An auxiliary method for leak monitoring in AUV cabin

Abstract

The present invention discloses an auxiliary method for monitoring leakage in an AUV cabin. During the operation of the AUV, a method based on internal sensor data fusion is used to eliminate the influence of false alarms in monitoring leakage in the cabin. The sensors carried by the AUV itself are used to monitor the operating states such as temperature and humidity, air pressure, leakage, depth, cabin deformation, motor speed, voltage and current, and the most important state variables related to leakage are extracted by correlation analysis. For two important parameters, cabin deformation and depth, the Kalman filter mathematical model is used to process the deformation data to obtain the optimal deformation value. Finally, an autoencoder method based on the fusion of multiple sensor numbers is used to assist in monitoring leakage in the underwater vehicle cabin. No additional sensors are introduced, and auxiliary monitoring of leakage is achieved by using basic cabin state monitoring data. The monitoring results are accurate and reliable, which increases the reliability of the system.

Description

Technical Field

[0001] The invention belongs to the technical field of leakage monitoring in an underwater vehicle cabin, and in particular relates to an auxiliary method for leakage monitoring in an AUV cabin. Background Art

[0002] Nowadays, autonomous underwater vehicles are increasingly recognized in the fields of national defense, marine geology, industry, etc. However, with the expansion of application areas, AUVs need to monitor and alarm whether they are leaking when performing tasks. At present, leakage detection for AUVs often adopts the form of short-circuit terminals. That is, a short-circuit detection circuit made of FPC flexible circuit board is attached to the leak-prone position. When there is water on the FPC, short-circuit detection will be triggered, and then leakage alarm information will be generated. However, AUV is an extremely complex electromechanical system, which includes a large number of electronic components and mechanical parts. Structural shedding, metal dust, shedding and damage of electronic components, etc. may cause AUV to have false leakage alarms.

[0003] Once the internal leak detection module sounds an alarm, whether it is a true leak alarm or a false alarm, the equipment will almost always be dismantled for repair. If an alarm sounds during the mission, the equipment can only surface urgently. This will significantly affect the stability and safety of the AUV, and also increase the workload.

[0004] In the prior art, there are almost no relevant solutions for false alarms. Therefore, in order to ensure that the AUV can complete the expected mission safely, efficiently and reliably, it is necessary to assist in judging the leakage alarm module of the AUV during the mission to reduce the impact of false alarms on equipment safety and mission progress. Summary of the invention

[0005] In view of the defects in the prior art, the present invention proposes an auxiliary method for leak monitoring in an AUV cabin in order to reduce the impact of false alarms on equipment safety and mission progress.

[0006] The present invention is implemented by adopting the following technical solution: an auxiliary method for leak monitoring in an AUV cabin, comprising the following steps:

[0007] Step A: Analyze the correlation between each variable and leakage based on the AUV cabin data to determine the key factors most related to cabin leakage;

[0008] The variables include air pressure, depth, main cabin temperature, main cabin humidity, motor temperature, cabin deformation, system battery voltage, system battery current, power battery voltage, and power battery current, and the key factors determined include depth, air pressure, cabin deformation, and main cabin temperature;

[0009] Step B: optimal estimation of cabin deformation based on Kalman filtering;

[0010] The cabin deformation data values ​​actually measured by multiple deformation sensors on the AUV cabin are used as observation values, and the deformation values ​​estimated by the depth value according to the numerical calculation model are used as prediction values. Based on the observation values ​​and prediction values, the optimal cabin deformation is estimated in combination with the Kalman filter equation.

[0011] Step C, establishing a leakage autoencoder through relevant data to perform leakage anomaly monitoring;

[0012] (1) The historical cabin data of the AUV is used to construct a data set, which includes six types of variables: air pressure, main cabin temperature, depth, optimal cabin deformation, initial temperature, and initial air pressure, and the training set, validation set, and test set are determined;

[0013] (2) Construct and train a leakage autoencoder, taking the six variables in the test set, namely depth, main cabin temperature, air pressure, optimal cabin deformation value, initial cabin temperature and initial air pressure, as input, and outputting the corresponding six reconstruction values ​​after being processed by the leakage autoencoder; the difference between the reconstruction value and the input is the reconstruction error, and the maximum value of the sum of the absolute values ​​of the reconstruction errors of the six variables in the test set is used as the judgment threshold;

[0014] (3) When the AUV's existing leakage sensor detects a leakage alarm, the current air pressure, main cabin temperature, depth, optimal cabin deformation, initial temperature and initial air pressure data are input into the trained leakage autoencoder to obtain the sum of the absolute values ​​of the reconstruction errors of the six variables at the current moment, which is recorded as the judgment parameter. When the judgment parameter is greater than the judgment threshold, the leakage is considered abnormal, otherwise it is a false leakage judgment.

