Method for evaluating marine environment influence ship navigation safety risk under high-dimensional small sample condition

Abstract

The invention discloses a marine environment influence ship navigation safety risk assessment method under a high-dimensional small sample condition. The method comprises the following steps: S1, establishing a B-R high-dimensional association risk sample expansion model; s2, coupling the B-R high-dimensional association risk sample expansion model with a generalized regression neural network model, and optimizing model parameters to obtain a BR-GRNN risk assessment model under the condition of high-dimensional small samples; and S3, evaluating the navigation safety risk of the ship influenced by the marine environment by using the BR-GRNN risk evaluation model under the condition of high-dimensional small samples, and clustering risk evaluation values to obtain a risk zoning map of the navigation area. According to the method, the accuracy of ship navigation safety risk assessment influenced by the marine environment under the high-dimensional small sample condition is improved, and the method has relatively high reliability and application potential.

Description

Technical Field

[0001] The present invention belongs to the technical field of safety risk assessment, and more specifically, relates to a method for assessing the risk of marine environment affecting ship navigation safety under high-dimensional small sample conditions. Background Art

[0002] Safe and efficient maritime transport is an important guarantee for the sustainable development of international trade and ensuring energy supply. With global warming, the marine environment has become complex and harsh, seriously affecting the navigation safety of ships. In 2023, there were 1,867 accidents involving ships over 3,000 tons, an increase of 330 from 2013. The navigation safety of ships under severe sea conditions is the basic goal of meteorological navigation. Objectively and accurately assessing the impact of the marine environment on the navigation safety of ships and scientifically and reasonably judging the risk of impact on the marine environment are important contents of meteorological and hydrological support.

[0003] Accurate and sufficient data information is the premise for correctly assessing risks. However, since risks are often difficult to observe or record, the data information and case samples of the correlation between marine environmental impact risks and marine hydrological and meteorological elements are extremely scarce or seriously missing. The available data information is only a small amount of experimental data samples or qualitative empirical knowledge. On the other hand, there are many marine environmental risk factors that affect the safety of ship navigation, and these factors have extremely complex nonlinear correlations. In summary, high-dimensional small samples are difficult problems that affect and restrict the risk assessment of ship navigation safety.

[0004] The mainstream methods of ship navigation safety risk assessment (such as weighted comprehensive method, fuzzy comprehensive evaluation, grey correlation analysis, etc.) are essentially still a linear weighted method. The weighted comprehensive method is simple and rough. Fuzzy comprehensive evaluation or grey correlation analysis is often used in practical applications. However, these two methods do not use risk value information when modeling. The established risk assessment model cannot characterize the nonlinear mapping relationship between risk assessment indicators and risks. Moreover, these methods all need to solve the weights. The hierarchical analysis method is a commonly used weighting method, but the comparison and judgment results of the hierarchical analysis method are relatively rough. Generally, the comparison and judgment matrix is ​​established based on the prior knowledge of experts. The traces of human manipulation are too strong and subjective influences cannot be avoided.

[0005] The atmosphere and ocean are highly complex nonlinear systems, which makes it difficult to describe the behavior of the systems through simple linear relationships. With the development of artificial intelligence, machine learning models such as generalized regression neural networks are gradually being applied to the field of risk assessment, because they can automatically learn high-level abstract features from data without the need for manual feature engineering, and can better capture the complex nonlinear relationship between meteorological and oceanographic data and navigation safety risks. However, machine learning models require a large number of training samples, and the high-dimensional small sample problem makes it impossible for machine learning models to be directly applied to the field of ship navigation safety risk assessment. The risk assessment method of the impact of the marine environment on ship navigation safety is in urgent need of improvement, development and innovation.

[0006] Therefore, there is an urgent need for a new risk assessment method for the impact of marine environment on ship navigation safety under high-dimensional and small sample conditions. Summary of the invention

[0007] Aiming at the objective fact and scientific problem that there are many factors affecting the risk of marine environment affecting the safety of ship navigation and few case samples, the present invention proposes a risk assessment method for the impact of marine environment on the safety of ship navigation under high-dimensional and small sample conditions.

