Underwater robot operation risk prediction method, system, equipment and medium

Abstract

This application discloses a method, system, device, and medium for predicting underwater robot operation risks, relating to the field of risk prediction. The method inputs a risk value vector to be predicted into a basic risk prediction model to obtain a first risk level prediction value; utilizes a hierarchical analysis method to obtain corresponding weights, and determines a second risk level value based on the weights; utilizes a trapezoidal fuzzy membership function to calculate the risk membership of each risk factor value, and determines the risk level corresponding to the maximum risk membership as a discrete risk level value, which is input into a risk level fuzzy mapping model to obtain a third risk level prediction value; calculates the number of different risk levels and inputs this into a risk prediction model based on risk level statistics to output a fourth risk level prediction value; and utilizes a particle swarm algorithm to determine the weights of the four risk level prediction values, thereby determining the comprehensive risk value of the underwater robot operation. This application achieves higher-accuracy underwater robot operation risk prediction using limited data.

Description

Technical Field

[0001] The present application relates to the technical field of risk prediction, and in particular to a method, system, equipment and medium for predicting risks in underwater robot operations. Background Art

[0002] In the academic field of underwater robot risk prediction, the current mainstream methods can be summarized into three categories: Bayesian networks, Markov chain models, and system dynamics modeling and analysis. While these methods can provide a certain degree of offline preliminary assessment of underwater robot risk status, the complexity of their implementation mechanisms, the intensiveness of computing resources, and the lack of real-time and application convenience significantly restrict their ability to meet the demand for immediate risk prediction systems in complex underwater operating environments.

[0003] Given the high risk and irreversible potential losses associated with underwater robot missions, building a risk prediction system capable of real-time response and high prediction accuracy is particularly urgent. Deep neural networks clearly possess the ability to perform real-time assessments, but as the number of neural network nodes increases, more data is required to achieve better predictions. However, datasets on underwater robot operational risk are scarce, and this scarcity of data is a key factor limiting model performance. Therefore, effective risk prediction strategies that reduce data dependency are needed to alleviate the data shortage issue and further enhance the effectiveness and reliability of risk prediction systems in practical applications. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, equipment and medium for predicting the risks of underwater robot operations, which can achieve more accurate dynamic prediction of underwater robot operation risks under the premise of limited data volume.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for predicting the risk of underwater robot operations, the method comprising:

[0007] Obtaining a risk value vector to be predicted of the underwater robot; the risk value vector to be predicted includes multiple risk factor values;

[0008] Inputting the risk value vector to be predicted into a basic risk prediction model to obtain a first risk level prediction value; the basic risk prediction model is obtained by training a fully connected neural network model using a first training set, the first training set including a sample risk value vector and a first risk level sample value;

[0009] Using the analytic hierarchy process, expert scoring is performed on each risk factor value in the risk value vector to be predicted to obtain a weight corresponding to each risk factor value, and based on the weight corresponding to each risk factor value and a risk averaging model, a second risk level value corresponding to the risk value vector to be predicted is determined;

[0010] Using a trapezoidal fuzzy membership function, respectively calculating the risk membership corresponding to each risk factor value in the risk value vector to be predicted, and determining the risk level corresponding to the maximum risk membership of each risk factor value as the discrete risk level value of the corresponding risk factor value;

[0011] Inputting the discrete risk level values ​​of each risk factor value into a risk level fuzzy mapping model to obtain a third risk level prediction value for each risk factor value; the risk level fuzzy mapping model is obtained by training the fully connected neural network model using a second training set, wherein the second training set includes discrete risk level values ​​of sample risk factor values ​​and corresponding third risk level sample values;

[0012] Counting the number of different risk levels in the third risk level prediction value of each risk factor value, and inputting the number of different risk levels into a risk prediction model based on risk level quantity statistics, and outputting a fourth risk level prediction value; the risk prediction model based on risk level quantity statistics is obtained by training the fully connected neural network model using a third training set, and the third training set includes the number of different risk levels in the third risk level sample values ​​of the sample risk factor values ​​and the corresponding fourth risk level sample values;

[0013] The particle swarm algorithm is used to determine the weights of the first risk level prediction value, the second risk level prediction value, the third risk level prediction value and the fourth risk level prediction value, so as to determine the comprehensive operation risk value of the underwater robot.

[0014] Optionally, the risk value vector to be predicted includes ocean current risk factor value, temperature risk factor value, water density risk factor value, reef risk factor value, aquatic plant risk factor value, swimming fish risk factor value, power system risk factor value, power system risk factor value, operation distance risk factor value and operation depth risk factor value.

[0015] Optionally, the fully connected neural network model includes an input layer, a hidden layer and an output layer connected in sequence.

[0016] Optionally, the training process of the basic risk prediction model specifically includes:

[0017] Inputting the sample risk value vector into the fully connected neural network model to obtain a predicted value of the first risk level of the sample;

[0018] A first loss function is constructed based on the first risk level prediction value of the sample and the first risk level sample value, and the network parameters of the fully connected neural network model are iteratively optimized according to the first loss function until the iterative optimization round reaches the maximum value or the first loss function reaches the minimum value, and the iterative optimization is stopped to obtain the basic risk prediction model.

