103] 2. Records of unaccepted solutions: abnormal solutions and normal but unaccepted solutions. Among them, abnormal solutions may include the value of the parameter to be adjusted caused by the parameter search exceeding the set range. Normal but unaccepted solutions may include: when the value change of the indicator to be optimized is caused by indicator jitter, the corresponding value of the parameter to be adjusted; when the value change of the indicator to be optimized is not caused by indicator jitter, but the value of the parameter to be adjusted causes the change of the deterioration limit indicator to exceed the deterioration limit threshold, the corresponding value of the parameter to be adjusted.

[0104] 3. Abnormal status records: abnormal parameter search range, abnormal indicators to be optimized, etc.

[0105] 4. Hypothesis test value calculation records: Since different hypothesis tests have different calculation methods, it is necessary to record the calculation process to locate problems such as abnormal input and calculation process errors.

[0106] Exemplarily, the middleware / OS scheduling module 210 may be arranged in Figure 1 In the communication middleware 112 shown in FIG. 1 , the service to be tuned 220 can be set Figure 1 In the autonomous driving business 111 shown; the inspection module 230 can be set in the communication middleware 112, or the inspection module 230 can also be set in the autonomous driving operating system 113, or the inspection module 230 can also be set in a module alone; the optimization algorithm module 240 can be set in the communication middleware 112, or the optimization algorithm module 240 can also be set in a module alone.

[0107] In some possible implementations, the inspection module 230 and the optimization algorithm module 240 may be provided in one module.

[0108] It should be understood that Figure 2 The architecture shown is only an example. Figure 2 The depicted systems may include more or fewer modules.

[0109] based on Figure 2 The parameter optimization system shown in the figure, this application proposes a parameter optimization method.

[0110] Figure 3 A schematic flow chart of a parameter optimization method provided in an embodiment of the present application is shown. The method 300 may be performed by Figure 2 The parameter optimization system 200 shown is performed, for example, by Figure 2 As shown, the inspection module 230 executes the method 300 including S301 to S303 .

[0111] S301, obtaining a first set of values ​​of a first performance indicator, where the first set of values ​​of the first performance indicator is generated when a first service runs within a first time period based on a first value of a parameter to be adjusted, the first performance indicator is associated with performance of the first service, and the first service runs on a first device.

[0112] Exemplarily, the first performance indicator may include the indicator to be optimized in the above embodiment, the first service may include the service to be optimized in the above embodiment, and the parameter to be adjusted may include the parameter to be adjusted in the above embodiment. The first performance indicator is associated with the performance of the first service, which can be understood as: the first performance indicator affects the performance of the first service.

[0113] Exemplarily, the first service is related to autonomous driving, for example, it may be a regulation and control service, a perception service, etc. in an autonomous driving function.

[0114] Exemplarily, the first set of values ​​of the first performance indicator includes multiple values ​​of the first performance indicator.

[0115] For example, in order to optimize the memory bandwidth of the CPU associated with the regulation and control business or the perception business, the memory system resource partitioning and monitoring (MPAM) can be taken as the parameter to be adjusted. Since the CPU memory bandwidth occupancy can be directly measured by the CPU utilization rate, the CPU utilization rate can be taken as the first performance indicator. Since the CPU utilization rate itself is also a constantly changing value, in the autonomous driving system, there may be thousands of threads running at the same time, and the operating environment is constantly changing. Therefore, the CPU utilization rate itself is constantly fluctuating under the influence of these factors. It is necessary to confirm whether the change in the CPU utilization rate value is an indicator jitter through hypothesis testing.

[0116] S302, based on the first group of values ​​of the first performance indicator and the second group of values ​​of the first performance indicator, determine that the value change of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value, and the second group of values ​​of the first performance indicator is generated when the first service runs based on the second value of the parameter to be adjusted during the second time period.

[0117] Exemplarily, the second set of values ​​of the first performance indicator includes a plurality of data of the first performance indicator.

[0118] Exemplarily, the length of the second time period may be the same as that of the first time period, or the length of the second time period may be different from that of the first time period.

[0119] In some possible implementations, the second value of the parameter to be adjusted may be a value before parameter optimization is performed, and the first value of the parameter to be adjusted may be a value obtained after the first optimization algorithm is iterated once or multiple times; or, the second value of the parameter to be adjusted may be a legal value output by the last iteration of the first optimization algorithm, and the first value of the parameter to be adjusted may be a value output by the current iteration of the first optimization algorithm. The legal value may be understood as a value that has passed the verification of the verification module.

[0120] In some possible implementations, it may be determined through hypothesis testing whether the change in the value of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value. Exemplarily, a hypothesis test is performed on the average value of the first group of values ​​of the first performance indicator and the average value of the second group of values ​​of the first performance indicator according to the first hypothesis; when the hypothesis test passes, it is determined that the change in the value of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value.

[0121] Exemplarily, the hypothesis test method may be as shown in Table 1. The first hypothesis may be any null hypothesis H0 in Table 1, and the hypothesis test passing means that the null hypothesis H0 is established.

