A ship integrated navigation fault detection method based on cascade neural network

Abstract

The present invention discloses a ship integrated navigation fault detection method based on cascade neural network, comprising: S1: establishing the integrated navigation system equation of surface ship SINS / GNSS and SINS / DVL / COMPASS; S2: selecting the variable of the integrated navigation state estimation error; S3: selecting the variable of the integrated navigation measurement error; S4: obtaining the covariance matrix P generated in the filtering process of the integrated navigation system k ; S5: Get the residual χ 2 Failure information statistics q of the test method k ; S6: Obtaining fault information of the integrated navigation system. Aiming at the problems that the fault detection algorithm based on the residual observer has the problem that the gradual fault is not easy to be identified and the fault threshold setting depends on prior knowledge, the present invention adopts a cascade neural network to achieve greater sensitivity to the gradual fault, improves the intelligence of the fault detection of the integrated navigation system, and has the robustness of the fault detection.

Description

Technical Field

[0001] The invention relates to the technical field of ship integrated navigation fault detection, and in particular to a ship integrated navigation fault detection method based on cascade neural network. Background Art

[0002] The normal operation of the integrated navigation system depends on the normal operation of each subsystem. The detection and separation of navigation sensor faults helps to improve the robustness of the integrated navigation system. In practical applications, due to the influence of external factors or the reasons of the components themselves, the output of the navigation sensor is inevitably faulty. Under the condition that the noise and measurement noise of the integrated navigation system are bounded and unknown, the fault detection algorithm based on the residual observer has the defects of relying on prior knowledge, difficult to set the threshold, and insensitivity to gradual faults. Summary of the invention

[0003] The present invention provides a ship integrated navigation fault detection method based on cascade neural network to overcome the above technical problems.

[0004] In order to achieve the above object, the technical solution of the present invention is:

[0005] A ship integrated navigation fault detection method based on cascade neural network, comprising:

[0006] S1: Establish the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships;

[0007] S2: Select the variable of the integrated navigation state estimation error:

[0008] S3: Select the variable of integrated navigation measurement error:

[0009] S4: According to the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships, obtain the discretized combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships to obtain the covariance matrix P generated in the filtering process of the combined navigation system k ;

[0010] S5: According to the variable of the integrated navigation state estimation error and the variable of the integrated navigation measurement error, the residual of the surface ship SINS / GNSS and SINS / DVL / COMPASS integrated navigation system at time k is obtained to obtain the residual χ 2 Failure information statistics q of the test method k ;

[0011] S6: Based on the covariance matrix P generated during the filtering process of the integrated navigation system k and the residual χ2 Failure information statistics q of the test method k ,The fault information of the combined navigation systems of SINS / GNSS and SINS / DVL / COMPASS of surface ships is obtained through cascade neural networks.

[0012] Furthermore, in S1, the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships are established as follows:

[0013]

[0014] In the formula: x k is the estimated error of the integrated navigation state at time k; k is the combined navigation measurement error at time k; f(.) is the nonlinear state transfer equation; H k is the measurement equation at time k; D k is the coefficient matrix of the system measurement noise at time k; w k is the system process noise at time k; v k is the system measurement noise at time k; k represents the time when the system is working.

[0015] Furthermore, in S2, the variable of the integrated navigation state estimation error is selected as follows:

[0016]

[0017] Where: is the pitch angle error; is the roll angle error; is the heading angle error; δv E is the eastward velocity error; δv N is the north velocity error; δv U is the celestial velocity error; δL is the latitude error; δλ is the longitude error; δh is the height error; ε E is the gyro zero bias in the east direction; ε N is the gyro bias in the north direction; U The gyro zero bias for the celestial direction; is the accelerometer zero bias in the east direction; is the accelerometer bias in the north direction; is the accelerometer zero bias in the celestial direction; ε and They are the gyroscope bias and accelerometer bias respectively.

[0018] Furthermore, in S3, the variables of the integrated navigation measurement error are selected as follows:

[0019]

[0020] Where: is the measurement error of GNSS; is the measurement error of DVL and COMPASS; is the position error of the SINS system at time k; is the position error of the GNSS system at time k; is the velocity error of the SINS system at time k; is the velocity error of the GNSS system at time k; is the velocity error of the DVL system at time k; is the heading angle error of the SINS system at time k; is the heading angle error of the COMPASS system at time k.