[0008] In order to solve at least one of the above technical problems, according to one aspect of the present invention, a method for assessing the risk of marine environment affecting ship navigation safety in a high-dimensional small sample situation is provided, comprising the following steps:

[0009] S1. Establish BR high-dimensional correlation risk sample expansion model;

[0010] The associated risk sample refers to the high-dimensional risk assessment index value X and the corresponding risk value Y. The i-th high-dimensional associated risk sample is expressed as:

[0011] (X i ,Y i )=(x i,1 ,x i,2 ,…,x i,n ,y i ) (1)

[0012] In the formula, n represents the number of high-dimensional risk assessment indicators;

[0013] S2, coupling the BR high-dimensional associated risk sample expansion model with the generalized regression neural network model, optimizing the model parameters, and obtaining the BR-GRNN risk assessment model under the high-dimensional small sample situation;

[0014] S3. The BR-GRNN risk assessment model under high-dimensional small sample conditions is used to assess the risk of the marine environment affecting the navigation safety of ships, and the risk assessment values ​​are clustered to obtain a risk zoning map of the navigation area.

[0015] Furthermore, step S1 specifically includes the following steps:

[0016] S11, establishing a high-dimensional expansion index array;

[0017] S12. Establish a relationship model between the high-dimensional index value X and the risk value Y, which is the expanded high-dimensional index data X new Calculate the corresponding risk value Y new .

[0018] Furthermore, in S11, the steps for establishing the high-dimensional expansion index array are as follows:

[0019] S111, vertically cut the high-dimensional risk assessment index data and generate n labeled arrays (l1, l2, ···, l n ), let l j Represents the row number of the j-th column data, i.e. l j =(1,2,…,n) T ;

[0020] Suppose the matrix of high-dimensional small sample risk assessment indicators is X, the size of X is m×n, where m≤30, and X can be expressed as:

[0021]

[0022] S112, bind the two columns of data with strong intra-group correlation, that is, (l1,l2,···,l n ) are combined with several indicator data with a Pearson correlation coefficient greater than or equal to 0.6 to obtain a new number array

[0023] S113, using Bootstrap method Each column of is resampled, and m rows of new samples are obtained. Repeat the operation until the desired sample size is reached. The expanded new sample is recorded as X PP , X PP Simulate the strong correlation information contained in the original high-dimensional risk data, X PP The structure is shown below:

[0024]

[0025] S114, the original high-dimensional index data X and the expanded X PP The new data array X after merging and expansion new ; After expansion, X new The size of becomes (k+1)m×n, where k is the expansion multiple and k>2.

[0026] Furthermore, in S12, the specific steps of the relationship model between the high-dimensional index value X and the risk value Y are:

[0027] S121, let each high-dimensional correlation sample be (X i ,Y i )=(x i,1 ,x i,2 ,…,x i,n ,y i ), where i = 1, 2, ..., m, and these m associated samples are used as training data for the multivariate linear regression model, thereby obtaining the relationship f between the high-dimensional index value X and the risk value Y;

[0028] S122, the expanded new sample X PP Input into the relation f, thus obtaining PP The corresponding risk value vector Y PP ;

[0029] S123, put X PP and Y PP Merge to obtain a high-dimensional correlation sample matrix (X PP ,Y PP ), and (X PP ,Y PP ) is combined with the original high-dimensional correlated small sample matrix (X, Y) to obtain a new high-dimensional correlated sample matrix (X new ,Y new ).

[0030] Furthermore, the specific steps of S2 are:

[0031] S21, based on the expanded high-dimensional correlation sample matrix (X new ,Y new ) train the generalized regression neural network model and establish a nonlinear optimization model for the smoothing factor σ; thereby having sufficient training samples to optimize the parameters of the generalized regression neural network model;

[0032] S22. Design an efficient algorithm to solve the nonlinear optimization model about the smoothness factor σ and obtain the optimal smoothness factor.

[0033] Furthermore, the specific steps of S21 are as follows: In order to effectively avoid underfitting and overfitting in the σ optimization process, with the help of k-fold cross validation, the objective function of the nonlinear optimization model is first established, and the process is as follows:

[0034] For the sample data set Grouping, let the kth subset be Among them, m kis the sample size of the subset, each subset data is used as the validation set, and the remaining k-1 groups of subset data are used as the training set, thereby obtaining k models. The average mean square error of the k model validation set is used as the nonlinear optimization objective function E of the cubic smoothing factor σ. The calculation formula is shown in (7) and (8):

[0035]

[0036] Furthermore, the nonlinear optimization model of the smooth factor σ can be established according to the nonlinear optimization objective function of the smooth factor σ:

[0037]

[0038] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method for assessing the risk of the impact of the marine environment on ship navigation safety under high-dimensional small sample conditions of the present invention.

[0039] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the method for assessing the risk of the impact of the marine environment on ship navigation safety under high-dimensional small sample conditions of the present invention.

