Uncertainty Analysis Method for Estimation Model of Marine Environment Design Parameters
By calculating the number of observed samples and degree of discreteness of marine environmental design parameters, the uncertainty of model itself and parameter estimation, combined with information entropy and Monte Carlo method, the problem of uncertainty analysis in the calculation of marine environmental design parameters is solved, and the problem of uncertainty analysis in the calculation of marine environmental design parameters is achieved. The overall uncertainty evaluation and factor analysis of multiple models provide the basis for model optimization and selection.
Patent Information
- Application Number
- CN202210124325.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2022-01-29
- Filing Date
- 2022-02-10
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-02-10
AI Technical Summary
In marine engineering design, the uncertainty of marine environmental design parameters is difficult to accurately calculate, especially in the case of small samples. Existing methods are difficult to distinguish the uncertainty caused by samples and other factors, and it is difficult to calculate the overall uncertainty of multivariate models. Spend.
By determining the uncertainty of the number of observed samples and the degree of discreteness, the uncertainty of the model itself and the parameter estimates are calculated, and the overall uncertainty is calculated by summing them to calculate the overall uncertainty, and the information entropy theory and Monte Carlo method are used to separate random and cognitive non-existence Determinism, and then analyze the overall uncertainty of the multivariate model.
The uncertainty analysis of marine environmental design parameters estimation model in small samples is realized, which can evaluate the overall uncertainty of the multivariate model and analyze it for different factors, providing a reference basis for selecting the optimal model.
Smart Images

Figure CN114492809B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of model uncertainty analysis, and in particular to an uncertainty analysis method for a marine environment design parameter estimation model. Background Art
[0002] Marine environmental design parameters are the basis of marine engineering design and construction. Since various marine environmental elements are often not independent of each other but have certain correlations, researchers pay more and more attention to the use of multivariate probability models to estimate marine environmental design parameters. However, in the process of estimating marine environmental design parameters, the selection of different data sampling methods, model distribution forms and parameter estimation methods makes the estimation results have greater uncertainty. Considering these uncertain factors in the design process of marine engineering helps to ensure the safety of engineering structures.
[0003] However, when the marine observation data is short, it is difficult for the samples to accurately describe the distribution characteristics of the marine observation data. This makes it difficult to obtain accurate results of marine environment design parameter estimation even if the appropriate model and parameter estimation method are selected. At the same time, since the calculation of the standard deviation and the mean requires the use of different sampling methods, model distribution forms or parameter estimation methods to obtain a variety of estimation results, it is difficult for existing methods to calculate the uncertainty caused by a specific sampling method, model distribution form or parameter estimation method, and it is also difficult to distinguish the uncertainty caused by the sample from the uncertainty caused by other factors and analyze them separately. Therefore, how to calculate the overall uncertainty of the multivariate model while calculating the uncertainty of the multivariate model caused by different factors and perform uncertainty analysis based on the various calculated uncertainties has become an urgent problem to be solved. Summary of the invention
[0004] In view of this, the purpose of the present invention is to overcome the defects in the prior art and provide an uncertainty analysis method for the marine environment design parameter estimation model, which can solve the shortcomings of the traditional uncertainty calculation method under small sample conditions, and can not only evaluate the overall uncertainty of the multivariate model, but also analyze a certain factor that causes the uncertainty of the multivariate model.
[0005] The uncertainty analysis method of the marine environment design parameter estimation model of the present invention comprises the following steps:
[0006] S1. Determine the uncertainty of the number of observed samples and the uncertainty of the degree of discreteness of the observed samples, and take the sum of the uncertainty of the number of observed samples and the uncertainty of the degree of discreteness of the observed samples as the random uncertainty of the estimation model;
[0007] S2. Determine the uncertainty of the inference model itself and the uncertainty of the estimated parameters of the inference model, and take the sum of the uncertainty of the inference model itself and the uncertainty of the estimated parameters of the inference model as the epistemic uncertainty of the inference model;
[0008] S3. Taking the sum of the random uncertainty and the epistemic uncertainty as the overall uncertainty of the inference model.