[0122] Table 1

[0123]

[0124] Among them, for the test statistic and represents the sample mean of the first group of values ​​of the first performance indicator, n represents the sample size of the first group of values ​​of the first performance indicator, σ represents the population standard deviation of the first group of values ​​of the first performance indicator, μ represents the population mean of the first group of values ​​of the first performance indicator, μ0 represents the population mean of the second group of values ​​of the first performance indicator, z represents the value of the statistic Z, α represents the confidence level, z α represents the z value obtained by looking up the z distribution table according to the confidence level α, t represents the value of the statistic t, and t α (n-1) represents the t value obtained by looking up the t distribution table according to the degree of freedom n-1 and the confidence level α, wherein α can be 0.90, or 0.95, or other values ​​between 0 and 1.

[0125] In an example, if the change in the value of the parameter to be adjusted does not lead to an increase in the value of the first performance indicator, when the value of the first performance indicator increases, it can be considered that it is caused by indicator jitter. Then the null hypothesis can be taken as μ≤μ0, the alternative hypothesis is μ>μ0, and the rejection region is z≥z α Furthermore, if z≥z is not satisfied α , then the original hypothesis μ≤μ0 holds, that is, when the parameter to be adjusted changes from the second value to the first value, the value of the first performance indicator decreases or remains unchanged, that is, the change is not caused by the jitter of the first performance indicator, that is, the hypothesis test passes. If z≥z α , then the alternative hypothesis μ>μ0 holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the value of the first performance indicator increases, and the change is caused by the jitter of the first performance indicator.

[0126] In another example, if the change in the value of the parameter to be adjusted does not lead to an increase in the value of the first performance indicator, when the value of the first performance indicator increases, it can be considered that it is caused by indicator jitter. Then the null hypothesis can be taken as μ≤μ0, the alternative hypothesis is μ>v0, and the rejection region is t≥t α (n-1). Furthermore, if t≥t is not satisfied α (n-1), then the original hypothesis μ≤μ0 holds, that is, when the parameter to be adjusted changes from the second value to the first value, the value of the first performance indicator decreases or remains unchanged, that is, the change is not caused by the jitter of the first performance indicator, that is, the hypothesis test passes. α(n-1), then the alternative hypothesis μ>μ0 holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the value of the first performance indicator increases, and the change is caused by the jitter of the first performance indicator.

[0127] By analogy, the test statistic in Table 1 is And the test statistic is The relationship between the null hypothesis, alternative hypothesis, and rejection region of the other rows can be deduced from the above description, which will not be repeated here. In actual use, a specific hypothesis testing method can be selected according to the specific properties of the first performance indicator (for example, whether the first performance indicator should increase or decrease after the parameter to be adjusted is adjusted).

[0128] In addition, the test statistics in Table 1 are and The corresponding hypothesis test method can determine whether the change of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value, and whether the first performance indicator is optimized or degraded. For example, the increase in the value of the first performance indicator indicates that the first performance indicator is optimized. If the null hypothesis is μ≥μ0, then when the null hypothesis μ≥μ0 is established, it means that the first performance indicator is optimized.

[0129] As described above, the first set of values ​​of the first performance indicator includes multiple values ​​of the first performance indicator. The population of the first set of values ​​of the first performance indicator can be all of the above multiple values, and the sample of the first set of values ​​of the first performance indicator can be some values ​​selected from the above multiple values.

[0130] For the test statistic and in, represents the sample mean of the first set of values ​​of the first performance indicator, represents the sample mean of the second set of values ​​of the first performance indicator, n1 represents the sample size of the first set of values ​​of the first performance indicator, n2 represents the sample size of the second set of values ​​of the first performance indicator, represents the population variance of the first set of values ​​of the first performance indicator, represents the population variance of the second set of values ​​of the first performance indicator, μ1 represents the population mean of the first set of values ​​of the first performance indicator, μ2 represents the population mean of the second set of values ​​of the first performance indicator, represents the combined sample variance, i.e., the variance of the samples of the first set of values ​​of the first performance indicator and the samples of the second set of values ​​of the first performance indicator after being combined, and δ represents the difference between the sample mean of the second set of values ​​of the first performance indicator and the sample mean of the first set of values ​​of the first performance indicator. In the embodiment of the present application, δ may be a deterioration limit threshold, for example, δ = ±0.1*μ2.

[0131] In one example, if the change in the value of the parameter to be adjusted causes the value of the first performance indicator to increase by no more than δ, when the value of the first performance indicator increases by more than δ, it can be considered that it is caused by indicator jitter. Then the original hypothesis can be μ1-μ2≤δ, the alternative hypothesis is μ1-μ2>δ, and the rejection region is z≥z α Furthermore, if z≥z is not satisfied α , then the original hypothesis μ1-μ2≤δ holds, that is, when the parameter to be adjusted changes from the second value to the first value, the increase in the value of the first performance indicator does not exceed δ, that is, the hypothesis test passes. If z≥z α , then the alternative hypothesis μ1-μ2>δ holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the value increase of the first performance indicator exceeds δ, and the value increase of the first performance indicator is caused by the jitter of the first performance indicator.