[0015] Furthermore, when determining the key factors in step A, the key parameters related to the air pressure are determined by correlation analysis:

[0016] (1) Use the Spearman correlation coefficient to calculate the correlation between air pressure and other variables:

[0017]

[0018] Among them, Ps is the correlation calculated by the Spearman correlation coefficient, x represents the level of air pressure, y represents the level of any variable among the other nine variables, i represents the i-th variable data, and n represents the total number of variable data. represents the average of the ranks of x and y;

[0019] (2) Conduct hypothesis test on correlation, propose null hypothesis H0: Ps = 0, hypothesis H1: Ps ≠ 0; with the null hypothesis established as the background, calculate the statistic z that follows the standard normal distribution * ;

[0020]

[0021] Find the cumulative distribution function of the standard normal distribution and calculate the cumulative distribution function value z * Corresponding to a certain point m, substitute it into the following formula to get the test value p:

[0022] p=2(1-m)

[0023] If p>0.05, the null hypothesis cannot be rejected at the 95% confidence level, that is, the null hypothesis H0 is met, there is no significant difference, and there is no correlation between the data; otherwise, the null hypothesis is rejected at the 95% confidence level, that is, the hypothesis H1 is met, there is a significant difference, and there is correlation between the data.

[0024] Furthermore, step B is specifically implemented in the following manner:

[0025] (1) First, the AUV cabin deformation data is used as the observed value, and the deformation value estimated by the depth value according to the numerical calculation model y'=Lx'+B is used as the predicted value. The predicted value is corrected in real time using the Kalman filter, and the iterative calculation is as follows:

[0026]

[0027] Z k =LD k +B+V k

[0028]

[0029] Among them, S k is the observed value of the cabin deformation at time k, W k-1 With V k Represent the Gaussian noise in the observation and prediction process, respectively, and the corresponding covariances are Q and R, Z k According to the predicted value of the shape variable at time k obtained by the model, D k is the depth measurement value, L and B are numerical calculation model parameters; P k-1 yes The covariance matrix, P k yes The covariance matrix of yes The covariance matrix, K k is the Kalman gain, is the optimal deformation at the previous moment, is the optimal deformation at the current moment, is the predicted value at time k obtained based on the optimal value at the previous moment; The initial value of is set to 0. P k , P k-1The initial value of the equal covariance matrix is ​​set to the identity matrix;

[0030] (2) After the processing in step (1), the filter value is obtained, and the data is iterated in a cyclic manner in chronological order based on the current filter value to solve the optimal deformation amount of the deformation sensor at different positions at the current moment;

[0031] (3) After filtering the deformation data collected by deformation sensors at different positions, the average is taken as the final optimal deformation of the cabin to measure the overall deformation degree of the AUV cabin.

[0032] Furthermore, in step C, the leakage autoencoder includes an input layer, an encoder, a decoder and an output layer. The encoder and the decoder have symmetrical structures and each includes two hidden layers. The two hidden layers of the encoder encode the six feature data of the compressed input and extract the relevant features of the four dimensions of output depth, optimal deformation of the cabin, main cabin temperature and air pressure. At the same time, these four feature quantities are used as the input quantities of the decoder. Finally, the output layer outputs 6 output features, namely the reconstruction quantity

[0033] Compared with the prior art, the advantages and positive effects of the present invention are:

[0034] This scheme uses Kalman filtering to combine the cabin deformation data at different positions of the vehicle with the theoretical model of cabin deformation during the operation of the AUV to obtain the optimal value of cabin deformation, which is used to eliminate the influence of position interference on cabin deformation; the leakage autoencoder is used to reconstruct variables such as temperature, air pressure, cabin deformation, depth, initial temperature and initial air pressure, and the reconstructed value is subtracted from the measured value to generate a residual as a judgment standard for abnormal monitoring;

[0035] No additional sensors were introduced, and auxiliary monitoring of leakage was achieved using basic cabin status monitoring data. The monitoring results were accurate and reliable, which increased the reliability of the system. The entire method was based on the causal relationship of leakage, and variables that were strongly correlated with leakage were selected, using unsupervised methods to monitor abnormal conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a schematic diagram of the auxiliary method flow in an embodiment of the present invention;

[0037] Figure 2 It is a schematic diagram of a block diagram of an autoencoder trainer according to an embodiment of the present invention;

[0038] Figure 3 Schematic diagram of the structure of the leakage autoencoder according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the specific embodiments disclosed below.

