A method and apparatus for evaluating the realism of raindrop distribution in simulated rainfall at a site

Abstract

This invention relates to a method and apparatus for assessing the realism of raindrop distribution in simulated rainfall at a site. The method includes: collecting microscopic data of real rainfall at various levels; calculating average raindrop distribution matrices for different levels based on the microscopic data of real rainfall, using them as an evaluation benchmark; collecting microscopic data of simulated rainfall at the site and calculating the raindrop distribution matrix of the simulated rainfall; calculating the similarity between the raindrop distribution matrix of the simulated rainfall and the average raindrop distribution matrices for each level, thereby determining the realism assessment result of the simulated rainfall at the site. Compared with the prior art, this invention starts from the level of the microscopic raindrop distribution matrix of rainfall, constructs an evaluation benchmark based on the raindrop distribution matrix of real rainfall samples, and achieves a quantitative assessment of the realism of raindrop distribution in simulated rainfall by quantifying the similarity between simulated rainfall samples and the evaluation benchmark.

Description

Technical Field

[0001] This invention relates to the field of simulated rainfall testing technology in enclosed sites, and in particular to a method and apparatus for evaluating the accuracy of raindrop distribution in simulated rainfall in enclosed sites. Background Technology

[0002] With the deepening development of autonomous driving technology, the safety and reliability of automobiles will face new and more severe challenges. Unlike traditional vehicles, the testing focus of autonomous vehicles is on the vehicle's ability to perceive and respond to the traffic environment. It is a test of the coupled system of people, vehicles, roads, and environment, and places greater emphasis on the construction of test scenarios. In particular, testing under special weather conditions such as rain, fog, and sunlight is an indispensable part of the research and development of autonomous driving technology.

[0003] Environmental perception systems, as key modules for autonomous vehicles to acquire information about their external environment, are easily affected by complex environmental triggering conditions, making them a major source of anticipated functional safety issues. Rainfall is a high-risk triggering condition. Therefore, it is necessary to explore the impact of rainfall on environmental perception systems. Existing technologies mostly utilize real-road testing, closed-site testing, and simulation testing. Among these, simulated rainfall testing in closed environments is currently the primary research method due to its controllability, repeatability, and suitability for sensor testing. Compared to autonomous driving weather scenario testing in natural environments, building simulated weather scenarios in closed environments allows for precise control of environmental parameters such as rainfall, fog, visibility, and illuminance. This makes the test background environment more accurate and reproducible, helping developers better understand the performance and responsiveness of autonomous driving systems under different conditions. This, in turn, allows for optimization of system design and control strategies, promoting the application and widespread adoption of autonomous driving technology.

[0004] Existing research explores the impact of rainfall triggering conditions on sensors through simulated rainfall in a field. Most analyses of these simulated rainfall tests are based on the settings of the simulated rainfall environment. However, there are many differences between simulated rainfall and real weather, such as fogging. Current methods for evaluating simulated rainfall equipment only consider single indicators such as rainfall intensity or falling velocity. Referring to relevant meteorological research, rainfall itself should be compared at the microscopic distribution level within space. Differences at the microscopic level of rainfall can lead to biases in experimental results, potentially resulting in low confidence levels in assessing the expected functional safety of the sensing system.

[0005] Therefore, it is necessary to assess the realism of raindrop distribution in simulated rainfall at the micro level in order to quantify the realism of simulated rainfall and ensure the confidence level of test results for autonomous vehicles. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art by providing a method and apparatus for evaluating the accuracy of raindrop distribution in simulated rainfall at a site, which can achieve a quantitative evaluation of the accuracy of raindrop distribution in simulated rainfall.

[0007] The objective of this invention can be achieved through the following technical solution: a method for evaluating the accuracy of raindrop distribution in site-based simulated rainfall, comprising the following steps:

[0008] S1. Collect microscopic data on actual rainfall at all levels;

[0009] S2. Based on real rainfall microdata, calculate the average raindrop distribution matrix for different levels as an evaluation benchmark;

[0010] S3. Collect microscopic data of simulated rainfall at the site and calculate the raindrop distribution matrix of the simulated rainfall;

[0011] S4. Calculate the similarity between the simulated rainfall drop distribution matrix and the average rainfall drop distribution matrix at each level;

[0012] S5. Based on the similarity calculated in step S4, determine the accuracy assessment result of the simulated rainfall at the site.