[0040] Compared with the existing technology, the beneficial effects of the above method of the present invention are: compared with the existing technology model, the present invention greatly improves the accuracy of the risk assessment of the impact of the marine environment on ship navigation safety under high-dimensional small sample conditions, and has greater reliability and application potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments will be briefly introduced below. Obviously, the drawings in the following description only relate to some embodiments of the present invention, but are not intended to limit the present invention.

[0042] Figure 1 A method flow chart of a preferred embodiment of the present invention;

[0043] Figure 2 A BR expansion technology flow chart of a preferred embodiment of the present invention;

[0044] Figure 3 A flowchart of a quantum genetic algorithm for solving the optimal smoothing factor according to a preferred embodiment of the present invention;

[0045] Figure 4 A schematic diagram showing the comparison between the risk assessment values ​​obtained by the four models and the actual values ​​under 30 training samples in a preferred embodiment of the present invention;

[0046] Figure 5 A schematic diagram showing comparison between risk assessment values ​​obtained by four models and actual values ​​under 20 training samples in a preferred embodiment of the present invention;

[0047] Figure 6 A schematic diagram showing comparison between risk assessment values ​​obtained by four models and actual values ​​under 10 training samples in a preferred embodiment of the present invention;

[0048] Figure 7 A schematic diagram showing comparison between risk assessment values ​​obtained by four models and actual values ​​under five training samples in a preferred embodiment of the present invention;

[0049] Figure 8 This is a schematic diagram of the comparison of average relative errors of four models under different small sample scenarios in a preferred embodiment of the present invention;

[0050] Fig. 9 This is a schematic diagram of the comparison of mean square errors of four models under different small sample scenarios according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0051] To make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0052] Unless otherwise defined, technical or scientific terms used herein shall have the same meaning as commonly understood by one of ordinary skill in the art to which the invention belongs.

[0053] Embodiment 1:

[0054] like Figure 1-9 As shown, the present invention proposes a method for assessing the risk of marine environment affecting ship navigation safety under high-dimensional small sample conditions, comprising the following steps:

[0055] S1. Establish a BR high-dimensional correlation risk sample expansion model.

[0056] like Figure 2 The figure shows the process flow of BR high-dimensional associated risk sample expansion technology.

[0057] The associated risk sample refers to the high-dimensional risk assessment index value X and the corresponding risk value Y. The i-th high-dimensional associated risk sample can be expressed as:

[0058] (X i ,Y i )=(x i,1 ,x i,2 ,…,x i,n ,yi ) (1)

[0059] Where n represents the number of high-dimensional risk assessment indicators.

[0060] Step S1 includes:

[0061] S11. Establish a high-dimensional expansion index array.

[0062] S111, vertically cut the high-dimensional risk assessment index data and generate n labeled arrays (l1, l2, ···, l n ), let l j Represents the row number of the j-th column data, i.e. l j =(1,2,…,n) T ;

[0063] Suppose the matrix of high-dimensional small sample risk assessment indicators is X, the size of X is m×n, where m≤30, and X can be expressed as:

[0064]

[0065] S112, bind the two columns of data with strong intra-group correlation, that is, (l1,l2,···,l n ) are combined with several indicator data with a Pearson correlation coefficient greater than or equal to 0.6 to obtain a new number array

[0066] S113, using Bootstrap method Each column of is resampled, and m rows of new samples are obtained. Repeat the operation until the desired sample size is reached. The expanded new sample is recorded as X PP , we can see that X PP It can simulate the strong correlation information contained in the original high-dimensional risk data, X PP The structure is shown below:

[0067]

[0068] S114, the original high-dimensional index data X and the expanded X PP The new data array X after merging and expansion new After the above expansion process, X new The size of becomes (k+1)m×n, where k is the expansion multiple and k>2.

[0069] S12. Establish a relationship model between the high-dimensional index value X and the risk value Y, which is the expanded high-dimensional index data X new Calculate the corresponding risk value Y new .

[0070] S121, let each high-dimensional correlation sample be (X i ,Y i )=(x i,1 ,x i,2 ,…,x i,n ,y i ), where i = 1, 2, ..., m, and these m associated samples are used as training data for the multivariate linear regression model, thereby obtaining the relationship f between the high-dimensional index value X and the risk value Y.

[0071] S122, the expanded new sample X PP Input into the relation f, thus obtaining PP The corresponding risk value vector Y PP .

[0072] S123, put X PP and Y PP Merge to obtain a high-dimensional correlation sample matrix (X PP ,Y PP ), and (X PP ,Y PP ) is combined with the original high-dimensional correlated small sample matrix (X, Y) to obtain a new high-dimensional correlated sample matrix (X new ,Y new ), which solves the technical problem of high-dimensional small sample correlation data in risk assessment modeling.