[0019] Optionally, the analytic hierarchy process is used to perform expert scoring on each risk factor value in the risk value vector to be predicted, and the weight corresponding to each risk factor value is obtained, specifically including:

[0020] Using the analytic hierarchy process, experts score each risk factor value in the risk value vector to be predicted to obtain a scoring matrix;

[0021] Performing eigendecomposition on the scoring matrix to obtain the maximum eigenvalue of the scoring matrix;

[0022] Performing a consistency check on the scoring matrix based on the maximum eigenvalue of the scoring matrix to obtain a consistency ratio;

[0023] Obtaining a random consistency index, and obtaining a test coefficient based on the random consistency index and the consistency ratio;

[0024] When the test coefficient is less than a preset threshold, the eigenvector corresponding to the maximum eigenvalue of the scoring matrix is ​​calculated;

[0025] The eigenvector is normalized and used as the weight corresponding to each risk factor value.

[0026] Optionally, based on the weights corresponding to the risk factor values ​​and the risk averaging model, determining the second risk level value corresponding to the risk value vector to be predicted specifically includes:

[0027] The weights corresponding to each risk factor value and the risk factor value are weighted averaged to obtain the weighted risk value;

[0028] The risk averaging model interval corresponding to the weighted risk value is determined, and a second risk level value corresponding to the risk value vector to be predicted is determined.

[0029] Optionally, when the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A1=(0, 0.1, 0.2, 0.3), the risk level corresponding to the maximum risk membership of the risk factor value is low risk, and the discrete risk level value of the risk factor value is 1;

[0030] When the maximum risk membership corresponding to the risk factor value falls within the range corresponding to A2=(0.25, 0.35, 0.45, 0.55), the risk level corresponding to the maximum risk membership of the risk factor value is low risk, and the discrete risk level value of the risk factor value is 2;

[0031] When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A3=(0.5, 0.6, 0.7, 0.8), the risk level corresponding to the maximum risk membership of the risk factor value is high risk, and the discrete risk level value of the risk factor value is 3;

[0032] When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A4=(0.75, 0.85, 0.95, 1), the risk level corresponding to the maximum risk membership of the risk factor value is high risk, and the discrete risk level value of the risk factor value is 4.

[0033] In a second aspect, the present application provides an underwater robot operation risk prediction system, which is used to implement the underwater robot operation risk prediction method, and the underwater robot operation risk prediction system includes:

[0034] A data acquisition unit is used to acquire a risk value vector to be predicted of the underwater robot; the risk value vector to be predicted includes multiple risk factor values;

[0035] a first risk level prediction value determination unit, configured to input the risk value vector to be predicted into a basic risk prediction model to obtain a first risk level prediction value; the basic risk prediction model is obtained by training a fully connected neural network model using a first training set, the first training set including a sample risk value vector and a first risk level sample value;

[0036] a second risk level value determination unit, configured to perform expert scoring on each risk factor value in the risk value vector to be predicted using an analytic hierarchy process to obtain a weight corresponding to each risk factor value, and determine a second risk level value corresponding to the risk value vector to be predicted based on the weight corresponding to each risk factor value and a risk averaging model;

[0037] a discrete risk level value determination unit, configured to calculate the risk membership corresponding to each risk factor value in the risk value vector to be predicted using a trapezoidal fuzzy membership function, and determine the risk level corresponding to the maximum risk membership of each risk factor value as the discrete risk level value of the corresponding risk factor value;

[0038] a third risk level prediction value determination unit, configured to input the discrete risk level values ​​of each risk factor value into a risk level fuzzy mapping model to obtain a third risk level prediction value for each risk factor value; the risk level fuzzy mapping model is obtained by training the fully connected neural network model using a second training set, the second training set including discrete risk level values ​​of sample risk factor values ​​and corresponding third risk level sample values;

[0039] a fourth risk level prediction value determination unit, configured to count the number of different risk levels in the third risk level prediction value of each risk factor value, input the number of different risk levels into a risk prediction model based on risk level quantity statistics, and output a fourth risk level prediction value; the risk prediction model based on risk level quantity statistics is obtained by training the fully connected neural network model using a third training set, the third training set including the number of different risk levels in the third risk level sample values ​​of the sample risk factor values ​​and the corresponding fourth risk level sample values;

[0040] The comprehensive operation risk value determination unit is used to use the particle swarm algorithm to determine the weight of the first risk level prediction value, the weight of the second risk level value, the weight of the third risk level prediction value and the weight of the fourth risk level prediction value, thereby determining the comprehensive operation risk value of the underwater robot.

[0041] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned underwater robot operation risk prediction methods.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned underwater robot operation risk prediction methods.