[0040] Aiming at the false alarm phenomenon of cabin leakage monitoring during the operation of AUV, the present invention proposes an auxiliary method for AUV cabin leakage monitoring, the basic principle of which is as follows: firstly, the AUV cabin data is used to extract the key factors related to the leakage; then, the Kalman filter is used to obtain the optimal cabin deformation to eliminate the influence of position interference on the cabin deformation; when the original leakage sensor alarms, the leakage autoencoder is used to reconstruct and detect the determined key factors and parameters such as the optimal cabin deformation, and the reconstructed value is subtracted from the input value, and the generated residual is used as the judgment standard for abnormal monitoring to further determine whether the cabin is leaking.

[0041] Specifically, Figure 1 As shown, the auxiliary method for AUV cabin leakage monitoring includes the following steps:

[0042] Step A: Analyze the correlation between various variables and leakage based on the AUV cabin data to determine the key factors most related to cabin leakage;

[0043] First, a large amount of AUV cabin data is used to analyze the correlation between various variables and leakage. Normally, cabin leakage is directly related to changes in cabin air pressure. If abnormal cabin air pressure changes can be detected, then it is possible to indirectly determine whether the cabin is leaking. In other words, the key factors related to leakage can be determined by determining key parameters related to air pressure. In general, the monitoring parameters of the AUV cabin status system include ten variables: air pressure, depth, main cabin temperature / humidity, motor temperature, cabin deformation, system battery voltage, current, and power battery voltage and current. However, some of these variables are not related to air pressure. Therefore, in order to determine the variables most related to air pressure, this embodiment uses correlation analysis to determine the sensor variables related to air pressure. The specific analysis is as follows:

[0044] For the monitoring parameters of the AUV cabin status system (ten variables including air pressure, depth, main cabin temperature and humidity, motor temperature, cabin deformation, system battery and power battery voltage and current), deformation is monitored by fixed-point sampling. From experience, it can be known that the pressure on the cabin body varies at different locations. The closer to the support structure, the greater the compressive strength. Therefore, there are differences in the data recorded by the strain gauge. The deformation data used below must reflect the overall deformation degree of the AUV. Therefore, the deformation data collected by deformation sensors at different positions are averaged, and the average value of multiple deformation data is used as a standard to measure the overall deformation degree of the AUV. This embodiment uses the Spearman correlation coefficient, which does not have strict requirements on data conditions, to calculate the correlation between air pressure and other variables, as follows:

[0045] (1) The Spearman correlation coefficient is used to calculate the correlation between air pressure and the other variables mentioned above. The calculation formula is shown in Formula 1:

[0046]

[0047] Among them, Os is the correlation calculated by the Spearman correlation coefficient, x represents the level of air pressure, and y represents the level of any variable among the other nine variables. The variable level is to arrange the variable values ​​from low to high over a period of time. The lower the value, the smaller the level, and the larger the value, the larger the level. i represents the i-th variable data, and n represents the total number of data. Represents the rank average of x and y.

[0048] (2) After calculating the correlation between air pressure and other variables, it is necessary to conduct hypothesis testing on the correlation. The null hypothesis H0 is Ps = 0, and the hypothesis H1 is Ps ≠ 0. With the null hypothesis established as the background, calculate the statistic z that follows the standard normal distribution * , as shown in formula 1-2;

[0049]

[0050] Find the cumulative distribution function of the standard normal distribution (the cumulative distribution function is the integral of the probability density function, which is a function that describes the probability of the output value of this random variable near a certain value point), and calculate the cumulative distribution function value z * Substitute the corresponding certain point m into the test value p, as shown in formula 1-3.