[0009] Furthermore, the uncertainty of the observed sample quantity U1 is determined according to the following formula:
[0010] U1=nΔH1;
[0011] Where n is the number of observation samples of the design parameters, and ΔH1 is the change in information entropy of the inference model caused by the increase in the number of observation samples.
[0012] Furthermore, the uncertainty U2 of the discrete degree of the observed sample is determined according to the following formula:
[0013] U2=σ x σ y ΔH2;
[0014] Among them, σ x is the observed sample standard deviation corresponding to the design parameter x, σ y is the observed sample standard deviation corresponding to the design parameter y, and ΔH2 is the change in information entropy caused by the increase in the product of the observed sample standard deviations.
[0015] Furthermore, the uncertainty U3 of the estimated model itself is determined according to the following formula:
[0016] U3=-∫∫f(x,y)ln[f(x,y)]dxdy;
[0017] Among them, f(x,y) is the joint distribution probability density function corresponding to the design parameters x and y.
[0018] Furthermore, the uncertainty U4 of the estimated model parameters is determined according to the following formula:
[0019]
[0020] Where m is the relative error W xi With W yi The number of intervals divided; the relative error W xi is the relative error caused by parameter estimation of the design parameter x; the relative error W yi is the relative error caused by parameter estimation of design parameter y; p xk is the probability of occurrence of the kth interval corresponding to the design parameter x; p yk is the probability of occurrence of the kth interval corresponding to the design parameter y.
[0021] Further, the relative error W is determined according to the following formula: xi With W yi :
[0022]
[0023] Wherein, (x0, y0) is the joint design value of the design parameters obtained by the calculation model once in N years; (x i ,y i ) is the joint design value of the design parameters obtained by the Monte Carlo method for the i-th simulation with a return period of N years.
[0024] Furthermore, the Monte Carlo method is used to simulate and obtain the joint design value of the design parameters once in N years, which specifically includes the following steps:
[0025] a. At time t = 0, a random initial state (x0, y0) is given;
[0026] b. From the conditional probability distribution f Y Sample data (x0,y1) from (y|x0);
[0027] c. From the conditional probability distribution f X Sampling from (x|y1) yields data (x1,y1);
[0028] d. Repeat steps bc, rotating the coordinate axis for sampling, and obtain data (x1, y2), (x2, y2), (x2, y3), ...
[0029] The beneficial effects of the present invention are as follows: the uncertainty analysis method of the marine environment design parameter estimation model disclosed in the present invention proposes a new overall uncertainty calculation method for the multivariate marine environment design parameter estimation model based on the information entropy theory, and at the same time uses the Monte Carlo method to solve the problem of specifically analyzing the influence of various factors on the model uncertainty under small sample conditions, and divides the factors affecting the model uncertainty into random factors and cognitive factors, and accordingly divides the overall uncertainty of the model into random uncertainty and cognitive uncertainty. Random uncertainty can be obtained by analyzing the influence of the number and discreteness of samples on the information entropy of the model; cognitive uncertainty can be obtained by calculating the information entropy of the model itself and the model prediction error. The present invention overcomes the shortcomings of traditional uncertainty calculation methods under small sample conditions, can evaluate the overall uncertainty of the multivariate model, and can also analyze a certain factor that causes the uncertainty of the multivariate model, and the overall uncertainty of the model can also be used as a reference for selecting the optimal model. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:
[0031] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION
[0032] The present invention is further described below in conjunction with the accompanying drawings, as shown in the drawings:
[0033] The uncertainty analysis method of the marine environment design parameter estimation model of the present invention comprises the following steps:
[0034] S1. Determine the uncertainty of the number of observed samples and the uncertainty of the degree of discreteness of the observed samples, and take the sum of the uncertainty of the number of observed samples and the uncertainty of the degree of discreteness of the observed samples as the random uncertainty of the estimation model;
[0035] S2. Determine the uncertainty of the inference model itself and the uncertainty of the estimated parameters of the inference model, and take the sum of the uncertainty of the inference model itself and the uncertainty of the estimated parameters of the inference model as the epistemic uncertainty of the inference model;
[0036] S3. Taking the sum of the random uncertainty and the epistemic uncertainty as the overall uncertainty of the inference model.