[0043] According to the specific embodiments provided in this application, this application has the following technical effects:

[0044] This application discloses a method, system, device, and medium for predicting underwater robot operation risks. This method, firstly, enhances the model's interpretability and prediction accuracy compared to single data-driven risk prediction methods by fusing expert prior knowledge with data. Secondly, expert knowledge is extracted as a feature model, and the evaluation results of each model are integrated through ensemble learning. This can address the problem of underwater robot risk prediction datasets being small and unable to drive deep network models. Finally, a particle swarm algorithm is used to optimize the weights of each risk level prediction value, further improving the accuracy of the corresponding risk level prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 A schematic diagram of a process flow for predicting underwater robot operation risks according to an embodiment of the present application;

[0047] Figure 2 A schematic diagram of a fully connected neural network model provided in one embodiment of the present application;

[0048] Figure 3 A schematic diagram of a trapezoidal fuzzy membership function provided in one embodiment of the present application;

[0049] Figure 4 A schematic diagram of the particle swarm weight optimization iterative process provided in one embodiment of the present application;

[0050] Figure 5 Schematic diagram of the change in loss value of the basic risk prediction model;

[0051] Figure 6 Schematic diagram of the changes in the evaluation accuracy of the basic risk prediction model;

[0052] Figure 7 Schematic diagram of the change of loss value in the evaluation process after particle swarm optimization;

[0053] Figure 8 This is a schematic diagram of the change in evaluation accuracy after particle swarm optimization;

[0054] Figure 9 A schematic diagram of the functional modules of an underwater robot operation risk prediction system provided in one embodiment of the present application;

[0055] Figure 10 A schematic diagram of the structure of a computer device provided in one embodiment of the present application.

[0056] Reference numerals:

[0057] Data acquisition unit 1, first risk level prediction value determination unit 2, second risk level value determination unit 3, discrete risk level value determination unit 4, third risk level prediction value determination unit 5, fourth risk level prediction value determination unit 6, comprehensive operation risk value determination unit 7. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0059] In response to the problem of scarcity and small amount of existing underwater robot operation risk data, this application integrates the knowledge and experience of domain experts into the data model to achieve the fusion of data and prior knowledge, and improves the risk prediction accuracy of the underwater robot operation risk prediction model driven by small sample data with less data volume. Specifically, the expert knowledge adopted in this application includes risk weights based on the hierarchical analysis method, risk level fuzzy mapping, and risk level quantitative statistics. The above three types of expert knowledge are integrated into the basic neural network model to form three risk prediction models. The idea of ​​ensemble learning is adopted, and the four underwater robot operation risk prediction level values ​​such as the basic risk level prediction model directly trained with the original data are assigned certain weights. The corresponding weight distribution is optimized through the particle swarm algorithm, thereby achieving an improvement in the accuracy of the risk prediction method.

[0060] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0061] In an exemplary embodiment, Figure 1 As shown, a method for predicting the risk of underwater robot operations is provided, comprising the following steps S1 to S7.

[0062] Step S1, obtaining a risk value vector to be predicted of the underwater robot; the risk value vector to be predicted includes multiple risk factor values.

[0063] Specifically, the risk value vector to be predicted includes ocean current risk factor value, temperature risk factor value, water density risk factor value, reef risk factor value, aquatic plant risk factor value, swimming fish risk factor value, power system risk factor value, power system risk factor value, operation distance risk factor value and operation depth risk factor value.

[0064] Step S2: input the risk value vector to be predicted into the basic risk prediction model to obtain a first risk level prediction value; the basic risk prediction model is obtained by training a fully connected neural network model using a first training set, and the first training set includes a sample risk value vector and a first risk level sample value.

[0065] As an optional implementation, in step S2, the fully connected neural network model includes an input layer, a hidden layer, and an output layer connected in sequence. The schematic diagram of the fully connected neural network model is as follows: Figure 2 As shown. Among them, the hidden layer uses the following formula as the activation function:

[0066] ReLU(y hid )=max(0,y hid ) (1)

[0067] Among them, max is a function that finds the maximum value of multiple variables; y hid is the output of the neurons in the previous layer of neural network.

[0068] The output layer uses the following formula as the probability estimate of the four risk levels of low risk, lower risk, higher risk and high risk:

[0069]

[0070] Among them, y o is the output of the output layer neural network neurons; y oi is the neural network output corresponding to each risk level, and i takes 1, 2, 3, and 4, corresponding to low risk, lower risk, higher risk, and high risk, respectively.

[0071] As an optional implementation, in step S2, the training process of the basic risk prediction model specifically includes:

[0072] Step S21: input the sample risk value vector into the fully connected neural network model to obtain a sample first risk level prediction value.

[0073] Step S22: construct a first loss function based on the first risk level prediction value of the sample and the first risk level sample value, and iteratively optimize the network parameters of the fully connected neural network model according to the first loss function until the iterative optimization round reaches a maximum value or the first loss function reaches a minimum value, stop the iterative optimization, and obtain the basic risk prediction model.

[0074] Specifically, the first training set is a small sample data set of existing underwater robot operation risk prediction. Each sample in the small sample data set contains a ten-dimensional input (i.e., sample risk value vector) and a one-dimensional output (i.e., first risk level sample value). The ten-dimensional input includes ocean current risk factor value, temperature risk factor value, water density risk factor value, reef risk factor value, aquatic plant risk factor value, swimming fish risk factor value, power system risk factor value, power system risk factor value, operation distance risk factor value, and operation depth risk factor value; in the output first risk level sample value, 1 represents low risk, 2 represents lower risk, 3 represents higher risk, and 4 represents high risk.

[0075] Step S3, using the hierarchical analysis method, experts score each risk factor value in the risk value vector to be predicted, obtain the weight corresponding to each risk factor value, and determine the second risk level value corresponding to the risk value vector to be predicted based on the weight corresponding to each risk factor value and the risk averaging model.