[0073] S2. The BR high-dimensional associated risk sample expansion model is coupled with the generalized regression neural network (GRNN) model, and the model parameters are optimized to obtain a BR-GRNN risk assessment model under high-dimensional small sample conditions.

[0074] The basic principles and network topology of the GRNN model are explained:

[0075] Assume that the joint probability density function of random variables X and Y is f(x,y), and the observed value of X is When the input is When , the evaluation data of Y relative to X is:

[0076]

[0077] Based on the Parzen-window nonparametric estimation method, using the sample data set The probability density function can be Make an estimate:

[0078]

[0079] Where n is the sample size, P is the dimension of the random variable X, and σ is the smoothing factor. The prediction output formula is as follows:

[0080]

[0081] in, The numerator in (6) is y obtained from the training sample data set i The weighted sum of

[0082] The network topology of the generalized regression neural network includes input layer, pattern layer, summation layer and output layer. The input is the X observation value The output is The sample dataset is

[0083] Step S2 includes:

[0084] S21, based on the expanded high-dimensional correlation sample matrix (X new ,Y new )The generalized regression neural network model is trained and a nonlinear optimization model about the smoothing factor σ is established.

[0085] After the original high-dimensional small sample data is expanded using the above-mentioned expansion model, the high-dimensional small sample can be expanded into a large sample, meeting the requirements of the generalized regression neural network model for the amount of training sample data, thereby having sufficient training samples to optimize the parameters of the generalized regression neural network model.

[0086] From formula (5), we can see that the generalized regression neural network model has an unknown parameter σ (smooth factor). The value of σ has a great influence on the evaluation performance of the network. If the value of σ is too large, the evaluation output value will be If the value of σ is too small, the output value will be too low. The value of σ is close to the value of y, which makes the evaluation effect of the test set worse, and overfitting occurs. In order to effectively avoid underfitting and overfitting in the process of σ optimization, the objective function of the nonlinear optimization model is first established with the help of k-fold cross validation. The specific establishment process is as follows:

[0087] For the sample data set Grouping, let the kth subset be (m k is the sample size of the subset), each subset data is used as a validation set, and the remaining k-1 groups of subset data are used as training sets, thereby obtaining k models. The average mean square error of the k validation sets in these models is used as the nonlinear optimization objective function E of the cubic smoothing factor σ. The calculation formulas are shown in (7) and (8).

[0088]

[0089] The above operation enables the samples in each round to be used to train the model, which can be as close to the distribution of the original samples as possible. The introduction of the validation set can also effectively avoid overfitting. Therefore, the nonlinear optimization model of the smoothing factor σ can be established according to the nonlinear optimization objective function of the smoothing factor σ:

[0090]

[0091] S22. Design an efficient algorithm to solve the nonlinear optimization model about the smoothness factor σ and obtain the optimal smoothness factor. The process of using quantum genetic algorithm to solve the optimal smoothness factor is as follows: Figure 3 shown.

[0092] The present invention designs a quantum genetic algorithm (QGA) to solve a nonlinear optimization model about a smoothing factor σ. The quantum genetic algorithm is based on quantum state vectors and genetic algorithms, integrates quantum state vector expression with genetic coding, encodes chromosomes based on the probability amplitude of quantum bits, and realizes the evolution of chromosomes with the help of quantum logic gates, thereby realizing the solution of the nonlinear optimization model and effectively avoiding local optimality.

[0093] First, the basic principles of quantum genetic algorithm are explained:

[0094] A quantum bit is a physical medium that acts as an information storage unit in a quantum computer and can be in a superposition of two quantum states at the same time. Where α and β are the probability amplitudes of the quantum state, which are a pair of complex constants and satisfy |α|+|β|=1; |0> and |1> represent the spin-down and spin-up states, respectively. Quantum gates are the execution structures of evolutionary operations in quantum genetic algorithms. They can transform the above quantum bit states to implement specific logical operations. Commonly used ones include quantum rotation gates and quantum Hadamard gates.

[0095] The QGA algorithm uses the qubit chromosome representation method. A qubit represents a gene and can express any superposition state at the same time. Based on the qubit encoding, the chromosome structure can be represented as:

[0096]

[0097] In the formula, q j represents the chromosome of the jth individual, k represents the number of quantum bits encoded for each gene, m is the number of chromosome genes, α and β are the probability amplitudes of |0> and |1>, respectively, and satisfy the normalization condition |α| 2 +|β| 2 =1,|α| 2 represents the probability that the quantum measurement value is 0, |β| 2 Represents the probability that the quantum measurement value is 1.