[0013] Furthermore, the microscopic data of real rainfall in step S1 and the microscopic data of simulated rainfall in step S3 both include the number of raindrop particles with different diameters and velocities, as well as rainfall intensity data.

[0014] Furthermore, the specific process of step S2 is as follows: the micro data of real rainfall at all levels are sliced ​​into different samples according to minutes, each sample contains a corresponding diameter velocity-particle number distribution matrix, and the corresponding level is determined based on the rainfall intensity data within one minute;

[0015] The average distribution is obtained by superimposing diameter velocity-particle number distribution matrices of the same level, and the average distribution heatmap of real rainfall at different levels is used as the benchmark for authenticity evaluation. The benchmark is iteratively updated as real rainfall samples are added.

[0016] Further, step S4 includes the following steps:

[0017] S41. Based on the formula of Structural Similarity (SSIM), calculate the basic similarity between the raindrop distribution matrix of the simulated rainfall and the average raindrop distribution matrix of each level. The similarity result is a value from 0 to 1. The calculation formula is as follows:

[0018]

[0019] Where, μ x μ is the mean of the elements of matrix x;y Let y be the mean of the elements of matrix y; x σ is the variance of the elements of matrix x; y σ is the variance of the elements of matrix y; xy C1 and C2 are the covariances of matrices x and y; C1 and C2 are constants to avoid the case where the denominator is 0.

[0020] S42. For pixels with abnormal distributions and values ​​in the simulated rainfall droplet distribution matrix, set a penalty term y. dist With y val This is used to distinguish simulated rainfall samples with significant differences in raindrop distribution. The calculation formula is as follows:

[0021]

[0022] Where X is the simulated rainfall droplet distribution matrix, Y is the baseline rainfall droplet distribution matrix, and P... dist The statistical simulation of rainfall samples includes the number of outlier elements whose values ​​are non-zero but whose corresponding positions in the baseline distribution matrix are zero. P represents the number of such outlier elements. val Then count the number of outlier elements in the sample to be tested whose values ​​differ by magnitude from the corresponding elements in the benchmark matrix;

[0023] S43, Based on two penalty terms P dist With P val The total penalty coefficient P is obtained. total Multiply the total penalty coefficient by the basic similarity calculation result to obtain the total similarity S between the raindrop distribution matrix of the simulated rainfall and the average raindrop distribution matrix of each level.

[0024] Furthermore, the formula for calculating the total similarity S in step S43 is as follows:

[0025] S(X,Y)=P total ·SSIM(X,Y)

[0026]

[0027] Where, N X This represents the number of non-zero pixels in the raindrop distribution matrix used to simulate rainfall.

[0028] Furthermore, step S5 specifically includes the following steps:

[0029] S51. Based on the simulated rainfall intensity data, a fuzzy membership degree for the rainfall level is set, thereby defining the boundary for the fuzzy rainfall level. The fuzzy membership degree is calculated as follows:

[0030]

[0031]

[0032] Where, μ little μ mid μ heavy These represent the fuzzy membership degrees for light, moderate, and heavy rain, respectively, and R is the simulated rainfall intensity value.

[0033] S52. Using fuzzy membership degree as weight, the similarity of raindrop distribution between the simulated rainfall and the benchmarks at all levels is calculated by weighted summation, and the realism of raindrop distribution (RRD) is obtained by comprehensive calculation.

[0034] S53. Based on the RRD assessment results of simulated rainfall, the site's simulated rainfall meets the requirements for the realism of raindrop distribution by comparing the RRD index calculation results of simulated rainfall with approximate rainfall intensity, or by comparing with the realism threshold.

[0035] Furthermore, the formula for calculating the raindrop distribution accuracy (RRD) in step S52 is as follows:

[0036] RRD = μ little *S little +μ mid *S mid +μ heavy *S heavy

[0037] Among them, S little S mid S heavy These represent the similarity of raindrop distribution between simulated rainfall and the three levels of realism assessment benchmarks (small, medium, and large), respectively.