[0051] p=2(1-m) (1-3)

[0052] If p>0.05, the null hypothesis cannot be rejected at a 95% confidence level, that is, the null hypothesis H0(ρ S=0), there is no significant difference, and there is no correlation between the data; if p < 0.05, the null hypothesis is rejected at a 95% confidence level, that is, the hypothesis H1 (ρ S ≠0), there is a significant difference, and there is a certain correlation between the data; the specific correlation analysis results are shown in Table 1:

[0053] Table 1 Correlation coefficients between air pressure and other variables

[0054]

[0055]

[0056] According to Table 1, the p-value of the Spearman correlation coefficient of the motor speed is 0.7175, which is greater than 0.05, so the null hypothesis cannot be rejected, indicating that there is no correlation. The p-values ​​of the remaining parameters are all less than 0.05, so the null hypothesis can be rejected, indicating that there is a certain correlation. After determining whether there is a correlation, the variables are classified according to the degree of correlation, and it can be obtained that the following have a strong correlation with the air pressure: depth, deformation, main cabin temperature and motor temperature. Since the AUV motor and cabin are designed as independent parts, they do not directly affect the air pressure changes in the AUV cabin, and the main cabin temperature and motor temperature are variables of the same dimension with similar change trends, so the main cabin temperature of the two is taken as the fusion variable in the leakage monitoring method, and then the key factors related to the leakage are finally determined to include depth, air pressure, cabin deformation and main cabin temperature.

[0057] Step B: Optimal estimation of cabin deformation based on Kalman filtering:

[0058] Since the cabin deformation is obtained by multiple deformation sensors attached to the AUV cabin, the deformation sensors will amplify the measurement noise when amplifying the weak cabin deformation. In addition, considering the rebound phenomenon of the cabin itself, it may cause a large drift value in the measurement value, so it is necessary to process the deformation data.

[0059] This embodiment establishes a Kalman filter model, using the deformation data values ​​of the cabin actually measured by multiple deformation sensors as the observed value, and using the deformation value estimated by the depth value according to the numerical calculation model (the form of the model is y'=Lx'+B) as the predicted value. Using the observed value and the predicted value, combined with the Kalman filter equation, the optimal estimated value of the cabin deformation can be obtained, and the specific analysis is as follows:

[0060] First, the AUV cabin deformation data is used as the observed value, and the deformation value estimated by the numerical calculation model using the depth value is used as the predicted value. The Kalman filter is used to perform real-time correction on the predicted value. The iterative calculation is shown in Formula (2-1) to Formula (2-6).

[0061]

[0062] Z k =LD k +B+V k (2-2)

[0063] P k - =P k-1 +Q (2-3)

[0064]

[0065] Among them, S k is the observed value of the cabin deformation at time k, W k-1 With V k Represent the Gaussian noise in the observation and prediction process, respectively, and their covariances are Q and R, Z k According to the predicted value of the shape variable at time k obtained by the model, D k is the depth measurement value, L and B are numerical calculation model parameters; P k-1 yes The covariance matrix, P k yes The covariance matrix of . yes The covariance matrix, K k is the Kalman gain, is the optimal deformation at the previous moment, is the optimal deformation at the current moment, is the predicted value at time k obtained based on the optimal value at the previous moment. The initial value of is set to 0. P k , P k-1 The initial value of the equal covariance matrix is ​​set to the identity matrix, so that after a period of iteration, the result of the filter output is independent of the initial value.

[0066] After the above processing, the filter value is obtained. Based on the current filter value, the data is iterated in chronological order to improve the reliability of the deformation data and reduce the false alarm rate. The deformation data in the AUV cabin needs to use the depth value to calculate the predicted value of the deformation variable according to the theoretical model reference, and the strain collection value is used as the observation value. The respective Kalman models are established, and the optimal deformation variables at different positions at the current moment are solved using Formula 2. The deformation data collected by the deformation sensors at different positions are filtered and averaged as the final optimal deformation of the cabin to measure the overall deformation degree of the AUV cabin.

[0067] Step C: Finally, an autoencoder is established through relevant data to perform leakage anomaly detection;

[0068] When the leakage sensor alarms, the air pressure, main cabin temperature, depth, optimal cabin deformation, initial temperature and initial air pressure data collected by the cabin sensor are input into the leakage autoencoder for leakage anomaly detection. Since the two variables of temperature and air pressure are greatly affected by the environment, human factors (such as vacuuming) will also have an intuitive change on the air pressure, and the change trend of the air pressure will also change with the change of the initial quantity. The experimental data is the performance data change, and the interception depth changes more, that is, the task execution stage with rich variable changes. Therefore, the initial cabin temperature and initial air pressure are introduced as feature data, among which the cabin air pressure and temperature when the AUV just enters the water are used as the initial cabin temperature and initial air pressure.