[0098] The update process based on the quantum revolving door is:

[0099] Among them, (α i ,β i ) and (α i ′,β i ′) represents the probability amplitude before and after the update of the i-th qubit rotating gate of the chromosome; θ i =Δθ i s(α i β i ) represents the rotation angle; Δθ i Indicates the angle of rotation; s(α i β i ) indicates the direction of rotation, which is determined by the initial conditions and other parameter strategies and is mainly used to ensure the convergence of the algorithm. The quantum update process rotates the chromosome code generated by the quantum bit code through the quantum rotating gate to search for the optimal code value.

[0100] Furthermore, the optimization steps of the smoothing factor σ are as follows:

[0101] Step 1: Set the basic parameters of the algorithm, including the population size N, the maximum number of iterations T, the binary length L of each variable, and the value range of the parameter σ to be optimized [σ min ,σ max ].

[0102] Step 2: Population initialization, set the quantum population to Where t is the current genetic generation; is a chromosome of an individual in the population, defined as is the parameter to be optimized, i.e., the quantum bit of the smoothing factor. Initialize Q(t) to get Q(t0), and set the probability amplitude of all chromosomes to

[0103] Step 3: Measure each individual in the initial population Q(t0) once to generate a binary solution set P(t0).

[0104] Step 4: Perform fitness evaluation on each determined solution in P(t0), and use the average mean square error of k-fold cross validation as the individual fitness value to reflect the prediction ability of GRNN. The fitness function formula is FD=E, where E is the average mean square error of k-fold cross validation.

[0105] Step 5: Determine whether the termination condition is met. If so, terminate; otherwise, proceed to the next step.

[0106] Step 6: Measure each individual in the population Q(t) once, generate a binary solution set P(t), and evaluate the fitness of each determined solution.

[0107] Step 7: Use the new quantum revolving door U′(t) to adjust the individuals and obtain a new population Q(t+1). Obtain P(t+1) by measuring each individual, and record the optimal individual and the corresponding fitness value based on the fitness evaluation of each determined solution.

[0108] Step 8: Increase the number of iterations by 1 and return to step 5.

[0109] The method for assessing the risk of marine environment affecting ship navigation safety under high-dimensional small sample conditions of the present invention also includes the following steps:

[0110] S3. The BR-GRNN model is used to assess the risk of the marine environment affecting the navigation safety of ships, and the risk assessment values ​​are clustered to obtain a risk zoning map of the navigation area.

[0111] In order to verify the effect of the model (BR-GRNN) of the present invention, the data in Table 7.12 on page 332 of the monograph "Diagnosis of Marine Environmental Characteristics and Risk Assessment of Maritime Military Activities" was used to verify the effectiveness of the model of the present invention (Table 1), and the risk assessment effects of BR-GRNN and traditional risk assessment methods (GRNN, fuzzy comprehensive evaluation, grey correlation analysis) under different numbers of training small samples were compared to verify the effectiveness and robustness of the BR-GRNN model under high-dimensional small sample conditions. As can be seen from Table 1, these data include marine environmental risk assessment indicators (wind speed, wave height, visibility, thunderstorm probability, low cloud cover) and corresponding risk values. These risk assessment indicators and corresponding risk value data are called high-dimensional correlation samples. The specific approach is as follows: First, select high-dimensional correlated small samples of different capacities as training data, which are 30, 20, 10 and 5 respectively, and the remaining 6, 10, 20 and 31 samples as validation data; secondly, use the BR-GRNN model to expand the training data of different capacities by 100 times, so that the original 30, 20, 10 and 5 samples become 3000, 2000, 1000 and 500 samples respectively; thirdly, based on the expanded samples, use efficient algorithms to optimize the smoothing factor of the BR-GRNN model, and obtain the optimal smoothing factor under different small sample conditions; finally, use the optimized BR-GRNN model to evaluate the risks of the remaining 6, 10, 20 and 31 samples to obtain risk assessment values, and use traditional methods (GRNN, fuzzy comprehensive evaluation, grey correlation analysis) to evaluate the risks of these samples. The comparison results of the risk assessment values ​​and actual values ​​under different small samples (30, 20, 10 and 5) are shown in the figure below. Figures 4 to 7The average relative error and mean square error between the risk assessment values ​​obtained by the four assessment models and the actual values ​​under different small sample conditions are shown in Tables 2 to 5. The average relative error and mean square error between the risk assessment values ​​obtained by the four assessment models and the actual values ​​change with the sample size as shown in Tables 2 to 5. Figure 8 and 9 shown.