[0021] Furthermore, in S4, the combined navigation system equations of the discretized surface ship SINS / GNSS and SINS / DVL / COMPASS are obtained as follows:

[0022]

[0023] In the formula, is the one-step prediction of the state error; P k|k-1 is the state error covariance matrix for one-step prediction; Φ k is the state transfer matrix of the discretization of f(.); B k is the coefficient matrix of the discretized system process noise; Q k is the system noise covariance matrix at time k; K k is the filter gain at time k; I is the unit matrix; where, E(.) is the expectation equation; is the one-step prediction of the state error; P k|k-1 is the state error covariance matrix for one-step prediction; R k is the measurement noise covariance matrix; P k is the covariance matrix generated during the filtering process of the integrated navigation system.

[0024] Further, in S5, the residual χ is obtained. 2 The fault information statistics method of the inspection method is as follows:

[0025] S51: Get the residuals of the surface ship SINS / GNSS and SINS / DVL / COMPASS integrated navigation systems at time k:

[0026]

[0027] Where: γ k represents the residual error of the integrated navigation system at time k; One-step prediction for state error;

[0028] S52: Get the theoretical value of the innovation variance:

[0029]

[0030] Where: S k is the theoretical value of the innovation variance; P k|k-1 is the state error covariance matrix for one-step prediction; R k is the measurement noise covariance matrix;

[0031] S53: Get the residual χ 2 The fault information statistics of the test method is

[0032]

[0033] Where: q k χ represents the residual 2 Failure information statistics of the test method.

[0034] Furthermore, the cascade neural network in S6 is composed of two parts: a time domain feature fault detection network and a STFT frequency domain fault detection network.

[0035] The time domain feature fault detection network can simultaneously extract the residual χ 2 Failure information statistics q of the test method k The statistical characteristics of the integrated navigation system and the covariance matrix P generated during the filtering process k The slow features are input into the BP neural network;

[0036] The frequency domain fault detection network of the STFT can extract the residual χ 2 Failure information statistics q of the test method k The frequency domain fault features are input into the CNN neural network;

[0037] Finally, the outputs of the time domain characteristic fault detection network and the frequency domain fault detection network of STFT are used through the decision network to obtain the fault information of the combined navigation systems of SINS / GNSS and SINS / DVL / COMPASS of surface ships.

[0038] Beneficial effects: The ship combined navigation fault detection method based on cascade neural network of the present invention aims at the problems that gradual faults are difficult to be identified and the fault threshold setting depends on prior knowledge in the fault detection algorithm based on residual observer. The time-frequency fault characteristics of the fault information statistics of the covariance matrix and the residual matrix generated in the filtering process of the combined navigation system are extracted, and the cascade neural network is used to achieve greater sensitivity to gradual faults, thereby improving the intelligence of the combined navigation system fault detection and having the robustness of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 is a flow chart of the fault detection method of the present invention;

[0041] Figure 2 Schematic diagram of STFT feature extraction and fault detection in an embodiment of the present invention;

[0042] Figure 3 is a schematic diagram of a cascade neural network structure in an embodiment of the present invention;

[0043] Figure 4a Schematic diagram of SINS / GNSS time domain characteristic fault detection results in an embodiment of the present invention;

[0044] Figure 4b Schematic diagram of SINS / GNSS frequency domain feature extraction results in an embodiment of the present invention;

[0045] Figure 4c Schematic diagram of SINS / GNSS frequency domain characteristic fault detection network results in an embodiment of the present invention

[0046] Figure 4d Schematic diagram of decision-level fusion fault detection network results in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] This embodiment provides a ship integrated navigation fault detection method based on cascade neural network, such as Figure 1 As shown, the following steps are included:

[0049] Specifically, the present embodiment mainly solves the intelligent fault detection of a ship strapdown inertial navigation system / global navigation satellite system (SINS / GNSS) and a ship strapdown inertial navigation system / Doppler log / gyrocompass (SINS / DVL / COMPASS) combined navigation system.