[0076] As an optional implementation, in step S3, the risk factor values ​​in the risk value vector to be predicted are scored by experts using the analytic hierarchy process to obtain the weight corresponding to each risk factor value, specifically including:

[0077] Step S31 , using the analytic hierarchy process, experts score each risk factor value in the risk value vector to be predicted to obtain a scoring matrix.

[0078] Experts in the field of underwater robot operations were asked to compare ten risk factors, including ocean currents, temperature, reefs, aquatic plants, fish, electricity, power, operating distance, and operating depth, in pairs. Scores were given according to the standard nine-level grading table of the hierarchical analysis method. The scoring results of each expert will form a 10-order square matrix. The mode of the elements in the same position in each 10-order square matrix is ​​taken as the final scoring result, and a 10-order square matrix is ​​obtained, which is the scoring matrix. Among them, the elements a are symmetrical about the main diagonal. ij a ji =1, and the matrix main diagonal element a ii = 0. For example, if the importance of the ocean current risk factor value relative to the temperature risk factor value is 3, then a in the matrix 12 =3, The importance of the temperature risk factor value to the reef risk factor value is but a 32 =2.

[0079] Step S32: Perform eigendecomposition on the scoring matrix to obtain the maximum eigenvalue of the scoring matrix. The maximum eigenvalue is obtained by calculating all eigenvalues ​​of the scoring matrix and taking the maximum value.

[0080] Step S33: Based on the maximum eigenvalue of the scoring matrix, a consistency check is performed on the scoring matrix to obtain a consistency ratio; the consistency ratio is calculated according to the following formula:

[0081]

[0082] Among them, λ is the maximum eigenvalue of the scoring matrix, and n is the order of the scoring matrix.

[0083] Step S34: Obtain a random consistency index, and obtain a test coefficient based on the random consistency index and the consistency ratio.

[0084] Considering that the increase in CI value may be caused by the increase in random factors due to the increase in matrix order, when testing whether the scoring matrix has satisfactory consistency, it is also necessary to compare CI with the random consistency index RI to obtain the test coefficient CR, which is calculated as follows:

[0085]

[0086] In the formula, the RI value is related to the order of the scoring matrix, and its relationship with the order n of the scoring matrix can be determined by looking up the standard RI value table.

[0087] Step S35: When the test coefficient is less than a preset threshold, the eigenvector corresponding to the maximum eigenvalue of the scoring matrix is ​​calculated.

[0088] If the CRCR exceeds the preset threshold of 0.1, the decision experts need to revise their initial judgment. After performing all necessary pairwise comparisons and revisions, if the preset threshold CR is below 0.1, the consistency test is passed. The eigenvalues ​​of the scoring matrix are then calculated, the maximum eigenvalue is taken, and the corresponding eigenvector is found.

[0089] Step S36: normalize the eigenvectors and use the normalized eigenvectors as weights corresponding to the risk factor values.

[0090] As an optional implementation, in step S3, based on the weights corresponding to the risk factor values ​​and the risk averaging model, determining the second risk level value corresponding to the risk value vector to be predicted specifically includes:

[0091] Step S37, performing weighted averaging on the weights corresponding to the risk factor values ​​and the risk factor values ​​to obtain a weighted risk value;

[0092] Step S38: determine the risk averaging model interval corresponding to the weighted risk value, and determine the second risk level value corresponding to the risk value vector to be predicted.

[0093] The normalized eigenvector is used as the weight w1 corresponding to each risk factor value. After obtaining the weight w1, each risk factor value is weighted and summed in a linear weighted manner. That is, for a risk vector r composed of ten risk factor values, its weighted risk value is w1 Tr. After obtaining the final weighted risk value, the second risk level is determined according to the risk averaging model. That is, if the weighted risk value is in the interval [0, 0.25], the second risk level is low risk; if the weighted risk value is in the interval (0.25, 0.5], the second risk level is relatively low risk; if the weighted risk value is in the interval (0.5, 0.75], the second risk level is relatively high risk; if the weighted risk value is in the interval (0.75, 1], the second risk level is high risk.

[0094] Step S4, using a trapezoidal fuzzy membership function, respectively calculates the risk membership corresponding to each risk factor value in the risk value vector to be predicted, and determines the risk level corresponding to the maximum risk membership of each risk factor value as the discrete risk level value of the corresponding risk factor value.

[0095] As an optional implementation, in step S4, the trapezoidal fuzzy membership function is as follows: Figure 3 As shown, when the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A1=(0, 0.1, 0.2, 0.3), the risk level corresponding to the maximum risk membership of the risk factor value is low risk, and the discrete risk level value of the risk factor value is 1.

[0096] When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A2=(0.25, 0.35, 0.45, 0.55), the risk level corresponding to the maximum risk membership of the risk factor value is low risk, and the discrete risk level value of the risk factor value is 2.

[0097] When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A3=(0.5, 0.6, 0.7, 0.8), the risk level corresponding to the maximum risk membership of the risk factor value is high risk, and the discrete risk level value of the risk factor value is 3.

[0098] When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A4=(0.75, 0.85, 0.95, 1), the risk level corresponding to the maximum risk membership of the risk factor value is high risk, and the discrete risk level value of the risk factor value is 4.