[0132] In another example, if the change in the value of the parameter to be adjusted causes the value of the first performance indicator to increase by no more than δ, when the value of the first performance indicator increases by more than δ, it can be considered that it is caused by indicator jitter. Then the original hypothesis can be μ1-μ2≤δ, the alternative hypothesis can be μ1-μ2>δ, and the rejection region is t≥t α (n1+n2-2). Furthermore, if t≥t α (n1+n2-2), then the original hypothesis μ1-μ2≤δ holds, that is, when the parameter to be adjusted changes from the second value to the first value, the increase in the value of the first performance indicator does not exceed δ, that is, the hypothesis test passes. If t≥t α (n1+n2-2), then the alternative hypothesis μ1-μ2>δ holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the value increase of the first performance indicator exceeds δ, and the value increase of the first performance indicator is caused by the jitter of the first performance indicator.

[0133] By analogy, the test statistic in Table 1 is And the statistics are The relationship between the null hypothesis, alternative hypothesis, and rejection region of the other rows can be deduced from the above description, which will not be repeated here. In actual use, a specific hypothesis testing method can be selected according to the specific properties of the first performance indicator (for example, the expected increase or decrease of the first performance indicator after the parameter to be adjusted is adjusted).

[0134] It is understandable that for the test statistic and The size of δ can characterize the degree of optimization or degradation of the first performance indicator. In the specific implementation process, by adjusting the size of δ, and then according to whether the size relationship between the difference between μ1 and μ2 and δ satisfies the original hypothesis, it can be determined whether the degree of optimization or degradation of the first performance indicator meets expectations.

[0135] For the test statistic represents the sample size of the first set of values ​​of the first performance indicator and the second set of values ​​of the first performance indicator, that is, the sample sizes of the two sets of samples are equal, μ D is the difference between the overall mean of the first set of values ​​of the first performance indicator and the overall mean of the second set of values ​​of the first performance indicator, represents the difference between the sample mean of the first set of values ​​of the first performance indicator and the sample mean of the second set of values ​​of the first performance indicator, S D The standard deviation of the samples of the first group of values ​​of the first performance indicator and the samples of the second group of values ​​of the first performance indicator. When the number of the first group of values ​​of the first performance indicator and the number of the second group of values ​​of the first performance indicator are the same, the test statistic can be used to Conduct hypothesis testing.

[0136] S303: Optimize the parameter to be adjusted according to the first value.

[0137] Exemplarily, the first value may be a value determined according to a first optimization algorithm, and the first optimization algorithm may be any optimization algorithm in the optimization algorithm module 240. Optimizing the parameter to be adjusted according to the first value may include: inputting the first value into the first optimization algorithm, so that the first optimization algorithm optimizes the parameter to be adjusted based on the first value.

[0138] In some possible implementations, the method also includes: obtaining a first set of values ​​of a second performance indicator within a first time period, the second performance indicator is associated with the performance of a second service, and the second service runs on the first device; determining a value change of the second performance indicator when the parameter to be adjusted changes from the second value to the first value based on the second set of values ​​of the second performance indicator and the first set of values ​​of the second performance indicator, the second set of values ​​of the second performance indicator being generated when the second service runs within the second time period based on the second value of the parameter to be adjusted; and optimizing the parameter to be adjusted based on the first value, including: optimizing the parameter to be adjusted based on the first value when the value change of the second performance indicator does not exceed a deterioration limit threshold.

[0139] Exemplarily, the second performance indicator may include the deterioration limit indicator in the above embodiment, and the deterioration limit threshold may include the deterioration limit threshold in the above embodiment. A hypothesis test may be performed on the average value of the first group of values ​​of the second performance indicator and the average value of the second group of values ​​of the second performance indicator according to the third hypothesis; when the hypothesis test passes, it is determined that the value change of the second performance indicator does not exceed the deterioration limit threshold. The third hypothesis may be the null hypothesis H0 described in the third or fourth row of Table 1.

[0140] When the hypothesis test method represented by the third or fourth row in Table 1 is applied to determine whether the value change of the second performance indicator exceeds the deterioration limit threshold, represents the sample mean of the first set of values ​​of the second performance indicator, represents the sample mean of the second set of values ​​of the second performance indicator, n1 represents the sample size of the first set of values ​​of the second performance indicator, n2 represents the sample size of the second set of values ​​of the second performance indicator, represents the population variance of the first set of values ​​of the second performance indicator, represents the population variance of the second set of values ​​of the second performance indicator, μ1 represents the population mean of the first set of values ​​of the second performance indicator, μ2 represents the population mean of the second set of values ​​of the second performance indicator, represents the combined sample variance, that is, the variance of the samples of the first set of values ​​of the second performance indicator and the samples of the second set of values ​​of the second performance indicator after merging, and δ represents the difference between the sample mean of the second set of values ​​of the second performance indicator and the sample mean of the first set of values ​​of the second performance indicator.