[0050] S1: Establish the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships:

[0051] Among them, SINS / GNSS refers to the ship strapdown inertial navigation system / global navigation satellite system; SINS / DVL / COMPASS refers to the ship strapdown inertial navigation system / Doppler log / gyrocompass combined navigation system;

[0052]

[0053] In the formula: x k is the estimated error of the integrated navigation state at time k; k is the combined navigation measurement error at time k; f(.) is the nonlinear state transfer equation; H k is the measurement equation at time k; D k is the coefficient matrix of the system measurement noise at time k; w k is the system process noise at time k; v k is the system measurement noise at time k; where w k and v k All satisfy the bounded l2[0,N] norm; k represents the time when the system works;

[0054] S2: Select the variable of the integrated navigation state estimation error:

[0055]

[0056] Where: is the pitch angle error; is the roll angle error; is the heading angle error; δv E is the eastward velocity error; δv N is the north velocity error; δv U is the celestial velocity error; δL is the latitude error; δλ is the longitude error; δh is the height error; ε E is the gyro zero bias in the east direction; ε N is the gyro bias in the north direction; U The gyro zero bias for the celestial direction; is the accelerometer zero bias in the east direction; is the accelerometer bias in the north direction; is the accelerometer zero bias in the celestial direction; ε and are the gyroscope bias and accelerometer bias, respectively; the subscripts E, N, and U correspond to the directions east, north, and sky, respectively;

[0057] S3: Select the variable of the integrated navigation measurement error: For the SINS / GNSS system, the integrated navigation measurement error at time k is the measurement error of GNSS For the SINS / DVL / COMPASS system, the combined navigation measurement error at time k is the measurement error of DVL and COMPASS

[0058]

[0059] Where: is the measurement error of GNSS; is the measurement error of DVL and COMPASS; is the position error of the SINS system at time k; is the position error of the GNSS system at time k; is the velocity error of the SINS system at time k; is the velocity error of the GNSS system at time k; is the velocity error of the DVL system at time k; is the heading angle error of the SINS system at time k; is the heading angle error of the COMPASS system at time k;

[0060] S4: According to the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships, the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships are obtained, i.e., the prediction equation group, so as to obtain the covariance matrix generated in the filtering process of the combined navigation system;

[0061] Specifically, in this embodiment, the Kalman filter is used to minimize the combined navigation state estimation error x through the prediction-update step. k And the covariance matrix P generated in the filtering process of the integrated navigation system k After discretization, the discretized surface ship SINS / GNSS and SINS / DVL / COMPASS combined navigation system equations are as follows:

[0062]

[0063] In the formula, is the one-step prediction of the state error; P k|k-1 is the state error covariance matrix for one-step prediction; Φ k is the state transfer matrix of the discretization of f(.); Bk is the coefficient matrix of the discretized system process noise; Q k is the system noise covariance matrix at time k; K k is the filter gain at time k; I is the unit matrix; where, E(.) is the expectation equation; is the one-step prediction of the state error at time k; P k|k-1 is the state error covariance matrix of the k-th moment one-step prediction; R k is the measurement noise covariance matrix;

[0064] S5: According to the variable of the integrated navigation state estimation error and the variable of the integrated navigation measurement error, the residual of the surface ship SINS / GNSS and SINS / DVL / COMPASS integrated navigation system at time k is obtained to obtain the residual χ 2 Failure information statistics q of the test method k ;

[0065] For surface ship SINS / GNSS and SINS / DVL / COMPASS integrated navigation systems, the residual χ 2 The inspection method is used to detect the ship integrated navigation system. The measurement residual information of the system at time k is as follows:

[0066] In S5, the residual χ is obtained. 2 The fault information statistics method of the inspection method is as follows:

[0067] S51: Get the residuals of the surface ship SINS / GNSS and SINS / DVL / COMPASS integrated navigation systems at time k:

[0068]

[0069] Where: γ k represents the residual error of the integrated navigation system at time k; One-step prediction of state error;

[0070] S52: Get the theoretical value of the innovation variance:

[0071]

[0072] Where: S k is the theoretical value of the innovation variance; P k|k-1 is the state error covariance matrix for one-step prediction; R k is the measurement noise covariance matrix;

[0073] S53: Get the residual χ 2 The fault information statistics of the test method is

[0074]

[0075] Where: q k χ represents the residual 2 Failure information statistics of the inspection method;

[0076] S6: Based on the covariance matrix P generated during the filtering process of the integrated navigation system k and the residual χ 2 Failure information statistics q of the test method k ,The fault information of the combined navigation systems of SINS / GNSS and SINS / DVL / COMPASS of surface ships is obtained through cascade neural networks.