[0069] like Figure 2 As shown in the figure, with the initial temperature and air pressure as the benchmark, a total of six variables are used to train the leakage autoencoder, and the autoencoder network structure, the learning rate of the input data and the number of iterations are adjusted to obtain the autoencoder structure with a smaller loss function. The AUV depth, the optimal deformation of the cabin, the temperature and air pressure of the main cabin are taken as the characteristic data of leakage anomaly. The historical cabin data of the AUV is used to construct a data set, with 70% of the data set as the training set, 15% of the data set as the validation set, and 15% of the data set as the test set.

[0070] The constructed leaky autoencoder encodes the output in four dimensions with two hidden layer MLP structures, compresses the six input feature data, extracts relevant features, and reconstructs the input data with a decoder with a symmetrical structure. Figure 3 , the encoder input layer contains 6 input units, hidden layer 1 contains 64 neurons and hidden layer 2 contains 16 neurons. Hidden layer 2 outputs 4 features. As the feature obtained by compressing the 6 input quantities. At the same time, these 4 features are used as the input quantity in hidden layer 1 of the decoder. At the same time, hidden layer 1 in the decoder contains 16 neurons and hidden layer 2 contains 64 neurons. Finally, the output layer contains 6 output features, i.e., the reconstruction quantity. The learning rate of the entire autoencoder is 0.0001, the number of iterations is 1000, and the optimization function uses Adam. The activation function uses RELU, which largely solves the problems of gradient explosion and gradient disappearance.

[0071] The input of the leaky autoencoder is six variables, which learns the features between data. The abnormal changes of each data will cause reconstruction errors of each data. Therefore, the maximum value of the sum of the absolute values ​​of the reconstruction errors of the six variables in the test set is used as the judgment threshold, as shown in Formula 3, where e' is the reconstruction error of each variable under leakage data, e represents the reconstruction error of different variables under normal conditions, and the subscript i refers to the six input data in turn.

[0072]

[0073] Therefore, the autoencoder uses the above six variables (air pressure, main cabin temperature, depth, optimal deformation value of the cabin, initial temperature and initial air pressure) as the input layer. After the data features are extracted through the hidden layers of the middle two layers, six reconstruction values ​​are output in the output layer, namely: reconstructed AUV depth, reconstructed optimal deformation of the cabin, reconstructed main cabin temperature, reconstructed air pressure, reconstructed initial cabin temperature, and reconstructed initial air pressure. The reconstruction values ​​are subtracted from the input values ​​to obtain the corresponding residuals, and the absolute values ​​of the residuals are summed to obtain the judgment parameters. When the judgment parameter is greater than the set judgment threshold, it is considered that a leakage anomaly has occurred, otherwise it is a leakage misjudgment.

[0074] In addition, it should be noted that in this embodiment, in step B, when estimating the cabin shape variable, it is not limited to the Kalman filter described in the embodiment, and other filtering algorithms such as extended Kalman filter, unscented Kalman filter, particle filter and other optimal estimation methods are also feasible; and step C can also use other types of autoencoders, such as convolutional autoencoders, recurrent neural network autoencoders, sparse autoencoders, stacked autoencoders, etc., which will not be elaborated in detail here.