[0038] Furthermore, in step S53, the authenticity threshold is determined based on the numerical distribution of similarity between the real rainfall sample and the distribution benchmark, and the lower limit of the 95% confidence interval based on the similarity distribution result of the real rainfall sample is used as the reference value of the authenticity threshold.

[0039] A device for assessing the realism of raindrop distribution in simulated rainfall at a site includes an overall frame consisting of a base, rollers, and a column. The rollers are installed at the bottom of the base, and a battery box is installed at the top of the base. An industrial control computer, a display screen, and a laser raindrop spectrometer are installed on the column. The laser raindrop spectrometer is located at the top of the column. The laser raindrop spectrometer is used to collect microscopic data of real and simulated rainfall and transmit them to the industrial control computer. The industrial control computer is used to execute a preset program to calculate the realism assessment results of the simulated rainfall at the site and transmit the results to the display screen for display.

[0040] Furthermore, the preset program in the industrial control computer includes a first program and a second program with switchable modes. The first program processes the micro data of real rainfall, uses real rainfall as the basis for evaluation, classifies real rainfall samples into levels based on rainfall intensity, and iteratively updates the average raindrop distribution matrix of different levels.

[0041] The second program processes the microscopic data of simulated rainfall, takes the simulated rainfall at the site as the object of evaluation, calculates the similarity between the raindrop distribution matrix of the simulated rainfall and the average raindrop distribution matrix of each level, and determines the authenticity evaluation result of the simulated rainfall at the site.

[0042] Compared with the prior art, the present invention has the following advantages:

[0043] This invention first collects microscopic data of real rainfall at various levels to calculate the average raindrop distribution matrix for different levels, serving as an evaluation benchmark. Then, it collects microscopic data of simulated rainfall at a site and calculates the raindrop distribution matrix of the simulated rainfall. Finally, by calculating the similarity between the simulated rainfall's raindrop distribution matrix and the average raindrop distribution matrices for each level, the realism assessment result of the simulated rainfall at the site is determined. This allows for the evaluation of the realism of raindrop distribution in simulated rainfall at a microscopic level, quantifying the realism of the simulated rainfall and ensuring the confidence level of the test results for autonomous vehicles.

[0044] In this invention, the microscopic data of rainfall includes the number of raindrop particles of different diameters and velocities, as well as rainfall intensity data. The microscopic data is sliced ​​into different samples by minute, and each sample contains a corresponding diameter-velocity-particle number distribution matrix. Based on the rainfall intensity value within one minute, a level is defined. For the microscopic data of real rainfall, the raindrop distribution matrices of rainfall intensity data within the same level are superimposed to obtain the average distribution. The average raindrop distribution matrix of real rainfall at different levels is used as the accuracy evaluation benchmark, and this benchmark is iteratively updated as real rainfall samples are added. On the one hand, since the diameter-velocity-particle number distribution matrix contains most of the information on the microscopic distribution within the rainfall space, the evaluation results can reflect the accuracy of the simulated rainfall at the microscopic level. On the other hand, using the average raindrop distribution matrix of each level of real rainfall as the evaluation benchmark can avoid the randomness of individual rainfall samples, obtain the common characteristics of the microscopic distribution of specific levels of rainfall in space, and thus enhance reliability.

[0045] This invention proposes a device for evaluating the accuracy of raindrop distribution in simulated rainfall at a site. The device uses a laser raindrop spectrometer to collect microscopic data of both real and simulated rainfall. An industrial control computer receives the microscopic rainfall data output from the laser raindrop spectrometer and executes a preset program. It then uses a raindrop distribution accuracy evaluation method to assess the accuracy of the simulated rainfall at the site. The preset program can switch between different modes: for real rainfall samples, iteratively updating the evaluation benchmark; for simulated rainfall samples, it performs accuracy evaluation based on a stored evaluation benchmark, ensuring real-time updating, storage, and output of rainfall data. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0047] Figure 2 This is a heatmap showing the distribution of some real rainfall samples in the example.