[0077] Specifically, the conventional method is based on the residual χ 2 The fault information statistics of the test method are used to establish the fault detection rules: according to the statistical characteristics of the residual, the threshold value T is set D , establish the following fault detection rules:

[0078]

[0079] Where: T D is the fault detection threshold;

[0080] The above residual-based χ 2 The detection algorithm is not sensitive to gradual faults. If the gradual fault is small at the beginning, The fault output will be tracked, so the residual is always small. In addition, the fault detection threshold value T D The setting depends on human experience and only takes q k It is difficult to judge whether there is a fault based on the amplitude of . Therefore, a cascade neural network is used to optimize it.

[0081] Extract the covariance P k The residual χ 2 Failure information statistics q of the test method k Based on the fault characteristics in the paper, a residual observer fault detection method based on cascade neural network is designed.

[0082] In this embodiment, the covariance matrix P generated during the filtering process of the integrated navigation system is used k and the residual χ 2 Failure information statistics q of the test method k As input, it is input into the cascade neural network. Preferably, the cascade neural network in this embodiment is composed of two parts: a time domain feature fault detection network and a STFT frequency domain fault detection network, that is, a short-time Fourier transform (STFT) frequency domain fault detection network.

[0083] The time domain feature fault detection network can simultaneously extract the residual χ 2 Failure information statistics q of the test method k The statistical characteristics of the integrated navigation system and the covariance matrix P generated during the filtering process k Then the statistical features and the slow features (i.e., time-domain fault features) are input into the BP neural network;

[0084] The frequency domain fault detection network of the STFT can extract the residual χ 2 Failure information statistics q of the test method k The frequency domain fault features are input into the CNN neural network;

[0085] Finally, the output of the time domain feature fault detection network and the frequency domain fault detection network of STFT are output through the decision network to output the fault information of the combined navigation system of surface ships SINS / GNSS and SINS / DVL / COMPASS. In order to realize the real-time and accurate identification of faults. The decision network is a universal feature in the cascade neural network, which has the function of obtaining the output of the cascade neural network according to the output of the input time domain feature fault detection network and the output of the frequency domain fault detection network of STFT. It does not belong to the technical point of the present invention, so it is not described in detail.

[0086] Specifically, the cascade neural network of this embodiment detects the P in the integrated navigation system by fusing the time domain features with the time domain feature fault detection network. k and q k For the one-dimensional signal q k Extraction, its morphological characteristics include the amplitude of the signal, statistical characteristics include the skewness and kurtosis of the signal; and time domain characteristics include the rate of change, mean, extreme value and covariance, etc. Considering the covariance P k The dimension is high, so a slow feature detection method is introduced to detect P k Perform dimensionality reduction feature extraction to obtain slow features. Input these features into the BP neural network; at the same time, use wavelet transform to transform q k After extracting the envelope of the signal change trend, the STFT frequency domain fault detection network is used to obtain q k The frequency domain fault features in the multi-dimensional frequency domain features are substituted into the CNN for training as follows: Figure 2 Finally, the training results of CNN and BP network are input into the decision network to realize the online fault detection network as shown in Figure 3 shown.

[0087] Specifically, in this embodiment, the extracted P k and q kThe time domain fault features are input into the BP neural network. The BP neural network contains three hidden layers, which are composed of 20, 30 and 10 neurons respectively. The activation functions are sigmod, sigomd and Tanh respectively. The output result is whether the time domain feature fault detection network has a fault. Considering q k After frequency domain feature extraction, the image is a two-dimensional image. In order to better extract features, it is input into the convolutional neural network CNN to extract features from the frequency domain data. The network contains 8 7*7 convolutional layers, and after using the Relu function to increase the nonlinear segmentation ability of the network, the result is input into 4 3*3 pooling layers, and then the Relu function is used again to enhance the nonlinear segmentation ability. After a 2*2 maximum pooling layer for feature extraction, the network is subjected to packet loss processing, and after being output to the full convolution layer, feature classification is performed to output the results. Finally, the training results of CNN and BP neural networks are input into the decision network to realize the online fault detection network, such as Figure 3 shown.