[0037] The overall uncertainty of the inference model is measured or evaluated through the overall uncertainty calculated above, and the overall uncertainty analysis of the inference model is realized. If the overall uncertainty is larger, the overall uncertainty of the inference model is larger, and if the overall uncertainty is smaller, the overall uncertainty of the inference model is smaller. Furthermore, by adjusting the parameter value in the overall uncertainty, the overall uncertainty is reduced, thereby reducing the overall uncertainty of the inference model, and then optimizing the inference model. In addition, the uncertainty of the inference model caused by a certain factor is measured or evaluated respectively through the above-calculated uncertainty of the number of observed samples, the uncertainty of the degree of discreteness of observed samples, the uncertainty of the inference model itself, and the uncertainty of the estimated parameters of the inference model, so as to realize the single uncertainty analysis of the inference model; at the same time, the random uncertainty and epistemic uncertainty of the inference model are measured or evaluated respectively by classifying the random uncertainty and epistemic uncertainty, so as to realize the random uncertainty and epistemic uncertainty analysis of the inference model. It should be noted that the several uncertainties of the above calculations are generally positively correlated with the uncertainty of the inference model, that is, as the uncertainty increases, the uncertainty of the inference model also increases; as the uncertainty decreases, the uncertainty of the inference model also decreases.
[0038] The present invention overcomes the shortcomings of traditional uncertainty calculation methods under small sample conditions. It can not only evaluate the overall uncertainty of the multivariate model, but also analyze a certain factor that causes the uncertainty of the multivariate model. The overall uncertainty of the model can also be used as a reference for selecting the optimal model.
[0039] In this embodiment, if X and Y represent two random variables of marine environment design parameters (such as wind speed and water surge), let f(x, y) be the joint distribution probability density function (also called model) of these two random variables. Then the joint information entropy of X and Y is defined as:
[0040] H f =-∫∫f(x,y)ln[f(x,y)]dxdy. (1)
[0041] The joint information entropy can be used to measure the uncertainty of the binary distribution model f(x,y). The smaller the joint information entropy of the model, the smaller its uncertainty; the larger the joint information entropy of the model, the greater its uncertainty.
[0042] When using the observation data of marine environment design parameters to establish the model f(x,y), the number and discreteness of the observation samples have a significant impact on the information entropy of the model. The former is defined as the uncertainty U1 caused by the number of samples, and the latter is defined as the uncertainty U2 caused by the discreteness of the samples. Since the observation samples are random, the sum of these two uncertainties is called random uncertainty U R . So we have:
[0043] U R =U1+U2 (2)
[0044] For joint information entropy, there is the following theorem:
[0045] Assume that the continuous random variables X and Y reach the maximum information entropy H when the joint probability density function is g(x,y) g Another joint probability density function of random variables X and Y under the same constraints is h(x,y), and its information entropy is H h Then the information entropy change satisfies the following formula:
[0046]
[0047] Where n is the number of samples of the continuous random variable, and γ is the likelihood function ratio of g(x,y) to h(x,y).
[0048] From the references we can know:
[0049] -∫∫g(x,y)ln[g(x,y)]dxdy=-∫∫h(x,y)ln[g(x,y)]dxdy (4)
[0050] Therefore, the information entropy becomes worse:
[0051]
[0052] The likelihood function ratio of h(x,y) to g(x,y) is:
[0053]
[0054] In the formula is the likelihood function of the corresponding distribution, and is the measured data vector, is the parameter vector. Taking the logarithm of both sides of equation (6), when n→∞, we can get:
[0055]
[0056] Multiply both sides of equation (7) by h(x,y) and integrate to obtain:
[0057]
[0058] Substituting equation (8) into equation (5) yields equation (3), and the proof is complete.