[0141] In one example, if the change in the value of the parameter to be adjusted is expected to result in the value of the second performance indicator not increasing by more than the deterioration limit threshold δ, for example, δ = 0.1*μ2, then the null hypothesis can be μ1-μ2≤δ, the alternative hypothesis can be μ1-μ2>δ, and the rejection region can be z≥z α Furthermore, if z≥z is not satisfied α , then the original hypothesis μ1-μ2≤δ holds, that is, when the parameter to be adjusted changes from the second value to the first value, the increase in the value of the second performance indicator does not exceed the deterioration limit threshold δ, that is, the hypothesis test passes. If z≥z α , then the alternative hypothesis μ1-μ2>δ holds, that is, when the parameter to be adjusted changes from the second value to the first value, the increase in the value of the second performance indicator exceeds the deterioration limit threshold δ.

[0142] In another example, if the change in the value of the parameter to be adjusted is expected to result in the value of the second performance indicator not increasing by more than the deterioration limit threshold δ, for example, δ = 0.1*μ2, then the null hypothesis can be μ1-μ2≤δ, the alternative hypothesis can be μ1-μ2>δ, and the rejection region can be t≥t α(n1+n2-2). Furthermore, if t≥t α (n1+n2-2), then the original hypothesis μ1-μ2≤δ holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the increase in the value of the second performance indicator does not exceed the deterioration limit threshold δ, that is, the hypothesis test passes. If t≥t α (n1+n2-2), then the alternative hypothesis μ1-μ2>δ holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the increase in the value of the first performance indicator exceeds the deterioration limit threshold δ.

[0143] As described above, the first set of values ​​of the second performance indicator includes multiple values ​​of the first performance indicator, then the population of the first set of values ​​of the second performance indicator may be all of the above multiple values, and the sample of the first set of values ​​of the second performance indicator may be a portion of the values ​​selected from the above multiple values. The concepts of the sample and population of the second set of values ​​of the second performance indicator are similar to those of the sample and population of the first set of values ​​of the second performance indicator.

[0144] Exemplarily, the second service and the first service may be the same service or different services. The second service and the first service share resources of the first device, such as CPU memory bandwidth, OS scheduling capability, and the like.

[0145] For example, in an autonomous driving software system, the CPU may share memory bandwidth with an artificial intelligence core (AICore), a GPU, an image signal processor (ISP), etc. When optimizing the CPU memory bandwidth, it may cause degradation of other memory bandwidths. For example, when the CPU accesses a certain memory, the memory is occupied by the AIcore, and the CPU usage rate may increase at this time. In order to optimize the CPU memory bandwidth, a certain parameter (such as the MPAM parameter) can be adjusted to increase the priority of the CPU when accessing a certain memory, thereby reducing the CPU usage rate and achieving the purpose of optimizing the CPU usage rate (or CPU memory bandwidth). However, increasing the priority of the CPU when accessing the memory will inevitably affect the priority of the AIcore when accessing the memory, for example, when the AIcore accesses the memory, the memory is occupied by the CPU, which in turn causes the frame rate of the AIcore to drop, that is, the frame rate of the AIcore (or the memory bandwidth of the AIcore) is degraded. Similarly, when optimizing the CPU usage rate, it may also cause degradation of the GPU frame rate, the ISP frame rate, etc. Therefore, when the CPU usage rate is taken as the first performance indicator, any one of the AICore frame rate, GPU frame rate, and ISP frame rate can be taken as the second performance indicator.

[0146] In some possible implementations, the method further includes: determining whether the variance change of the first performance indicator is within a preset range according to the first set of values ​​of the first performance indicator and the second set of values ​​of the first performance indicator; optimizing the parameter to be adjusted according to the first value, including: optimizing the parameter to be adjusted according to the first value when the variance change of the first performance indicator is within the preset range. Exemplarily, the preset range may be a range smaller than the variance of the second set of values ​​of the first performance indicator, or may also be a range larger than the variance of the second set of values ​​of the first performance indicator, or may also be other ranges.

[0147] Exemplarily, the variance change of the first performance indicator being within the preset range may indicate that the indicator jitter of the first performance indicator is less than or equal to a preset threshold. Specifically, determining that the variance change of the first performance indicator is within the preset range may include: performing a hypothesis test on the variance of the first group of values ​​of the first performance indicator and the variance of the second group of values ​​of the first performance indicator according to the second hypothesis; when the hypothesis test passes, determining that the variance change of the first performance indicator is within the preset range.

[0148] Illustratively, the second hypothesis may be any null hypothesis H0 in Table 2, and a pass in the hypothesis test means that the null hypothesis H0 is established.