[0099] Specifically, the trapezoidal fuzzy membership function can be expressed as It needs to satisfy 0≤α≤l≤m≤β, where α is The lower boundary of The upper boundary of [l, m] is the relative most likely value interval. For the fuzzy mapping of the risk level of each risk factor, a comment set A = {low risk, lower risk, higher risk, high risk} is established. The representation can be expressed as: A1=(0,0.1,0.2,0.3), A2=(0.25,0.35,0.45,0.55), A3=(0.5,0.6,0.7,0.8), A4=(0.75,0.85,0.95,1).

[0100] After the trapezoidal fuzzy membership function is determined, the membership of each risk factor value in the ten risk factors is calculated to belong to low risk, relatively low risk, relatively high risk and high risk respectively. According to the maximum membership principle, that is, the current risk value is sorted and the maximum value is taken for the membership of each risk level. The risk level corresponding to the risk value of each of the ten risk factors is determined according to the risk level corresponding to the maximum membership value. After determining the risk level, the risk value corresponding to each risk factor value in the ten-dimensional input is mapped to discrete risk level values: 1, 2, 3, 4, corresponding to low risk, relatively low risk, relatively high risk and high risk respectively.

[0101] Step S5: Input the discrete risk level value of each risk factor value into the risk level fuzzy mapping model to obtain the third risk level prediction value of each risk factor value; the risk level fuzzy mapping model is obtained by training the fully connected neural network model using the second training set, and the second training set includes the discrete risk level values ​​of the sample risk factor values ​​and the corresponding third risk level sample values.

[0102] Step S6, counting the number of different risk levels in the third risk level prediction value of each risk factor value, and inputting the number of different risk levels into the risk prediction model based on risk level quantity statistics, and outputting the fourth risk level prediction value; the risk prediction model based on risk level quantity statistics is obtained by training the fully connected neural network model using a third training set, and the third training set includes the number of different risk levels in the third risk level sample value of the sample risk factor value and the corresponding fourth risk level sample value.

[0103] Specifically, the number of each risk level is also an important priori feature. Based on the third risk level prediction value of each risk factor value obtained in step S5, the number of each risk level after fuzzy mapping is counted. The number of level statistics corresponding to the ten-dimensional risk factors, that is, the number of low-risk risk factors, the number of relatively low-risk risk factors, the number of relatively high-risk risk factors, and the number of high-risk risk factors are used as input. Figure 2 The fully connected neural network model shown is trained with parameters to obtain a risk prediction model based on the quantitative statistics of risk levels.

[0104] In step S7, a particle swarm algorithm is used to determine the weights of the first risk level prediction value, the second risk level prediction value, the third risk level prediction value, and the fourth risk level prediction value, thereby determining the comprehensive operation risk value of the underwater robot.

[0105] After the four risk features are extracted using the above steps S2-S6, the idea of ​​ensemble learning combined prediction is adopted to treat the result of each feature prediction as an independent unit, assign a specific weight, and optimize the weight through the particle swarm algorithm to improve the accuracy of the corresponding results. The particle swarm weight optimization iterative process diagram is shown in the figure. Figure 4 shown.

[0106] In the particle swarm algorithm, each particle as a basic optimization unit includes two attributes: position p i and speed v i , that is, each particle in the group has its own velocity vector and position vector. In this application, the position vector is mapped to the weight w of each risk level using the following formula: i :

[0107]

[0108] Where q is the dimension of the position vector x, i is an integer and i∈[1,n], where q is 4, which is the four different risk level prediction methods in steps S2-S6 above.

[0109] After the mapping is completed, in the 10-dimensional search space, the particle swarm size is 50. Among these 50 particles, the position vector of the i-th particle at time k in the n-dimensional space is p i (k)=(p i1 (k),p i2 (k)...p i4 (k)), the particle has a velocity vector v i (k)=(v i1 (k),v i2 (k)...v i4 (k)), the optimal position searched by the particle itself is p ib =(p ib1 ,p ib2 ...p ib4 ), the global optimal position of all particles is p best =(p b1 ,p b2 …p b4 ), the speed and position update formulas of the i-th particle used in this application are formulas (6) and (7) respectively.

[0110] v i (k+1)=0.6v i(k)+0.3r1(p best -p i (k))+0.5r2(p ib -p i (k))(6)

[0111] p i (k+1)=p i (k)+v i (k+1) (7)

[0112] Among them, r1 and r2 are n-dimensional vectors composed of random numbers, r 1i ∈[0,1],r 2i ∈[0,1]. The optimization goal of the particle swarm in this application is to minimize the following formula:

[0113]

[0114] in, is the risk level output by the model under the current weight, y i is the actual risk level in the dataset. The maximum number of iterations during iterative updating is 100, the velocity range constraint is -0.5≤v≤0.5, and the position constraint is 0.05≤x≤1.

[0115] A specific embodiment is provided below to further illustrate the underwater robot operation risk prediction method proposed in this application.

[0116] Step 1: This example uses an existing small-scale dataset for underwater robot operation risk prediction. The dataset contains 390 samples, including 118 high-risk samples, 92 relatively high-risk samples, 87 relatively low-risk samples, and 93 low-risk samples. The input risk value vector of the first sample in the underwater robot operation risk dataset is taken as the risk value vector to be predicted, and is recorded as r = [0.579, 0.199, 0.012, 0.191, 0.02, 0.378, 0.213, 0.059, 0.007, 0.96] T .