[0075] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An auxiliary method for leak monitoring in an AUV cabin, characterized in that: The following steps are involved: Step A: Analyze the correlation between each variable and leakage based on the AUV cabin data to determine the key factors most related to cabin leakage; The variables include air pressure, depth, main cabin temperature, main cabin humidity, motor temperature, cabin deformation, system battery voltage, system battery current, power battery voltage, and power battery current, and the key factors determined include depth, air pressure, cabin deformation, and main cabin temperature; Step B, optimal estimation of cabin deformation based on Kalman filtering; using the cabin deformation data values ​​actually measured by multiple deformation sensors on the AUV cabin as observation values, using the deformation values ​​estimated by the numerical calculation model using the depth value as the prediction value, and estimating the optimal cabin deformation based on the observation value and the prediction value in combination with the Kalman filter equation; specifically: (1) First, the AUV cabin deformation data is used as the observed value, and the deformation value estimated by the depth value according to the numerical calculation model y'=Lx'+B is used as the predicted value. The predicted value is corrected in real time using the Kalman filter, and the iterative calculation is as follows: From k =LD k +B+V k Among them, S k is the observed value of the cabin deformation at time k, W k-1 With V k Represent the Gaussian noise in the observation and prediction process, respectively, and the corresponding covariances are Q and R, Z k According to the predicted value of the shape variable at time k obtained by the model, D k is the depth measurement value, L and B are numerical calculation model parameters; P k-1 yes The covariance matrix, P k yes The covariance matrix of yes The covariance matrix, K k is the Kalman gain, is the optimal deformation at the previous moment, is the optimal deformation at the current moment, is the predicted value at time k obtained based on the optimal value at the previous moment; The initial value of is set to 0. P k , P k-1 The initial value of the covariance matrix is ​​set to the identity matrix; (2) After the processing in step (1), the filter value is obtained, and the data is iterated in a cyclic manner in chronological order based on the current filter value to solve the optimal deformation amount of the deformation sensor at different positions at the current moment; (3) After filtering the deformation data collected by deformation sensors at different positions, the average is taken as the final optimal deformation of the cabin to measure the overall deformation degree of the AUV cabin; Step C, establishing a leakage autoencoder through relevant data to perform leakage anomaly monitoring; (1) The historical cabin data of the AUV is used to construct a data set, which includes six types of variables: air pressure, main cabin temperature, depth, optimal cabin deformation, initial temperature, and initial air pressure, and the training set, validation set, and test set are determined; (2) Construct and train a leakage autoencoder, taking the six variables in the test set, namely depth, main cabin temperature, air pressure, optimal cabin deformation value, initial cabin temperature and initial air pressure, as input, and outputting the corresponding six reconstruction values ​​after being processed by the leakage autoencoder; the difference between the reconstruction value and the input is the reconstruction error, and the maximum value of the sum of the absolute values ​​of the reconstruction errors of the six variables in the test set is used as the judgment threshold; (3) When the AUV's existing leakage sensor detects a leakage alarm, the current air pressure, main cabin temperature, depth, optimal cabin deformation, initial temperature and initial air pressure data are input into the trained leakage autoencoder to obtain the sum of the absolute values ​​of the reconstruction errors of the six variables at the current moment, which is recorded as the judgment parameter. When the judgment parameter is greater than the judgment threshold, the leakage is considered abnormal, otherwise it is a false leakage judgment.

2. The auxiliary method for AUV cabin leakage monitoring according to claim 1, characterized in that: When determining the key factors in step A, the key parameters related to air pressure are determined by correlation analysis: (1) Use the Spearman correlation coefficient to calculate the correlation between air pressure and other variables: Among them, Ps is the correlation calculated by the Spearman correlation coefficient, x represents the level of air pressure, y represents the level of any variable among the other nine variables, i represents the i-th variable data, and n represents the total number of variable data. represents the average of the ranks of x and y; (2) Conduct hypothesis test on correlation, propose null hypothesis H0: Ps = 0, hypothesis H1: Ps ≠ 0; with the null hypothesis established as the background, calculate the statistic z that follows the standard normal distribution * ; Find the cumulative distribution function of the standard normal distribution and calculate the cumulative distribution function value z * Corresponding to a certain point m, substitute it into the following formula to get the test value p: p=2(1-m) If p>0.05, the null hypothesis cannot be rejected at the 95% confidence level, that is, the null hypothesis H0 is met, there is no significant difference, and there is no correlation between the data; otherwise, the null hypothesis is rejected at the 95% confidence level, that is, the hypothesis H1 is met, there is a significant difference, and there is correlation between the data.

3. The auxiliary method for AUV cabin leakage monitoring according to claim 1, characterized in that: In step C, the leakage autoencoder includes an input layer, an encoder, a decoder and an output layer. The encoder and the decoder are symmetrical in structure and each includes two hidden layers. The two hidden layer MLP structures of the encoder encode the six feature data of the compressed input and extract the relevant features of the four dimensions of output depth, optimal deformation of the cabin, main cabin temperature and air pressure. At the same time, these four feature quantities are used as the input quantities of the decoder. Finally, the output layer outputs 6 output features, namely the reconstruction quantities.

Images