[0048] Figure 3 The rainfall level classification standard is used in the example;

[0049] Figure 4 A three-level average baseline distribution heatmap was constructed for this embodiment;

[0050] Figure 5 This is a heat map showing the distribution of simulated rainfall samples at some sites in the example.

[0051] Figure 6 The distribution of evaluation results for real rainfall samples in the examples;

[0052] Figure 7 This is a case study for evaluating the realism of simulated rainfall samples in the embodiments;

[0053] Figure 8 This is a schematic diagram of the overall structure of the device of the present invention;

[0054] The markings in the diagram are as follows: 11. Base, 12. Roller, 13. Column, 2. Battery box, 3. Industrial computer, 4. Display screen, 5. Laser raindrop spectrometer. Detailed Implementation

[0055] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0056] Example

[0057] like Figure 1 As shown, a method for evaluating the realism of raindrop distribution in site-based simulated rainfall includes the following steps:

[0058] S1. Collect microscopic data on actual rainfall at all levels;

[0059] S2. Based on real rainfall microdata, calculate the average raindrop distribution matrix for different levels as an evaluation benchmark;

[0060] S3. Collect microscopic data of simulated rainfall at the site and calculate the raindrop distribution matrix of the simulated rainfall;

[0061] S4. Calculate the similarity between the simulated rainfall drop distribution matrix and the average rainfall drop distribution matrix at each level;

[0062] S5. Based on the similarity calculated in step S4, determine the accuracy assessment result of the simulated rainfall at the site.

[0063] This embodiment applies the above-described solution, and its main contents include:

[0064] I. Use equipment to collect microscopic data on actual rainfall at all levels;

[0065] In this embodiment, microscopic data samples of real rainfall were collected (the collected microscopic data includes the number of raindrop particles of different diameters and velocities, as well as rainfall intensity data). The rainfall microscopic data was divided into different samples by minute slices, and the intensity was classified based on the corresponding rainfall intensity value within one minute. A total of 712 sets of real rainfall samples were collected, including 391 sets of light rain samples, 141 sets of moderate rain samples, and 180 sets of heavy rain samples. A heatmap of some sample distributions is shown below. Figure 2 As shown, the horizontal axis represents different particle size levels, the vertical axis represents different velocity levels, and the pixel represents the number of raindrop particles of the corresponding particle size and velocity.

[0066] Second, based on real rainfall micro data, calculate the average distribution heat map of different levels (i.e. the average raindrop distribution matrix of different levels) as the evaluation benchmark. Specifically, the average distribution is obtained by superimposing the raindrop distribution matrices of rainfall intensity samples within the same level, and the average distribution heat map of real rainfall at different levels is used as the accuracy evaluation benchmark. The benchmark is iteratively updated as real rainfall samples are added.

[0067] In this embodiment, real rainfall samples are divided into three levels according to rainfall intensity: light, medium, and heavy. The classification criteria are as follows: Figure 3 As shown, the average distribution heatmap calculated from the distribution heatmaps of samples within the same rainfall level serves as the benchmark distribution for subsequent accuracy assessment. The three-level average distribution heatmaps obtained based on the aggregated actual rainfall samples are shown below. Figure 4 As shown.

[0068] III. Collecting microscopic data of simulated rainfall at the site;

[0069] In this embodiment, microscopic data samples of simulated rainfall were collected at the site, resulting in 34 sets of simulated rainfall samples. A heat map showing the distribution of some of these simulated rainfall samples is shown below. Figure 5 As shown.

[0070] Fourth, the similarity between the simulated rainfall distribution heat map of the site and the baseline distribution heat maps of different levels is calculated to determine the accuracy assessment results;

[0071] In this embodiment, the similarity between the simulated rainfall sample and the evaluation benchmark distribution heat map is calculated, and the result is a value from 0 to 1, so as to facilitate intuitive analysis and comparison. The value can directly reflect the distinguishability, and the closer it is to 1, the more realistic the simulated rainfall is.

[0072] V. Based on the realism assessment results, determine whether the simulated rainfall at the site meets the requirements for the realism of raindrop distribution.