[0117] Step 2: Input the risk value vector r to be predicted into the basic risk prediction model to obtain the first risk level prediction value. Among them, for the trained basic risk prediction model, the final evaluation accuracy is about 80% under the premise of ensuring generalization performance. The loss value change diagram is as follows: Figure 5 As shown, the evaluation accuracy diagram is as follows Figure 6 shown.

[0118] Step 3, according to the steps of the hierarchical analysis method, invite experts in the field of underwater robot operations to complete the pairwise comparison of the ten risk factors, and score them according to the standard nine-level grading table of the hierarchical analysis method. The scoring results of each expert will form a 10-order square matrix, and the mode of the elements in the same position of each 10-order square matrix will be the final scoring matrix. The scoring matrix is ​​then subjected to a consistency test. In this embodiment, λ is calculated to be 10.3, n is 10, the RI value is 1.49, and the calculated consistency ratio CR value is 0.034, which does not exceed the threshold of 0.1 and can pass the consistency test. The eigenvector is calculated based on the maximum eigenvalue and normalized to obtain the weight vector of the risk factor:

[0119] w1=[0.092,0.031,0.04,0.075,0.151,0.151,0.075,0.225,0.094,0.064] T .

[0120] After obtaining the weights, the ten risk factors in a risk sample are linearly weighted to form a ten-digit input vector and the weight vector for inner product to obtain the final risk value, thereby determining the risk level. Taking the risk value vector r to be predicted as an example, the weighted calculation is w T ·r, the risk value is 0.148. According to the uniform distribution, the risk value in the interval [0, 0.25] is low risk, the risk value in (0.25, 0.5] is relatively low risk, the risk value in (0.5, 0.75] is relatively high risk, and the risk value in (0.75, 1] ​​is high risk. The second risk level value of this risk sample is determined to be low risk.

[0121] Step 4: The trapezoidal fuzzy membership function established in this embodiment is as follows: Figure 3 As shown, the risk factor with low risk after risk value mapping is labeled 1, the risk factor with relatively low risk after risk value mapping is labeled 2, the risk factor with low risk after higher risk value mapping is labeled 3, and the risk factor with high risk after risk value mapping is labeled 4.

[0122] After the trapezoidal fuzzy membership function is determined, the membership degrees of low risk, relatively low risk, relatively high risk, and high risk are calculated for the risk value corresponding to each risk factor. The final risk level of each risk factor risk data is determined based on the maximum membership principle. Taking the ocean current risk factor in the risk value vector r to be predicted as an example, its ocean current risk value is 0.579, and its membership degrees for the four risk levels are 0, 0, 1, and 0. Its maximum membership degree belongs to relatively high risk, and the risk level is labeled 3. After being mapped to discrete risk levels, it is input into the risk level fuzzy mapping model to obtain the third risk level prediction value for each risk factor value.

[0123] In step 5, based on the risk level fuzzy mapping in step 4, the number of risk levels after fuzzy mapping is counted. The four-dimensional data of the number of risk factors corresponding to the ten-dimensional risk factors, i.e., the number of low-risk risk factors, the number of relatively low-risk risk factors, the number of relatively high-risk risk factors, and the number of high-risk risk factors, are used as input to a risk prediction model based on risk level statistics, and the fourth risk level prediction value is output.

[0124] Step 6: Use the particle swarm optimization algorithm to determine the weights of the first risk level prediction value, the second risk level prediction value, the third risk level prediction value, and the fourth risk level prediction value, thereby determining the comprehensive operation risk value of the underwater robot. Figure 7 and Figure 8 As shown, the model prediction accuracy is 95%. The weight vector finally obtained in this embodiment is w = [0.163, 0.174, 0.632, 0.032] T .

[0125] Using the dataset corresponding to this embodiment to test each model in the above steps, the model prediction accuracy reached 90%, which is 15% higher than the basic neural network sub-model. This example shows that this application improves the accuracy of underwater robot operation risk prediction.

[0126] Beneficial effects of this application:

[0127] 1. The underwater robot operation risk prediction method provided in this application enhances the interpretability and prediction accuracy of the model compared to a single data-driven risk prediction method by integrating experts' prior knowledge with data.

[0128] 2. The underwater robot operation risk prediction method provided in this application extracts expert knowledge into a feature model and integrates the evaluation results of each model through integrated learning. It can solve the problem that the underwater robot risk prediction data set is small and cannot drive the deep network model.

[0129] 3. The underwater robot operation risk prediction method provided in this application is real-time and practical compared with previous evaluation methods, and can be actually applied to the actual operation environment of underwater robots.

[0130] Based on the same inventive concept, embodiments of the present application also provide an underwater robot operation risk prediction system for implementing the aforementioned underwater robot operation risk prediction method. The solution provided by this system is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the underwater robot operation risk prediction system provided below can be found in the limitations of the underwater robot operation risk prediction method described above and will not be repeated here.