[0059] When n→∞, the parameter value obtained by maximum likelihood estimation is the parameter value of the true distribution, so the likelihood function value can be considered and remains unchanged. So the likelihood function ratio γ also remains unchanged. But because n→∞, H g,h tends to 0. That is, when the number of samples is large enough, the joint information entropy of the multivariate model increases with the increase of the number of samples. Assuming that there is a linear relationship between the number of samples and the joint information entropy of the multivariate model, the following inference can be obtained:
[0060] The uncertainty caused by the number of samples can be calculated according to the following formula:
[0061] U1=nΔH1 (9)
[0062] Where n is the number of samples, and ΔH1 is the change in model information entropy caused by the increase in the number of samples.
[0063] Furthermore, if the joint probability density function of continuous random variables X and Y is f(x,y), their joint information entropy is H(x,y). The mathematical expectation and standard deviation of X are μ x and σ x The mathematical expectation and standard deviation of Y are μ y and σ y After X and Y are standardized, X′=(X-μ x ) / σ x and Y′=(Y–μ y ) / σ y The joint probability density function of X′ and Y′ is f′(x′,y′), and the joint information entropy is H f′ . Then we have the following relationship:
[0064] Hf =H f′ +ln(σ x σ y ) (10)
[0065] From formula (1), we can know that:
[0066] H f′ =-∫∫f′(x′,y′)ln[f′(x′,y′)]dx′dy′ (11)
[0067] Easy to get:
[0068] f′(x′,y′)=f(σ x x′+μ x ,σ y y′+μ y )σ x σ y (12)
[0069] Substituting formula (12) into formula (11), we can obtain:
[0070]
[0071] By transposing the terms in equation (13), we can obtain equation (10).
[0072] According to the above, the difference between the joint information entropy of the multivariate model and the joint information entropy of the standardized model is a constant. This constant is the logarithm of the product of the standard deviations of the two variable samples. Because the standard deviation represents the degree of dispersion of the sample, the uncertainty of the model increases with the increase of the dispersion of the data. Assuming that there is a linear relationship between the product of the standard deviation of the sample and the joint information entropy of the multivariate model, the following inference can be obtained:
[0073] The uncertainty caused by the sample dispersion can be calculated according to the following formula:
[0074] U2=σ x σ y ΔH2 (14)
[0075] Where σ x With σ y is the standard deviation of the sample, and ΔH2 is the change in information entropy caused by the increase in the product of the sample standard deviation.
[0076] Choosing different multivariate models to estimate the design parameters of the marine environment will bring greater uncertainty to the calculation results. Different model parameter estimation methods will affect the model's fit to the sample data. Therefore, both the model distribution form and the parameter estimation method will affect the uncertainty of the model. The former is defined as the uncertainty of the model itself U3, and the latter is defined as the uncertainty caused by parameter estimation U4. Since the distribution form and parameter estimation method of the multivariate model have certain cognitive limitations, the sum of these two uncertainties is called cognitive uncertainty U C . So we have:
[0077] U C =U3+U4 (15)
[0078] The uncertainty of the model itself can be directly obtained from the information entropy, that is:
[0079] U3=-∫∫f(x,y)ln[f(x,y)]dxdy (16)
[0080] Where f is the probability density function of the joint distribution.