[0073] The specific process for calculating the similarity between the simulated rainfall samples at the site and the assessment baseline distribution heatmap includes:

[0074] First, based on the Structural Similarity in Context (SSIM) formula, the basic similarity between the simulated rainfall samples at the site and the raindrop distribution matrices of each assessment benchmark is calculated. The similarity result is a value between 0 and 1, and the calculation formula is as follows:

[0075]

[0076] Where: μ x μ is the mean of the elements of matrix x; y σ is the mean of the elements of matrix y; x σ is the variance of the elements of matrix x; y σ is the variance of the elements of matrix y; xy C1 and C2 are the covariances of matrices x and y; C1 and C2 are constants to avoid the case where the denominator is 0.

[0077] Subsequently, a penalty term P is set for pixels with abnormal distributions and values ​​in the simulated rainfall sample distribution heatmap. dist With P val This is used to distinguish simulated rainfall samples with significant differences in raindrop distribution. The calculation formula is as follows:

[0078]

[0079] Where: X is the raindrop distribution matrix of the simulated rainfall sample to be evaluated, Y is the baseline raindrop distribution matrix, and P... dist The statistical simulation of rainfall samples includes the number of outlier elements whose values ​​are non-zero but whose corresponding positions in the baseline distribution matrix are zero. P represents the number of such outlier elements. val Then count the number of outlier elements in the sample to be tested whose values ​​differ by magnitude from the corresponding elements in the benchmark matrix.

[0080] Combining the two penalty factors above, the total penalty coefficient P is determined. totalMultiplying the total penalty coefficient by the base similarity calculation result yields the raindrop distribution similarity S between the simulated rainfall and the baseline distribution heatmaps at all levels. The calculation formula is as follows:

[0081]

[0082] S(Y,Y)=P total ·SSIM(Y,Y)

[0083] Where: N X This represents the number of non-zero pixels in the heatmap of the distribution of simulated rainfall samples to be evaluated.

[0084] The specific process for determining the authenticity assessment results includes:

[0085] Based on the rainfall intensity information of the simulated rainfall samples to be evaluated, a fuzzy membership degree for the rainfall level is set, thereby defining the boundary of the fuzzy rainfall level classification.

[0086] Fuzzy membership degree is calculated as follows:

[0087]

[0088] Where: μ little μ mid μ heavy These represent the fuzzy membership degrees for light, moderate, and heavy rain, respectively, and R represents the rainfall intensity value of the rainfall sample to be measured.

[0089] Using fuzzy membership degrees as weights, the raindrop distribution similarity between the sample to be evaluated and each level of benchmark is weighted and summed to obtain the raindrop distribution realism (RRD). The formula for calculating the raindrop distribution realism is as follows:

[0090] RRD = μ little *S little +μ mid *S mid +μ heavy *S heavy

[0091] Wherein: S little S mid S heavy These represent the similarity of raindrop distribution between the sample to be evaluated and the three levels of realism assessment benchmarks: small, medium, and large.

[0092] In this embodiment, the similarity between the actual rainfall sample and the distribution benchmark is calculated using the above method, resulting in the following: Figure 6 The similarity distribution of the real samples is shown.

[0093] The Kolmogorov-Smirnov (KS) test was used to assess the sample similarity distribution. The test results showed that the calculated RRD followed a normal distribution, with a significance value (P) greater than 0.2. Therefore, based on the distribution of actual rainfall similarity, this embodiment defined the lower limit of the 95% confidence interval as the accuracy threshold, with a numerical result of 0.76.

[0094] This embodiment evaluates the realism of raindrop distribution for all simulated rainfall samples, with the sample having the highest realism in raindrop distribution being, for example... Figure 7 As shown, its RRD value is 0.74, which is closest to the actual raindrop distribution. It can be found that the selected high-fidelity samples can match the overall distribution pattern of actual rainfall under the corresponding rainfall intensity, especially in terms of the overall number of raindrops, which is highly consistent with the actual rainfall.