[0131] In an exemplary embodiment, Figure 9 As shown, a system for predicting the risk of underwater robot operations is provided, including: a data acquisition unit 1, a first risk level prediction value determination unit 2, a second risk level value determination unit 3, a discrete risk level value determination unit 4, a third risk level prediction value determination unit 5, a fourth risk level prediction value determination unit 6 and a comprehensive operation risk value determination unit 7.

[0132] The data acquisition unit 1 is used to acquire a risk value vector to be predicted of the underwater robot; the risk value vector to be predicted includes multiple risk factor values.

[0133] The first risk level prediction value determination unit 2 is used to input the risk value vector to be predicted into the basic risk prediction model to obtain the first risk level prediction value; the basic risk prediction model is obtained by training the fully connected neural network model using the first training set, and the first training set includes a sample risk value vector and a first risk level sample value.

[0134] The second risk level value determination unit 3 is used to use the hierarchical analysis method to perform expert scoring on each risk factor value in the risk value vector to be predicted, obtain the weight corresponding to each risk factor value, and determine the second risk level value corresponding to the risk value vector to be predicted based on the weight corresponding to each risk factor value and the risk averaging model.

[0135] The discrete risk level value determination unit 4 is used to use a trapezoidal fuzzy membership function to calculate the risk membership corresponding to each risk factor value in the risk value vector to be predicted, and determine the risk level corresponding to the maximum risk membership of each risk factor value as the discrete risk level value of the corresponding risk factor value.

[0136] The third risk level prediction value determination unit 5 is used to input the discrete risk level value of each risk factor value into the risk level fuzzy mapping model to obtain the third risk level prediction value of each risk factor value; the risk level fuzzy mapping model is obtained by training the fully connected neural network model using the second training set, and the second training set includes the discrete risk level values ​​of the sample risk factor values ​​and the corresponding third risk level sample values.

[0137] The fourth risk level prediction value determination unit 6 is used to count the number of different risk levels in the third risk level prediction value of each risk factor value, and input the number of different risk levels into the risk prediction model based on risk level quantity statistics, and output the fourth risk level prediction value; the risk prediction model based on risk level quantity statistics is obtained by training the fully connected neural network model using a third training set, and the third training set includes the number of different risk levels in the third risk level sample value of the sample risk factor value and the corresponding fourth risk level sample value.

[0138] The comprehensive operation risk value determination unit 7 is used to use the particle swarm algorithm to determine the weight of the first risk level prediction value, the weight of the second risk level value, the weight of the third risk level prediction value and the weight of the fourth risk level prediction value, thereby determining the comprehensive operation risk value of the underwater robot.

[0139] In an exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for predicting risks of underwater robot operations.

[0140] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for predicting risks of underwater robot operations is implemented.

[0141] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 10 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting the risk of underwater robot operations is implemented.

[0142] Those skilled in the art will understand that Figure 10The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0143] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0144] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0145] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0146] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0147] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for predicting underwater robot operation risks, characterized in that: The underwater robot operation risk prediction method includes: Obtaining a risk value vector to be predicted of the underwater robot; the risk value vector to be predicted includes multiple risk factor values; Inputting the risk value vector to be predicted into a basic risk prediction model to obtain a first risk level prediction value; the basic risk prediction model is obtained by training a fully connected neural network model using a first training set, the first training set including a sample risk value vector and a first risk level sample value; Using the analytic hierarchy process, expert scoring is performed on each risk factor value in the risk value vector to be predicted to obtain a weight corresponding to each risk factor value, and based on the weight corresponding to each risk factor value and a risk averaging model, a second risk level value corresponding to the risk value vector to be predicted is determined; Using a trapezoidal fuzzy membership function, respectively calculating the risk membership corresponding to each risk factor value in the risk value vector to be predicted, and determining the risk level corresponding to the maximum risk membership of each risk factor value as the discrete risk level value of the corresponding risk factor value; Inputting the discrete risk level values ​​of each risk factor value into a risk level fuzzy mapping model to obtain a third risk level prediction value for each risk factor value; the risk level fuzzy mapping model is obtained by training the fully connected neural network model using a second training set, wherein the second training set includes discrete risk level values ​​of sample risk factor values ​​and corresponding third risk level sample values; Counting the number of different risk levels in the third risk level prediction value of each risk factor value, and inputting the number of different risk levels into a risk prediction model based on risk level quantity statistics, and outputting a fourth risk level prediction value; the risk prediction model based on risk level quantity statistics is obtained by training the fully connected neural network model using a third training set, and the third training set includes the number of different risk levels in the third risk level sample values ​​of the sample risk factor values ​​and the corresponding fourth risk level sample values; The particle swarm algorithm is used to determine the weights of the first risk level prediction value, the second risk level prediction value, the third risk level prediction value and the fourth risk level prediction value, so as to determine the comprehensive operation risk value of the underwater robot.

2. The underwater robot operation risk prediction method according to claim 1, characterized in that: The risk value vector to be predicted includes ocean current risk factor value, temperature risk factor value, water density risk factor value, reef risk factor value, aquatic plant risk factor value, swimming fish risk factor value, power system risk factor value, power system risk factor value, operation distance risk factor value and operation depth risk factor value.

3. The underwater robot operation risk prediction method according to claim 1, characterized in that: The fully connected neural network model includes an input layer, a hidden layer and an output layer that are connected in sequence.