[0081] Assume the joint distribution probability density function is in is the model parameter estimated from the sample. (x0,y0) is the joint design value of the marine environment design parameter obtained by the model for a return period of N years. i ,y i ) is the joint design value of N-year return period obtained by the Monte Carlo method for the i-th simulation. Then the relative error caused by parameter estimation can be expressed as:
[0082]
[0083] The error W xi With W yi Divide into m intervals, and count the frequency p of the kth interval xk With p yk , then the uncertainty caused by parameter estimation can be calculated according to the following formula:
[0084]
[0085] The random uncertainty of the estimated model is called the first type of uncertainty, and the epistemic uncertainty of the estimated model is called the second type of uncertainty. The sum of the first type of uncertainty and the second type of uncertainty is called the overall uncertainty U of the model, that is:
[0086] U=U R +U C (19)
[0087] In this embodiment, the Monte Carlo method can be used to easily generate simulated data that meets a specific multivariate model. The present invention uses the Gibbs sampling method. It should be noted that although the algorithms developed on the basis of the Gibbs sampling method, such as clustering algorithms and hybrid Gibbs sampling methods, have higher computational efficiency. However, the focus of the present invention is not the pros and cons of the algorithm, but whether simulated data can be obtained more conveniently. Since the Gibbs sampling method is easy to program and easy to apply in engineering, it is adopted in the present invention. If the probability density function of a binary distribution model is f(x, y), the steps of generating simulated data using the Gibbs algorithm are as follows:
[0088] a. At time t = 0, a random initial state (x0, y0) is given;
[0089] b. From the conditional probability distribution f Y Sample data (x0,y1) from (y|x0);
[0090] c. From the conditional probability distribution f X Sampling from (x|y1) yields data (x1,y1);
[0091] d. Repeat steps bc, rotating the coordinate axis for sampling, and obtain data (x1, y2), (x2, y2), (x2, y3), ...
[0092] Taking a certain sea area as an example, the uncertainty analysis method of the marine environment design parameter estimation model of the present invention is further described below:
[0093] In marine engineering, the one-dimensional models commonly used to fit extreme value data include Gumbel, Weibull, Pearson-Ⅲ and generalized extreme value distribution (GEV) models. Maximum entropy distribution (ME) is a distribution model derived under the condition of maximizing the information entropy of the model. This model can make the least assumptions about the model under the condition of known information, so the ME model can minimize the impact of human settings. Because it is subject to the least constraints, theoretically the ME model has the greatest uncertainty compared to other models. The present invention uses the above five models to fit the annual extreme wind speed V w and annual extreme water increase H s The ME model can also fit V w and H s Data; GEV model and Weibull model can fit V well respectively w Data and H s Data. Use the above three models to fit V w and H s The parameters obtained from the data are shown in Table 1.
[0094] Table 1
[0095]
[0096] In marine engineering, Copula functions are often used to connect two marginal distributions to construct a binary distribution model. Commonly used Copula functions include Gumbel Copula, Clatyon Copula, Frank Copula, Gaussian Copula, etc. In addition to this method, a two-dimensional maximum entropy distribution (2-DME) model can be obtained under the condition that the model information entropy is maximum. Because the information entropy of the 2-DME model is the largest, its own uncertainty should be the largest among all binary distribution models. For further comparative analysis, the present invention will calculate the uncertainty of the 9 binary distribution models listed in Table 2. The probability density function expressions and their parameters of different binary distribution models are given in Table 2.
[0097] Table 2
[0098]
[0099] Note: g1(x,y)=x,g2(x,y)=x 2 ,g3(x,y)=ln(x),g4(x,y)=y,g5(x,y)=y 2 ,g6(x,y)=ln(y),g7(x,y)=xy
[0100] The first type of uncertainty of the multivariate model is directly related to the number of samples and the degree of sample dispersion. However, due to the late development of the observation technology of marine environmental design parameters (or elements), it is difficult to obtain hundreds or even thousands of years of marine environmental design parameter observation data in engineering practice. Therefore, it is also difficult to calculate the first type of uncertainty using measured data. The present invention uses the Monte Carlo method to simulate sampling for different binary distribution models, thereby obtaining several groups (each group contains 1000 data) of simulated V w With H s Corresponding to the 9 models mentioned above, the simulated data are recorded as D 2,2-DME , D 2,GC-Ⅰ , D 2,GC-Ⅱ , D 2,CC-Ⅰ , D 2,CC-Ⅱ , D 2,FC-Ⅰ , D 2,FC-Ⅱ , D 2,GaC-Ⅰ , D 2,GaC-Ⅱ By setting the analog data to V w With H s In the coordinate system with the axes as coordinates, it can be seen that the simulated data obtained by each binary model through the Monte Carlo method can be well consistent with the annual extreme value data.