[0095] This embodiment also proposes a device for evaluating the accuracy of raindrop distribution in simulated rainfall at a site, such as... Figure 8 As shown, the device includes a base 11, rollers 12, a column 13, a battery box 2, an industrial control computer 3, a display screen 4, and a laser raindrop spectrometer 5. The base 11, rollers 12, and column 13 together form the frame structure of the device. The battery box 2 is fixedly installed on the base 11. The industrial control computer 3 and the display screen 4 are installed in the central part of the column 13. The laser raindrop spectrometer 5 is installed at the top of the column 13.

[0096] The battery box 2 supplies power to the industrial computer 3, the display screen 4, and the laser raindrop spectrometer 5 via a wiring harness. The battery box 2 is installed at a low position, which can lower the center of gravity of the overall device. At the same time, the casing of the battery box 2 is waterproof to ensure the safety and reliability of the power supply.

[0097] The laser raindrop spectrometer 5 is used to collect microscopic data of real and simulated rainfall, including the number of raindrop particles with different velocities and diameters, as well as the fitted rainfall intensity information. The collected microscopic data is transmitted to the industrial control computer 3, which stores and processes the rainfall data and transmits the processed realism evaluation results to the display screen 4 in real time for result display.

[0098] In practical applications, the industrial control computer 3 has a corresponding program pre-set. After receiving the rainfall sample, the industrial control computer 3 will take different processing methods: real rainfall samples are divided by minute and into different levels based on rainfall intensity, and the average distribution heat map is calculated as the evaluation benchmark; simulated rainfall samples are evaluated by the authenticity evaluation method to obtain the authenticity evaluation result.

[0099] In addition, display screen 4 shows the accuracy assessment results, as well as the heat map information of the collected samples.

[0100] In this embodiment, the industrial control computer 3, the display screen 4, and the laser raindrop spectrometer 5 are connected by a wiring harness to complete real-time data transmission.

[0101] In summary, this scheme starts from the level of the microscopic raindrop distribution matrix of rainfall, constructs an evaluation benchmark based on the raindrop distribution matrix of real rainfall samples, and achieves a quantitative evaluation of the realism of the simulated rainfall raindrop distribution by quantifying the similarity between the simulated rainfall samples and the evaluation benchmark.

Claims

1. A method for evaluating the realism of raindrop distribution in site-based simulated rainfall, characterized in that, Includes the following steps: S1. Collect microscopic data on actual rainfall at all levels; S2. Based on real rainfall microdata, calculate the average raindrop distribution matrix for different levels as an evaluation benchmark; S3. Collect microscopic data of simulated rainfall at the site and calculate the raindrop distribution matrix of the simulated rainfall; S4. Calculate the similarity between the simulated rainfall drop distribution matrix and the average rainfall drop distribution matrix at each level; S5. Based on the similarity calculated in step S4, determine the accuracy assessment result of the simulated rainfall at the site. Step S4 includes the following steps: S41. Based on the Structural Similarity in Memory (SSIM) formula, calculate the basic similarity between the raindrop distribution matrix of the simulated rainfall and the average raindrop distribution matrix of each level. The similarity result is a value from 0 to 1. The calculation formula is as follows: ， in, For matrix The element mean; For matrix The element mean; It is a matrix The element variance; It is a matrix The element variance; It is a matrix and Covariance; , It is a constant to avoid the case where the denominator is 0; S42. Set a penalty term for pixels with abnormal distributions and values ​​in the raindrop distribution matrix of simulated rainfall. and This is used to distinguish simulated rainfall samples with significant differences in raindrop distribution. The number of outlier elements in a statistically simulated rainfall sample whose element values ​​are non-zero but whose corresponding positions in the baseline distribution matrix are zero. Then count the number of outlier elements in the sample to be tested whose values ​​differ by magnitude from the corresponding elements in the benchmark matrix; S43, Based on two penalty terms and The total penalty coefficient is obtained. Multiplying the total penalty coefficient by the base similarity calculation result yields the total similarity between the simulated rainfall droplet distribution matrix and the average rainfall droplet distribution matrix at each level. ; Total similarity in step S43 The calculation formula is: ， ， in, To simulate the raindrop distribution matrix of rainfall, As the baseline raindrop distribution matrix, This represents the number of non-zero pixels in the raindrop distribution matrix used to simulate rainfall.