[0112] Table 1 Marine environmental risk assessment test samples

[0113]

[0114]

[0115] Depend on Figure 4 It can be seen that when 30 training samples are used, the risk assessment value obtained by the BR-GRNN model is highly consistent with the experimental value, and the deviation is very small, while the GRNN model has a large deviation at the 33rd, 34th, 35th and 36th sample points. It can be seen that the BR-GRNN model is significantly better than the GRNN model under the condition of small samples; and the fuzzy comprehensive evaluation and grey correlation analysis have a large deviation at other sample points except for the 32nd sample point. As shown in Table 2, when 30 training samples are used, the average relative error and mean square error of BR-GRNN are very small, which are 63.5% to 86.1% lower than the other three models. The average relative error of BR-GRNN is only about 0.08, and the evaluation accuracy is as high as 92%, while the evaluation accuracy of the GRNN model is only 77%, and the evaluation accuracy of fuzzy comprehensive evaluation and grey correlation analysis is only about 39% and 45%. In general, the evaluation errors of fuzzy comprehensive evaluation and grey correlation analysis are large. Compared with the GRNN model, the BR-GRNN model has a very large improvement and has high accuracy and evaluation precision.

[0116] Table 2 Error comparison between the BR-GRNN model and the other three models when using 30 training samples

[0117]

[0118] Depend on Figure 5It can be seen that when 20 training samples are used, the risk assessment value obtained by the BR-GRNN model has a small deviation from the actual value, except for the large deviation at the 26th point, while the GRNN model has a large deviation at the 26th, 30th, 34th and 35th sample points. It can be seen that the BR-GRNN model is significantly better than the GRNN model under the condition of small samples; while the fuzzy comprehensive evaluation and grey correlation analysis have very large deviations at other sample points except for the small deviations at the 24th, 25th, 29th and 32nd sample points. As shown in Table 3, when 20 training samples are used, the average relative error and mean square error of BR-GRNN are very small, 42% to 65.9% lower than the other three models. The average relative error of BR-GRNN is only 0.165, and the evaluation accuracy reaches 83.5%, while the evaluation accuracy of the GRNN model is only 71.6%, and the evaluation accuracy of the fuzzy comprehensive evaluation and grey correlation analysis is only about 52% and 53%. In general, compared with the other three models, the BR-GRNN model has a great improvement and has a very high evaluation accuracy.

[0119] Table 3 Error comparison between the BR-GRNN model and the other three models when using 20 training samples

[0120]

[0121]

[0122] Depend on Figure 6 It can be seen that when 10 training samples are used, except for some deviations at the 18th, 21st and 26th points, in general, the risk assessment value obtained by the BR-GRNN model is still very consistent with the actual value, while the GRNN model has a large deviation at most of the subsequent points starting from the 23rd point. Considering that BR-GRNN only uses 10 training samples, it can be considered that the evaluation effect of the model is still good. Therefore, under the condition of small samples, the BR-GRNN model is significantly better than the GRNN model; while the fuzzy comprehensive evaluation and grey correlation analysis have large evaluation deviations at most sample points. As shown in Table 4, when 10 training samples are used, the average relative error and mean square error of BR-GRNN are very small, which are 22.7% to 54% lower than the other three models. The average relative error of BR-GRNN is only 0.215, and the evaluation accuracy rate reaches 78.5%, while the evaluation accuracy rate of the GRNN model is only 66.8%, and the evaluation accuracy rates of the fuzzy comprehensive evaluation and grey correlation analysis are only about 54% and 58%. In general, compared with the other three models, the BR-GRNN model has obvious advantages and higher evaluation accuracy.

[0123] Table 4 Error comparison between the BR-GRNN model and the other three models when using 10 training samples

[0124]

[0125] Depend on Figure 7 It can be seen that when only 5 training samples are used, the risk assessment value obtained by the BR-GRNN model has a certain deviation from the actual value, but the risk assessment value and the actual value are still well correlated, and the Pearson correlation coefficient reaches 0.623. The risk assessment value obtained by the GRNN model deviates too much from the actual value, and the correlation coefficient between the risk assessment value and the actual value is only 0.418. The evaluation effect of fuzzy comprehensive evaluation and grey correlation analysis is still very poor, but better than that of the GRNN model. As shown in Table 5, when 5 training samples are used, the average relative error and mean square error of BR-GRNN are the smallest, which is 19% to 51% lower than the other three models. The average relative error of BR-GRNN is 0.347, and the evaluation accuracy still reaches 65.3%, while the evaluation accuracy of the GRNN model is only 48.3%, and the evaluation accuracy of fuzzy comprehensive evaluation and grey correlation analysis is only about 51% and 55%. In general, considering that there are only 5 training samples, the evaluation effect of the BR-GRNN model is still acceptable, significantly better than the other three models.