[0149] Table 2

[0150]

[0151] For the test statistic n represents the sample size of the first set of values ​​of the first performance indicator, S 2 represents the sample variance of the first set of values ​​of the first performance indicator, σ 2 represents the population variance of the first set of values ​​of the first performance indicator, represents the population variance of the second set of values ​​of the first performance indicator, Representation Statistics The value of , a represents the confidence level, It means that the chi-square distribution table is obtained based on the degree of freedom n-1 and the confidence level a. Wherein, a can be 0.90, or, a can be 0.95, or, a can be other values ​​between 0 and 1. For example, if it is expected that the change of the value of the parameter to be adjusted will not lead to an increase in the indicator jitter of the first performance indicator, the original hypothesis can be taken as The alternative hypothesis is The rejection region is Furthermore, if you do not meet The null hypothesis That is, when the parameter to be adjusted changes from the second value to the first value, the jitter of the first performance indicator does not increase, that is, the hypothesis test passes, and the parameter to be adjusted can be further optimized according to the first value of the parameter to be adjusted. The alternative hypothesis holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the jitter of the first performance indicator increases.

[0152] For the test statistic n1 represents the sample size of the second set of values ​​of the first performance indicator, n2 represents the sample size of the first set of values ​​of the first performance indicator, represents the population variance of the second set of values ​​of the first performance indicator, represents the population variance of the first set of values ​​of the first performance indicator, F represents the value of the statistic F, a represents the confidence level, and F a (n1-1, n2-1) represents the F value obtained by looking up the F distribution table according to the degrees of freedom n1-1, the degrees of freedom n2-1 and the confidence a. Wherein, a can be 0.90, or, a can also be 0.95, or, a can also be other values ​​between 0 and 1. For example, if it is expected that the change in the value of the parameter to be adjusted will not lead to an increase in the indicator jitter of the first performance indicator, the original hypothesis can be taken as The alternative hypothesis is The rejection region is F ≥ F a (n1-1,n2-1). Furthermore, if F≥F a (n1-1,n2-1), then the original hypothesis If F≥F a (n1-1,n2-1), then the alternative hypothesis holds true, that is, when the parameter to be adjusted changes from the second value to the first value, the jitter of the first performance indicator increases.

[0153] In some possible implementations, the parameter to be adjusted is optimized according to the first numerical value when at least two of the following conditions a to c are met: a. when the parameter to be adjusted changes from the second numerical value to the first numerical value, the numerical change of the first performance indicator is not caused by the jitter of the first performance indicator; b. the numerical change of the second performance indicator does not exceed the deterioration limit threshold (such as δ in Table 1); c. the numerical change of the first performance indicator does not exceed the deterioration limit threshold (such as δ in Table 1).

[0154] Exemplarily, if it is set that when two of the above a to c are satisfied, the parameter to be adjusted is optimized according to the first value, then when any of the above two items is not satisfied, the value of the parameter to be adjusted is re-acquired. For example, when a and b are satisfied, the parameter to be adjusted is optimized according to the first value, then when any of a and b is not satisfied, the value of the parameter to be adjusted is re-acquired, and so on.

[0155] Exemplarily, if it is set that when all three items a to c are satisfied, the parameter to be adjusted is optimized according to the first value, then when any item a to c is not satisfied, the value of the parameter to be adjusted is re-acquired.

[0156] In some possible implementations, optimizing the parameter to be adjusted according to the first value includes: adjusting the first value to a third value of the parameter to be adjusted so that the operation result of the first device meets the set condition. For example, if the parameter to be adjusted is an MPAM parameter, then the operation result of the first device meets the set condition, which may include: the CPU usage of the first service meets the set condition, such as reaching a set CPU usage range.

[0157] In some possible implementations, when controlling the first optimization algorithm to optimize the parameter to be adjusted based on the first numerical value, an optimization log is recorded, and the optimization log includes at least one of the following items: the reason why the first numerical value of the parameter to be adjusted is modified, the numerical value determined for the parameter to be adjusted that exceeds the set range, the numerical value of the parameter to be adjusted when the first performance indicator fails the hypothesis test, and the numerical value of the parameter to be adjusted when the first performance indicator passes the hypothesis test but the second performance indicator fails the hypothesis test.

[0158] The first performance indicator failing the hypothesis test may include at least one of the following: the change of the first performance indicator is caused by indicator jitter; or, when the first performance indicator is required to be optimized, the degradation degree of the first performance indicator is high; or, the jitter of the first performance indicator exceeds the set range. The second performance indicator failing the hypothesis test may include: the degradation degree of the second performance indicator exceeds the deterioration limit threshold.

[0159] Exemplarily, the optimization log may also include other entries in the optimization-related log information in the above embodiment.

[0160] The parameter optimization method provided in the embodiment of the present application can determine whether the change in the value of the performance indicator during the parameter optimization process is caused by indicator jitter, and whether the value of the parameter to be adjusted can be used for the next algorithm iteration by hypothesis testing, which helps to reduce the interference of jitter noise on the algorithm optimization process and improve the convergence ability of the algorithm; and, through hypothesis testing, optimization with the goal of reducing indicator jitter can be completed. Through multi-objective optimization, while optimizing the first performance indicator, it is ensured that the degradation of the second performance indicator is limited to a certain range, which helps the optimization process meet the optimization requirements of multiple performance indicators. In addition, during the parameter optimization process, recording the abnormal values ​​of the parameters to be adjusted and / or the causes of the abnormalities helps to improve the problem location capability during the optimization process.