[0101] Calculation and analysis of the first type of uncertainty: Using the simulated data obtained above to study the relationship between the information entropy of the model and the number of samples, it can be seen that for various models, the information entropy increases first and then remains stable as the number of samples increases. That is, the uncertainty of the estimated model first increases and then remains stable as the number of samples increases. In addition, before the model information entropy reaches the first peak, the model information entropy increases sharply with the number of samples almost linearly. The measured data used in the present invention only contains 23 annual extreme value sampling data. In this case, the model information entropy increases linearly with the number of samples. Based on this feature, ΔH1 in formula (9) can be obtained. Taking n as 23, the uncertainty U1 caused by the number of samples can be obtained. The results are shown in Table 3.
[0102] Table 3
[0103]
[0104] From the above analysis, it can be seen that there is a close relationship between the standard deviation of the statistic reflecting the degree of sample dispersion and the information entropy of the model. For the binary model, the larger the product of the standard deviations of the samples of the two marine environmental variables, the larger the information entropy of the model. The larger the information entropy of the model, the greater the uncertainty of the model. In other words, the more discrete the sample data, the greater the uncertainty of the model obtained using the sample. Using simulated data to study the relationship between the degree of sample dispersion and the information entropy of the model, it can be seen that the linear regression curve obtained by the least squares method and its 95% confidence interval cover most of the data points. Moreover, the information entropy of the 9 models is positively correlated with the product of the sample standard deviation. This verifies the results of the theoretical analysis. By studying the relationship between the degree of sample dispersion and the information entropy of the model through simulated data, the information entropy change ΔH2 caused by the change in the degree of sample dispersion of each model can be obtained. Measured V w With H s The standard deviation of the data is σ x With σ y They are 4.7233 and 0.3044 respectively. Substituting the above data into formula (14), we can get the uncertainty U2 caused by the sample dispersion. Summing the uncertainty caused by the sample size and the sample dispersion, we can get the first type of uncertainty U R The calculation results are shown in Table 3.
[0105] Calculation and analysis of the second type of uncertainty: According to the model parameter estimation results in Tables 1 and 2, the uncertainty of the model itself U3 can be calculated using formula (16), and the results are shown in Table 3. By showing the probability density contour maps of each model, it can be seen that when the same Copula function is used to construct a binary model, the probability density contour maps of the model are very close. The calculation result of the model's own uncertainty U3 is also very close. However, when different Copula functions are used to construct a binary model, the probability density contour maps of the model are quite different. This shows that compared with the choice of marginal distribution, whether the multivariate model can better describe the relationship between marginal distributions is a factor that has a greater impact on model uncertainty.
[0106] The joint return period is defined as 1 / (1-F(x,y)), where F(x,y) is the cumulative distribution function corresponding to the model. According to the parameters listed in Table 1-2, we can obtain w With H s The 100-year joint design value (x0, y0) of each model with equal probability. The Monte Carlo method is used to perform 10,000 simulations, and 1,000 V values of different models are obtained in each simulation. w With H s Data. Using the empirical distribution function, the V w With H s Joint design value (x i ,y i ). Thus, 10,000 groups of V w With H s Combined design value. The relative error caused by parameter estimation is calculated by formula (17). The error is evenly divided into 50 intervals, the frequency of each interval is counted, and then the uncertainty U4 caused by parameter estimation is obtained by formula (18). The uncertainty U4 caused by parameter estimation is related to the second type of uncertainty U C As shown in Table 3, by analyzing the relative errors of the prediction results of each model, it can be seen that the relative error range of the two-dimensional maximum entropy distribution is significantly larger than that of other models. Correspondingly, the uncertainty caused by its parameter estimation in Table 3 is also the largest among all models. In other words, the uncertainty of the estimated model increases as the relative error range caused by the estimated model parameter estimation becomes wider.