[0126] Table 5 Error comparison between the BR-GRNN model and the other three models when using 5 training samples

[0127]

[0128] Figure 8 and 9The variation trend of the average relative error and mean square error of the four models under different training samples is shown. For the BR-GRNN and GRNN models, the average relative error and mean square error increase with the reduction of training samples, while the average relative error and mean square error of fuzzy comprehensive evaluation and grey relational analysis tend to decrease. The main reason is that fuzzy comprehensive evaluation and grey relational analysis do not use the information of risk value when modeling. The main process of modeling is weighting, and the weight is solved by subjective method without training samples. When different numbers of training samples are used, the BR-GRNN model is always the best, and the GRNN model ranks second. The evaluation errors of fuzzy comprehensive evaluation and grey relational analysis are similar and large. When only 5 training samples are used, the average relative error and mean square error of the BR-GRNN model are 0.347 and 0.409 respectively, which are smaller than the errors of fuzzy comprehensive evaluation and grey relational analysis under 30 samples, that is, the evaluation accuracy of the BR-GRNN model under 5 training samples still reaches 65.3%. When 10 training samples are used, the average relative error and mean square error of the BR-GRNN model are 0.215 and 0.311, respectively, which are much smaller than the other three models. When 20 and 30 training samples are used, the average relative error of the BR-GRNN model is only 0.165 and 0.085, the mean square error is only 0.197 and 0.117, and the evaluation accuracy is as high as 83.5% and 91.5%, respectively. It can be seen that when the number of training samples is greater than or equal to 20, the BR-GRNN model can obtain satisfactory evaluation results; when the number of training samples is only 5 or 10, the BR-GRNN model can still obtain an acceptable evaluation result.

[0129] In summary, the experimental results show that the BR-GRNN model of the present invention has great reliability and application potential in high-dimensional small sample scenarios:

[0130] (1) Regardless of whether the number of small samples is 30, 20, 10 or 5, the BR-GRNN model is significantly better than the GRNN model, fuzzy comprehensive evaluation and grey relational analysis.

[0131] (2) According to the calculation results of the average relative errors of different small sample situations, compared with the GRNN model, the BR-GRNN model has improved by about 32.8% to 63.5%; compared with the fuzzy comprehensive evaluation, the BR-GRNN model has improved by about 30.0% to 86.1%; compared with the grey relational analysis, the BR-GRNN model has improved by about 21.6% to 84.8%.

[0132] (3) When the sample size is between 20 and 30, the average evaluation accuracy of the BR-GRNN model is 83.5% to 91.5%; when the sample size is 10, the average evaluation accuracy of the BR-GRNN model reaches 78.5%, which is still a good result; when the sample size is 5, the average evaluation accuracy of the BR-GRNN model is 65.3%.

[0133] According to the obtained BR-GRNN model, the impact of the marine environment on the navigation safety of ocean-going ships can be evaluated, and a risk zoning map of the navigation area can be obtained, which can provide auxiliary decision-making support for ocean-going ships to avoid severe weather risks.

[0134] Embodiment 2:

[0135] The computer-readable storage medium of this embodiment stores a computer program thereon, which, when executed by a processor, implements the steps in the method for assessing the risk of the marine environment affecting the safety of ship navigation under a high-dimensional small sample situation in Example 1.

[0136] The computer-readable storage medium of this embodiment may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal; the computer-readable storage medium of this embodiment may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card, a secure digital card, a flash memory card, etc. equipped on the terminal; further, the computer-readable storage medium may also include both an internal storage unit of the terminal and an external storage device.

[0137] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or is to be output.

[0138] Embodiment 3:

[0139] The computer device of this embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for assessing the risk of marine environment impact on ship navigation safety under high-dimensional small sample conditions in Example 1 are implemented.

[0140] In this embodiment, the processor may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, readily available programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0141] Those skilled in the art will appreciate that the disclosed content of the embodiments may be provided as methods, systems, or computer program products. Therefore, the present solution may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Moreover, the present solution may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.

[0142] The present solution is described with reference to the flowchart and / or block diagram of the method and computer program product according to the embodiment of the present solution. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions; these computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0143] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0144] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0145] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0146] The examples described in the present invention are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the concept and scope of the present invention. Without departing from the design concept of the present invention, various modifications and improvements made to the technical solutions of the present invention by engineers and technicians in this field should all fall within the protection scope of the present invention.