2. The method for evaluating the realism of raindrop distribution in site-based simulated rainfall according to claim 1, characterized in that, The microscopic data of real rainfall in step S1 and the microscopic data of simulated rainfall in step S3 both include the number of raindrop particles with different diameters and velocities, as well as rainfall intensity data.

3. The method for evaluating the realism of raindrop distribution in site-based simulated rainfall according to claim 2, characterized in that, The specific process of step S2 is as follows: the micro data of real rainfall at all levels are sliced ​​into different samples according to minutes, each sample contains the corresponding diameter velocity-particle number distribution matrix, and the corresponding level is determined based on the rainfall intensity data within one minute. The average distribution is obtained by superimposing diameter velocity-particle number distribution matrices of the same level, and the average distribution heatmap of real rainfall at different levels is used as the benchmark for authenticity evaluation. The benchmark is iteratively updated as real rainfall samples are added.

4. The method for evaluating the realism of raindrop distribution in site-based simulated rainfall according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51. Based on the simulated rainfall intensity data, a fuzzy membership degree for the rainfall level is set, thereby defining the boundary for the fuzzy rainfall level. The fuzzy membership degree is calculated as follows: ， ， ， in, , , These represent the fuzzy membership degrees for light, moderate, and heavy rain, respectively, and R is the simulated rainfall intensity value. S52. Using fuzzy membership degree as weight, the similarity of raindrop distribution between simulated rainfall and various benchmarks is weighted and summed to obtain the raindrop distribution realism RRD. S53. Based on the RRD assessment results of simulated rainfall, the site's simulated rainfall meets the requirements for the realism of raindrop distribution by comparing the RRD index calculation results of simulated rainfall with approximate rainfall intensity, or by comparing with the realism threshold.

5. The method for evaluating the realism of raindrop distribution in site-based simulated rainfall according to claim 4, characterized in that, The formula for calculating the raindrop distribution accuracy (RRD) in step S52 is as follows: ， in, , , These represent the similarity of raindrop distribution between simulated rainfall and the three levels of realism assessment benchmarks (small, medium, and large), respectively.

6. The method for evaluating the realism of raindrop distribution in site-based simulated rainfall according to claim 4, characterized in that, In step S53, the authenticity threshold is determined based on the numerical distribution of similarity between the real rainfall sample and the distribution benchmark, and the lower limit of the 95% confidence interval based on the similarity distribution result of the real rainfall sample is used as the reference value of the authenticity threshold.

7. A device for evaluating the realism of raindrop distribution in simulated rainfall at a site, employing the method for evaluating the realism of raindrop distribution in simulated rainfall at a site as described in claim 1, characterized in that, The system includes an overall frame consisting of a base (11), rollers (12) and a column (13). The rollers (12) are installed at the bottom of the base (11). A battery box (2) is installed on the top of the base (11). An industrial control computer (3), a display screen (4) and a laser rain spectrometer (5) are installed on the column (13). The laser rain spectrometer (5) is located at the top of the column (13). The laser rain spectrometer (5) is used to collect microscopic data of real rainfall and simulated rainfall and transmit them to the industrial control computer (3). The industrial control computer (3) is used to execute a preset program to calculate the realism assessment results of the simulated rainfall on the site and transmit them to the display screen (4) for result display.

8. The raindrop distribution accuracy assessment device for site-based simulated rainfall according to claim 7, characterized in that, The preset program in the industrial control computer (3) includes a first program and a second program with switchable modes. The first program processes the micro data of real rainfall, uses real rainfall as the basis for evaluation, divides the real rainfall sample levels based on rainfall intensity, and iteratively updates the average raindrop distribution matrix of different levels. The second program processes the microscopic data of simulated rainfall, takes the simulated rainfall at the site as the object of evaluation, calculates the similarity between the raindrop distribution matrix of the simulated rainfall and the average raindrop distribution matrix of each level, and determines the authenticity evaluation result of the simulated rainfall at the site.

Images