[0107] Calculation and analysis of overall uncertainty: The AIC information criterion is a method based on the concept of information entropy and is often used to describe the degree of model fit. This criterion believes that the smaller the AIC value of the model, the better the model fits. Table 3 shows the calculation results of the uncertainty and AIC value of each model, where the maximum value is marked with * in the upper right corner and the minimum value is marked with # in the upper right corner. As shown in Table 3, the AIC information criterion believes that 2-DME has the worst fit and CC-Ⅰ has the best fit. This conclusion is consistent with the conclusions obtained from the uncertainty U2 caused by the sample discreteness and the uncertainty U3 of the model itself. However, it is different from the conclusions obtained from the uncertainty U1 caused by the sample number and the uncertainty U4 caused by the model parameter estimation. This shows that although the commonly used goodness of fit test method can evaluate the quality of model fit, it cannot reflect the uncertainty brought to the model by the sample number and parameter estimation. Therefore, the use of AIC to evaluate the goodness of fit of the model may result in model overfitting. That is, although the model fits the sample well, the extrapolation result is not accurate due to the large uncertainty of the model. Judging the quality of the model from the perspective of model uncertainty, the following conclusions can be drawn: According to the first type of uncertainty U R , GaC-Ⅰ model is the best, and 2-DME model is the worst. According to the second type of uncertainty U C , GC-Ⅰ model is the best, and 2-DME model is the worst. The results obtained based on the overall uncertainty U of the model are consistent with the results obtained based on the first type of uncertainty U of the model. R The same results are obtained.
[0108] The impact of model uncertainty on marine environment design parameters: Assumption V w With H s The probability is equivalent. By calculating the multi-year joint design value of each model under different joint return periods, it can be seen that whether it is V w Or H s, and no matter whether the extreme value distribution (Type-Ⅰ) or the maximum entropy distribution (Type-Ⅱ) is used as the marginal distribution, the joint design values obtained by each model have the same trend of change. That is, the positions of the joint design value curves of each model are 2-DME, FC-Ⅰ / Ⅱ, GaC-Ⅰ / Ⅱ, GC-Ⅰ / Ⅱ, and CC-Ⅰ / Ⅱ from high to low. This trend is consistent with the change trend of the model's own uncertainty U3. Moreover, taking the calculation results of 2-DME as a reference, it can be seen that the joint design value curves of FC-Ⅰ, GaC-Ⅰ, GC-Ⅰ, and CC-Ⅰ models are all lower than the joint design value curves of FC-Ⅱ, GaC-Ⅱ, GC-Ⅱ, and CC-Ⅱ models. At the same time, the own uncertainty U3 of FC-Ⅰ, GaC-Ⅰ, GC-Ⅰ, and CC-Ⅰ models in Table 3 is also smaller than the own uncertainty of FC-Ⅱ, GaC-Ⅱ, GC-Ⅱ, and CC-Ⅱ. Therefore, it can be considered that the larger the uncertainty U3 of the model itself is, the wider the range of model distribution is, and the larger the joint design value of the model is under the same joint recurrence period. That is, the uncertainty of the estimated model increases as the range of the estimated model itself is wider.
[0109] The above analysis is about the uncertainty of the estimation model of the design parameters of the marine environment in a certain sea area. According to the source model of uncertainty, uncertainty is divided into two categories. The first type of uncertainty is random uncertainty. The source of random uncertainty is samples with random properties. Both the number of samples and the degree of discreteness of samples will have a significant impact on the random uncertainty of the model. With the increase of the number of samples, the random uncertainty of the model has a trend of first increasing linearly and then tending to be stable. With the increase of the degree of discreteness of the samples, the random uncertainty of the model tends to increase linearly, that is, the uncertainty of the estimation model increases with the increase of the degree of discreteness of the observed samples. The second type of uncertainty is epistemic uncertainty. The source of epistemic uncertainty is the model itself. Both the distribution form of the model and the model parameter estimation method will have a direct impact on the epistemic uncertainty of the model. For a certain sea area, the model with the smallest random uncertainty is the GaC-Ⅰ model, but the model with the smallest epistemic uncertainty is the GC-Ⅰ model.