[0161] Combined with the above Figures 1 to 3 The parameter optimization method provided in the embodiment of the present application is described in detail. Figure 4 and Figure 5 The parameter optimization device provided in the embodiment of the present application is described in detail. It should be understood that the description of the parameter optimization device embodiment corresponds to the description of the method embodiment, so the contents not described in detail can be referred to the method embodiment above, and will not be repeated here for the sake of brevity.

[0162] Figure 4 A schematic block diagram of a parameter optimization device 400 provided in an embodiment of the present application is shown. The device 400 may include a Figure 3 Furthermore, each unit in the device 400 is respectively for implementing Figure 3 The corresponding process of the method embodiment in FIG.

[0163] Specifically, the device 400 includes a first acquisition unit 410 and a processing unit 420, wherein the first acquisition unit 410 is used to: obtain a first set of values ​​of a first performance indicator, the first set of values ​​of the first performance indicator is generated when the first service is running within a first time period based on the first value of the parameter to be adjusted, the first performance indicator is associated with the performance of the first service, and the first service is running on a first device; the processing unit 420 is used to: determine, based on the first set of values ​​of the first performance indicator and the second set of values ​​of the first performance indicator, whether the value change of the first performance indicator is caused by the parameter to be adjusted changing from the second value to the first value, the second set of values ​​of the first performance indicator is generated when the first service is running within a second time period based on the second value of the parameter to be adjusted; and optimize the parameter to be adjusted according to the first value.

[0164] Optionally, the first acquisition unit 410 is also used to: obtain a first set of values ​​of a second performance indicator within a first time period, the second performance indicator is associated with the performance of a second service, and the second service runs on the first device; the processing unit 420 is also used to: determine, based on the second set of values ​​of the second performance indicator and the first set of values ​​of the second performance indicator, a value change of the second performance indicator when the parameter to be adjusted changes from the second value to the first value, the second set of values ​​of the second performance indicator being generated when the second service runs within the second time period based on the second value of the parameter to be adjusted; when the value change of the second performance indicator does not exceed the deterioration limit threshold, the parameter to be adjusted is optimized according to the first value.

[0165] Optionally, processing unit 420 is used to: perform a hypothesis test on the average value of a first group of values ​​of the first performance indicator and the average value of a second group of values ​​of the first performance indicator according to a first hypothesis; when the hypothesis test passes, determine that the change in the value of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value.

[0166] Optionally, the processing unit 420 is also used to: determine whether the variance change of the first performance indicator is within a preset range based on the first set of values ​​of the first performance indicator and the second set of values ​​of the first performance indicator; when the variance change of the first performance indicator is within the preset range, optimize the parameter to be adjusted according to the first value.

[0167] Optionally, the processing unit 420 is also used to: perform a hypothesis test on the variance of the first group of values ​​of the first performance indicator and the variance of the second group of values ​​of the first performance indicator according to the second hypothesis; when the hypothesis test passes, determine that the variance change of the first performance indicator is within a preset range.

[0168] Optionally, the apparatus 400 further includes a second acquisition unit, configured to reacquire the value of the parameter to be adjusted when the value change of the first performance indicator is not caused by the parameter to be adjusted changing from the second value to the first value.

[0169] Optionally, the processing unit 420 is used to: adjust the first value to a third value of the parameter to be adjusted, so that the operation result of the first device meets the set condition.

[0170] Optionally, the device 400 also includes a recording unit, which is used to: when the parameter to be adjusted is optimized according to the first numerical value, record an optimization log, and the optimization log includes at least one of the following items: the reason why the first numerical value of the parameter to be adjusted is modified, the numerical value determined for the parameter to be adjusted that exceeds the set range, the numerical value of the parameter to be adjusted when the first performance indicator fails the hypothesis test, and the numerical value of the parameter to be adjusted when the first performance indicator passes the hypothesis test but the second performance indicator fails the hypothesis test.

[0171] Exemplarily, the device 400 may be arranged at Figure 1In the automatic driving system 100 shown, or it can also be set in Figure 2 For example, the first acquisition unit 410 and the processing unit 420 can be arranged in Figure 2 In the inspection module 230 shown in FIG. 1 , the second acquisition unit may be set Figure 2 In the optimization algorithm module 240 shown, the recording unit may be provided in a module connected to the inspection module 230 and the optimization algorithm 240. The operations performed by the first acquisition unit 410 and the processing unit 420 may be performed by one processor, or may be performed by different processors. In a specific implementation process, the one or more processors may be processors provided in an electronic device; or, the device 400 may be a chip provided in an electronic device. The electronic device may be a computing platform of an intelligent driving device (such as a vehicle); or, the electronic device may also be a computer including a virtual autonomous driving system or an autonomous driving system simulator.