4. The underwater robot operation risk prediction method according to claim 3, characterized in that: The training process of the basic risk prediction model specifically includes: Inputting the sample risk value vector into the fully connected neural network model to obtain a predicted value of the first risk level of the sample; A first loss function is constructed based on the first risk level prediction value of the sample and the first risk level sample value, and the network parameters of the fully connected neural network model are iteratively optimized according to the first loss function until the iterative optimization round reaches the maximum value or the first loss function reaches the minimum value, and the iterative optimization is stopped to obtain the basic risk prediction model.

5. The underwater robot operation risk prediction method according to claim 1, characterized in that: Using the analytic hierarchy process, each risk factor value in the risk value vector to be predicted is scored by experts to obtain the weight corresponding to each risk factor value, specifically including: Using the analytic hierarchy process, experts score each risk factor value in the risk value vector to be predicted to obtain a scoring matrix; Performing eigendecomposition on the scoring matrix to obtain the maximum eigenvalue of the scoring matrix; Performing a consistency check on the scoring matrix based on the maximum eigenvalue of the scoring matrix to obtain a consistency ratio; Obtaining a random consistency index, and obtaining a test coefficient based on the random consistency index and the consistency ratio; When the test coefficient is less than a preset threshold, the eigenvector corresponding to the maximum eigenvalue of the scoring matrix is ​​calculated; The eigenvector is normalized and used as the weight corresponding to each risk factor value.

6. The underwater robot operation risk prediction method according to claim 5, characterized in that: Determining the second risk level value corresponding to the risk value vector to be predicted based on the weights corresponding to the risk factor values ​​and the risk averaging model specifically includes: The weights corresponding to each risk factor value and the risk factor value are weighted averaged to obtain the weighted risk value; The risk averaging model interval corresponding to the weighted risk value is determined, and a second risk level value corresponding to the risk value vector to be predicted is determined.

7. The underwater robot operation risk prediction method according to claim 1, characterized in that: When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A1=(0, 0.1, 0.2, 0.3), the risk level corresponding to the maximum risk membership of the risk factor value is low risk, and the discrete risk level value of the risk factor value is 1; When the maximum risk membership corresponding to the risk factor value falls within the range corresponding to A2=(0.25, 0.35, 0.45, 0.55), the risk level corresponding to the maximum risk membership of the risk factor value is low risk, and the discrete risk level value of the risk factor value is 2; When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A3=(0.5, 0.6, 0.7, 0.8), the risk level corresponding to the maximum risk membership of the risk factor value is high risk, and the discrete risk level value of the risk factor value is 3; When the maximum risk membership corresponding to the risk factor value belongs to the range corresponding to A4=(0.75, 0.85, 0.95, 1), the risk level corresponding to the maximum risk membership of the risk factor value is high risk, and the discrete risk level value of the risk factor value is 4.

8. An underwater robot operation risk prediction system, characterized in that: The underwater robot operation risk prediction system is used to implement the underwater robot operation risk prediction method according to claims 1 to 7, and the underwater robot operation risk prediction system includes: A data acquisition unit is used to acquire a risk value vector to be predicted of the underwater robot; the risk value vector to be predicted includes multiple risk factor values; a first risk level prediction value determination unit, configured to input the risk value vector to be predicted into a basic risk prediction model to obtain a first risk level prediction value; the basic risk prediction model is obtained by training a fully connected neural network model using a first training set, the first training set including a sample risk value vector and a first risk level sample value; a second risk level value determination unit, configured to perform expert scoring on each risk factor value in the risk value vector to be predicted using an analytic hierarchy process to obtain a weight corresponding to each risk factor value, and determine a second risk level value corresponding to the risk value vector to be predicted based on the weight corresponding to each risk factor value and a risk averaging model; a discrete risk level value determination unit, configured to calculate the risk membership corresponding to each risk factor value in the risk value vector to be predicted using a trapezoidal fuzzy membership function, and determine the risk level corresponding to the maximum risk membership of each risk factor value as the discrete risk level value of the corresponding risk factor value; a third risk level prediction value determination unit, configured to input the discrete risk level values ​​of each risk factor value into a risk level fuzzy mapping model to obtain a third risk level prediction value for each risk factor value; the risk level fuzzy mapping model is obtained by training the fully connected neural network model using a second training set, the second training set including discrete risk level values ​​of sample risk factor values ​​and corresponding third risk level sample values; a fourth risk level prediction value determination unit, configured to count the number of different risk levels in the third risk level prediction value of each risk factor value, input the number of different risk levels into a risk prediction model based on risk level quantity statistics, and output a fourth risk level prediction value; the risk prediction model based on risk level quantity statistics is obtained by training the fully connected neural network model using a third training set, the third training set including the number of different risk levels in the third risk level sample values ​​of the sample risk factor values ​​and the corresponding fourth risk level sample values; The comprehensive operation risk value determination unit is used to use the particle swarm algorithm to determine the weight of the first risk level prediction value, the weight of the second risk level value, the weight of the third risk level prediction value and the weight of the fourth risk level prediction value, thereby determining the comprehensive operation risk value of the underwater robot.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the underwater robot operation risk prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the underwater robot operation risk prediction method according to any one of claims 1 to 7 is implemented.

Images