[0110] In a certain sea area, among the nine binary models, the 2-DME model has the largest information entropy and is subject to the least artificial settings. Theoretically, the 2-DME model should be the model that best fits the natural state. However, precisely because it is subject to the least constraints, the overall uncertainty and various uncertainties of the 2-DME model are the largest. At the same time, its goodness of fit is the worst. Among the nine models, the GaC-Ⅰ model has the smallest overall uncertainty, but its goodness of fit is not the best but ranks third. Compared with selecting a marginal distribution with a better fitting effect, whether the binary model can better describe the relationship between the two marginal distributions is a factor that has a greater impact on the model's cognitive uncertainty. When the goodness of fit of the model is close, the better model can be selected by referring to the overall uncertainty of the model, that is, the model with a lower overall uncertainty is used as the better model. Doing so can not only ensure a better fitting effect but also avoid the situation where the model uncertainty is too high due to too few assumptions about the model. By analyzing the joint design values of wind speed and water increase under different models in a certain sea area, it can also be concluded that the inherent uncertainty U3 among the various uncertainties of the model has a more obvious impact on the joint design value.
[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.
Claims
1. An uncertainty analysis method for a marine environment design parameter estimation model, characterized in that: The steps include: S1. Determine the uncertainty of the number of observed samples and the uncertainty of the degree of discreteness of the observed samples, and take the sum of the uncertainty of the number of observed samples and the uncertainty of the degree of discreteness of the observed samples as the random uncertainty of the estimation model; S2. Determine the uncertainty of the inference model itself and the uncertainty of the estimated parameters of the inference model, and take the sum of the uncertainty of the inference model itself and the uncertainty of the estimated parameters of the inference model as the epistemic uncertainty of the inference model; The uncertainty of the estimated model parameter U4 is determined according to the following formula: Where m is the relative error W xi With W yi The number of divided intervals; the relative error W xi is the relative error caused by parameter estimation of the design parameter x; the relative error W yi is the relative error caused by parameter estimation of design parameter y; p xk is the probability of occurrence of the kth interval corresponding to the design parameter x; p yk is the probability of occurrence of the kth interval corresponding to the design parameter y; The relative error W is determined according to the following formula: xi With W yi : Wherein, (x0, y0) is the joint design value of the design parameters obtained by the calculation model once in N years; (x i ,y i ) is the N-year joint design value of the design parameters obtained by the Monte Carlo method for the i-th simulation; S3. Taking the sum of the random uncertainty and the epistemic uncertainty as the overall uncertainty of the inference model.
2. The uncertainty analysis method of the marine environment design parameter estimation model according to claim 1 is characterized by: Determine the uncertainty of the observed sample quantity U1 according to the following formula: U1=nΔH1; Where n is the number of observation samples of the design parameters, and ΔH1 is the change in information entropy of the inference model caused by the increase in the number of observation samples.
3. The uncertainty analysis method of the marine environment design parameter estimation model according to claim 1 is characterized by: Determine the uncertainty U2 of the discrete degree of the observed sample according to the following formula: U2=σ x s y ΔH2; Among them, σ x is the observed sample standard deviation corresponding to the design parameter x, σ y is the observed sample standard deviation corresponding to the design parameter y, and ΔH2 is the change in information entropy caused by the increase in the product of the observed sample standard deviations.
4. The uncertainty analysis method of the marine environment design parameter estimation model according to claim 1 is characterized by: The uncertainty U3 of the estimated model itself is determined according to the following formula: U3=-∫∫f(x,y)ln[f(x,y)]dxdy; Among them, f(x,y) is the joint distribution probability density function corresponding to the design parameters x and y.
5. The uncertainty analysis method of the marine environment design parameter estimation model according to claim 1 is characterized by: The Monte Carlo method is used to simulate and obtain the joint design value of the design parameters once in N years, which specifically includes the following steps: a. At time t = 0, a random initial state (x0, y0) is given; b. From the conditional probability distribution f Y Sample data (x0,y1) from (y|x0); c. From the conditional probability distribution f X Sampling from (x|y1) yields data (x1,y1); d. Repeat steps bc, rotating the coordinate axis for sampling, and obtain data (x1, y2), (x2, y2), (x2, y3), ...
Citation Information
Patent Citations
Parameter uncertainty quantitative modeling method under interval data sample condition
CN113962145A