[0172] In the embodiment of the present application, the processor is a circuit with signal processing capability. In one implementation, the processor can be a circuit with instruction reading and running capability, such as a CPU, a microprocessor, a GPU (which can be understood as a microprocessor), or a digital signal processor (digital signal processor, DSP), etc.; in another implementation, the processor can realize certain functions through the logical relationship of the hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable, such as the processor is a hardware circuit implemented by a dedicated integrated circuit (application-specific integrated circuit, ASIC) or a programmable logic device (programmable logic device, PLD), such as a field programmable gate array (field programmable gate array, FPGA). In a reconfigurable hardware circuit, the processor loads a configuration document to implement the process of hardware circuit configuration, which can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as a neural network processing unit (neural network processing unit, NPU), a tensor processing unit (tensor processing unit, TPU), a deep learning processing unit (deep learning processing unit, DPU), etc.

[0173] In the specific implementation process, the various units in the above device can be fully or partially integrated together, or can also be implemented independently. In one implementation, these units are integrated together and implemented in the form of a system-on-a-chip (SoC).

[0174] Figure 5 It is a schematic block diagram of a parameter optimization device provided in an embodiment of the present application. Figure 5 The parameter optimization device 500 shown may include: a processor 510, a transceiver 520, and a memory 530. The processor 510, the transceiver 520, and the memory 530 are connected via an internal connection path, the memory 530 is used to store instructions, and the processor 510 is used to execute the instructions stored in the memory 530 to implement the methods in the above embodiments. Optionally, the memory 530 may be coupled to the processor 510 via an interface, or may be integrated with the processor 510.

[0175] It should be noted that the transceiver 520 may include but is not limited to a transceiver device such as an input / output interface to achieve communication between the device 500 and other devices or a communication network.

[0176] The memory 530 may be a read-only memory (ROM), a static storage device, a dynamic storage device or a random access memory (RAM).

[0177] The transceiver 520 uses a transceiver device such as but not limited to a transceiver to implement communication between the apparatus 500 and other devices or a communication network to receive / send data / information used to implement the methods in the above embodiments.

[0178] An embodiment of the present application further provides an electronic device, which includes the above parameter optimization device 400 or the above parameter optimization device 500.

[0179] An embodiment of the present application further provides a computer program product, which includes a computer program code. When the computer program code runs on a computer, the computer implements the methods in the above embodiments of the present application.

[0180] An embodiment of the present application further provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the computer implements the methods in the above embodiments of the present application.

[0181] An embodiment of the present application also provides a chip, including a circuit, for executing the methods in the above embodiments of the present application.

[0182] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0183] In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in this article is a kind of association relationship that describes associated objects, indicating that three relationships can exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In this application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0184] The prefixes such as "first" and "second" used in the embodiments of the present application are only used to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers used to distinguish description objects in the embodiments of the present application does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the claims or embodiments, and the use of such prefixes should not constitute an unnecessary limitation.

[0185] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0186] In the various embodiments of the present application, unless otherwise specified or logically conflicting, the terms and / or descriptions between the various embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0187] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0188] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0189] The above is only 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 who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A parameter optimization method, characterized in that: include: Obtaining a first set of values ​​of a first performance indicator, where the first set of values ​​of the first performance indicator is generated when a first service is running in a first time period based on a first value of a parameter to be adjusted, the first performance indicator is associated with performance of the first service, and the first service is running on a first device; Determining, according to a first group of values ​​of the first performance indicator and a second group of values ​​of the first performance indicator, that a value change of the first performance indicator is caused by a change of the parameter to be adjusted from a second value to the first value, the second group of values ​​of the first performance indicator being generated when the first service is running in a second time period based on the second value of the parameter to be adjusted; The parameter to be adjusted is optimized according to the first value.

2. The method according to claim 1, characterized in that The method further comprises: Acquire a first set of values ​​of a second performance indicator within the first time period, where the second performance indicator is associated with performance of a second service, and the second service runs on the first device; Determine, according to a second group of values ​​of the second performance indicator and a first group of values ​​of the second performance indicator, a value change of the second performance indicator when the parameter to be adjusted changes from the second value to the first value, the second group of values ​​of the second performance indicator being generated when the second service runs in the second time period based on the second value of the parameter to be adjusted; The optimizing the parameter to be adjusted according to the first value includes: When the value change of the second performance indicator does not exceed the deterioration limit threshold, the parameter to be adjusted is optimized according to the first value.

3. The method according to claim 1 or 2, characterized in that: The step of determining that the change in the value of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value includes: Performing a hypothesis test on an average value of a first group of values ​​of the first performance indicator and an average value of a second group of values ​​of the first performance indicator according to a first hypothesis; When the hypothesis test passes, it is determined that the value change of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value.

4. The method according to any one of claims 1 to 3, characterized in that The method further comprises: Determining whether a variance change of the first performance indicator is within a preset range according to a first group of values ​​of the first performance indicator and a second group of values ​​of the first performance indicator; The optimizing the parameter to be adjusted according to the first value includes: When the variance change of the first performance indicator is within the preset range, the parameter to be adjusted is optimized according to the first value.