Claims

1. A method for assessing the risk of marine environment affecting ship navigation safety under high-dimensional small sample conditions, characterized in that: The steps include: S1. Establish BR high-dimensional correlation risk sample expansion model; The associated risk sample refers to the high-dimensional risk assessment index value X and the corresponding risk value Y. The i-th high-dimensional associated risk sample is expressed as: (X i ,Y i )=(x i,1 ,x i,2 ,…,x i,n ,y i ) (1) In the formula, n represents the number of high-dimensional risk assessment indicators; S2, coupling the BR high-dimensional associated risk sample expansion model with the generalized regression neural network model, optimizing the model parameters, and obtaining the BR-GRNN risk assessment model under the high-dimensional small sample situation; S3. The BR-GRNN risk assessment model under high-dimensional small sample conditions is used to assess the risk of the marine environment affecting the navigation safety of ships, and the risk assessment values ​​are clustered to obtain a risk zoning map of the navigation area.

2. The method according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11, establishing a high-dimensional expansion index array; S12. Establish a relationship model between the high-dimensional index value X and the risk value Y, which is the expanded high-dimensional index data X new Calculate the corresponding risk value Y new .

3. The method according to claim 2, characterized in that In S11, the steps for establishing the high-dimensional expansion index array are as follows: S111, vertically cut the high-dimensional risk assessment index data and generate n labeled arrays (l1, l2, ···, l n ), let l j Represents the row number of the j-th column data, i.e. l j =(1,2,…,n) T ; Suppose the matrix of high-dimensional small sample risk assessment indicators is X, the size of X is m×n, where m≤30, and X can be expressed as: S112, bind the two columns of data with strong intra-group correlation, that is, (l1,l2,···,l n ) are combined with several indicator data with a Pearson correlation coefficient greater than or equal to 0.6 to obtain a new number array S113, using Bootstrap method Each column of is resampled, and m rows of new samples are obtained. Repeat the operation until the desired sample size is reached. The expanded new sample is recorded as X PP , X PP Simulate the strong correlation information contained in the original high-dimensional risk data, X PP The structure is shown below: S114, the original high-dimensional index data X and the expanded X PP The new data array X after merging and expansion new ; After expansion, X new The size of becomes (k+1)m×n, where k is the expansion multiple and k>2.

4. The method according to claim 2, characterized in that: In S12, the specific steps of the relationship model between the high-dimensional index value X and the risk value Y are: S121, let each high-dimensional correlation sample be (X i ,Y i )=(x i,1 ,x i,2 ,…,x i,n ,y i ), where i = 1, 2, ..., m, and these m associated samples are used as training data for the multivariate linear regression model, thereby obtaining the relationship f between the high-dimensional index value X and the risk value Y; S122, the expanded new sample X PP Input into the relation f, thus obtaining PP The corresponding risk value vector Y PP ; S123, put X PP and Y PP Merge to obtain a high-dimensional correlation sample matrix (X PP ,Y PP ), and (X PP ,Y PP ) is combined with the original high-dimensional correlated small sample matrix (X, Y) to obtain a new high-dimensional correlated sample matrix (X new ,Y new ).

5. The method according to claim 2, characterized in that: The specific steps of S2 are: S21, based on the expanded high-dimensional correlation sample matrix (X new ,Y new ) train the generalized regression neural network model and establish a nonlinear optimization model for the smoothing factor σ; thereby having sufficient training samples to optimize the parameters of the generalized regression neural network model; S22. Design an efficient algorithm to solve the nonlinear optimization model about the smoothness factor σ and obtain the optimal smoothness factor.

6. The method according to claim 5, characterized in that The specific steps of S21 are: In order to effectively avoid underfitting and overfitting in the σ optimization process, with the help of k-fold cross validation, the objective function of the nonlinear optimization model is first established. The process is as follows: For the sample data set Grouping, let the kth subset be Among them, m k is the sample size of the subset, each subset data is used as the validation set, and the remaining k-1 groups of subset data are used as the training set, thereby obtaining k models. The average mean square error of the k model validation set is used as the nonlinear optimization objective function E of the cubic smoothing factor σ. The calculation formula is shown in (7) and (8): Furthermore, the nonlinear optimization model of the smooth factor σ can be established according to the nonlinear optimization objective function of the smooth factor σ:

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method for risk assessment of the impact of marine environment on ship navigation safety in a high-dimensional small sample situation as described in any one of claims 1 to 6 are implemented.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for risk assessment of the impact of the marine environment on ship navigation safety in a high-dimensional small sample situation according to any one of claims 1 to 6 are implemented.

Images