5. The method according to claim 4, characterized in that The method further comprises: Performing a hypothesis test on the variance of the first set of values ​​of the first performance indicator and the variance of the second set of values ​​of the first performance indicator according to a second hypothesis; When the hypothesis test passes, it is determined that the variance change of the first performance indicator is within the preset range.

6. The method according to any one of claims 1 to 5, characterized in that The method further comprises: When the value change of the first performance indicator is not caused by the change of the parameter to be adjusted from the second value to the first value, the value of the parameter to be adjusted is reacquired.

7. The method according to any one of claims 1 to 6, characterized in that The optimizing the parameter to be adjusted according to the first value includes: The first value is adjusted to a third value of the parameter to be adjusted so that the operation result of the first device meets the set condition.

8. The method according to any one of claims 1 to 7, characterized in that The method further comprises: When optimizing the parameter to be adjusted according to the first value, recording an optimization log, the optimization log including at least one of the following: the reason why the first value of the parameter to be adjusted is modified, the value determined for the parameter to be adjusted that exceeds the set range, the value of the parameter to be adjusted when the first performance indicator fails the hypothesis test, and the value of the parameter to be adjusted when the first performance indicator passes the hypothesis test but the second performance indicator fails the hypothesis test; The first business is related to autonomous driving.

9. A parameter optimization device, characterized in that: It includes a first acquisition unit and a processing unit, wherein: The first acquisition unit is used to: acquire a first set of values ​​of a first performance indicator, where the first set of values ​​of the first performance indicator is generated when the first service runs within a first time period based on the first value of the parameter to be adjusted, the first performance indicator is associated with the performance of the first service, and the first service runs on a first device; The processing unit is used to: determine, based on a first group of values ​​of the first performance indicator and a second group of values ​​of the first performance indicator, whether the value change of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value, wherein the second group of values ​​of the first performance indicator is generated when the first service runs in a second time period based on the second value of the parameter to be adjusted; and optimize the parameter to be adjusted according to the first value.

10. The device according to claim 9, characterized in that The first acquisition unit is further used for: Acquire a first group of values ​​of a second performance indicator within the first time period, where the second performance indicator is associated with performance of a second service, where the second service runs on the first device, and where the first service and / or the second service are related to autonomous driving; The processing unit is also used for: Determine, according to a second group of values ​​of the second performance indicator and a first group of values ​​of the second performance indicator, a value change of the second performance indicator when the parameter to be adjusted changes from the second value to the first value, the second group of values ​​of the second performance indicator being generated when the second service runs in the second time period based on the second value of the parameter to be adjusted; When the value change of the second performance indicator does not exceed the deterioration limit threshold, the parameter to be adjusted is optimized according to the first value.

11. The device according to claim 9 or 10, characterized in that The processing unit is used for: Performing a hypothesis test on an average value of a first group of values ​​of the first performance indicator and an average value of a second group of values ​​of the first performance indicator according to a first hypothesis; When the hypothesis test passes, it is determined that the value change of the first performance indicator is caused by the change of the parameter to be adjusted from the second value to the first value.

12. The device according to any one of claims 9 to 11, characterized in that The processing unit is also used for: Determining whether a variance change of the first performance indicator is within a preset range according to a first group of values ​​of the first performance indicator and a second group of values ​​of the first performance indicator; When the variance change of the first performance indicator is within the preset range, the parameter to be adjusted is optimized according to the first value.

13. The device according to claim 12, characterized in that The processing unit is also used for: Performing a hypothesis test on the variance of the first set of values ​​of the first performance indicator and the variance of the second set of values ​​of the first performance indicator according to a second hypothesis; When the hypothesis test passes, it is determined that the variance change of the first performance indicator is within the preset range.

14. The device according to any one of claims 9 to 13, characterized in that The device further includes a second acquiring unit, configured to: When the value change of the first performance indicator is not caused by the change of the parameter to be adjusted from the second value to the first value, the value of the parameter to be adjusted is reacquired.

15. The device according to any one of claims 9 to 14, characterized in that The processing unit is used for: The first value is adjusted to a third value of the parameter to be adjusted so that the operation result of the first device meets the set condition.

16. The device according to any one of claims 9 to 15, characterized in that The device also includes a recording unit, which is used to: When optimizing the parameter to be adjusted according to the first value, recording an optimization log, the optimization log including at least one of the following: the reason why the first value of the parameter to be adjusted is modified, the value determined for the parameter to be adjusted that exceeds the set range, the value of the parameter to be adjusted when the first performance indicator fails the hypothesis test, and the value of the parameter to be adjusted when the first performance indicator passes the hypothesis test but the second performance indicator fails the hypothesis test; The first business is related to autonomous driving.

17. A parameter optimization device, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program stored in the memory, so that the control device executes the method according to any one of claims 1 to 8.

18. An electronic device, characterized in that: The electronic device comprises the apparatus according to any one of claims 9 to 17.

19. A computer-readable storage medium, characterized in that: Instructions are stored thereon, and when the instructions are executed by a processor, the processor implements the method according to any one of claims 1 to 8.

20. A chip, characterized in that: The chip comprises a circuit for executing the method according to any one of claims 1 to 8.

Images