[0088] The combined navigation residual q of multiple simulations under normal and fault conditions is k and covariance P k The input network was pre-trained 10,000 times in the time domain feature fault detection network and the frequency domain fault detection network, and the decision network was trained 22,000 times. According to the loss value and accuracy of the cross-experiment, the batch size was determined to be 50, the learning rate was set to 0.0001, and the training target (minimum mean square error, MMSE) was set to 10 -4 , the learning rate is set to 0.001, and the training target is set to MMSE of 10 -6 To avoid overfitting, the packet loss rate is set to 0.5 for the frequency domain fault detection sub-network. For the decision-level fusion network, the training rate and MMSE settings are the same as those of the time domain feature fault detection network.

[0089] The q generated in the actual surface ship integrated navigation process k and P k When input into the trained network, fault detection in the integrated navigation subsystem can be completed.

[0090] A specific embodiment is as follows:

[0091] For the SINS / GNSS integrated navigation system, a real ship test is carried out according to the method of this embodiment, and the integrated navigation subsystem is tested for faults. Figure 4a It can be seen that for the SINS / GNSS integrated navigation system, for the time domain feature fault detection network, most of the SINS gradual faults between 100 and 150 seconds are detected; however, some false alarms occur after 250 seconds. Figure 4bAs shown, it can be Figure 4c It can be seen that the frequency domain fault detection has a large number of false alarms between 150 and 200 seconds. For the fusion results of the decision network, it not only achieves a lower false alarm rate than the frequency domain fault detection network and the frequency domain fault detection network, but also successfully detects most faults, including SINS soft faults, such as Figure 4d As shown. It can be shown that compared with the existing methods, this patent solves the problem that the fault detection algorithm based on the residual observer has the problem that the gradual fault is not easy to identify and the fault threshold setting depends on prior knowledge. The covariance matrix P generated in the filtering process of the integrated navigation system k and the residual χ 2 Failure information statistics q of the test method k The time-frequency fault features are extracted and optimized by neural network. On-line intelligent fault detection is realized with less real ship data and it is more sensitive to gradual faults.

[0092] Aiming at the problem that faults existing in a ship integrated navigation system may pollute the output of a federated filter and affect the navigation performance, the present invention constructs a ship integrated navigation fault detection method based on a cascade neural network, and aims at the defects that a residual observer is insensitive to gradual faults and relies on prior knowledge, and designs a fault detection algorithm that integrates a multi-time domain feature fusion fault detection network and a short-time Fourier transform (STFT) frequency domain fault detection network.

[0093] Aiming at the defects of residual observer such as insensitivity to gradual faults and dependence on prior knowledge, a fault detection method integrating time domain feature fault detection network and STFT frequency domain fault detection network (short-time Fourier transform frequency domain fault detection network) is designed. Aiming at the problems that gradual faults are not easy to be identified and the fault threshold setting depends on prior knowledge in the fault detection algorithm based on residual observer, the time-frequency fault features of covariance matrix and residual matrix are extracted. After getting rid of the traditional judgment threshold T D In addition to setting defects that rely on human experience, the detection of gradual faults is also more stable. Online intelligent fault detection is achieved, and it is more sensitive to gradual faults.

[0094] Beneficial effects:

[0095] The present invention extracts the time domain features and frequency domain features of the covariance signal generated in the residual and integrated navigation system filtering process from the output of the traditional residual observer-based fault detection algorithm. By inputting the extracted features into the designed cascade network, the online intelligent detection of the ship integrated navigation subsystem is realized, which is more sensitive to gradual faults while not relying on artificial experience to set the detection threshold parameters. At the same time, it gets rid of the constraint that the traditional neural network fault detection relies on a large number of measured data samples for training, and can complete the fault detection of the subsystem with less data.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship integrated navigation fault detection method based on cascade neural network, characterized in that: include: S1: Establish the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships; S2: Select the variable of the integrated navigation state estimation error: S3: Select the variable of integrated navigation measurement error: S4: According to the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships, obtain the discretized combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships to obtain the covariance matrix P generated in the filtering process of the combined navigation system k ; S5: According to the variable of the integrated navigation state estimation error and the variable of the integrated navigation measurement error, the residual of the surface ship SINS / GNSS and SINS / DVL / COMPASS integrated navigation system at time k is obtained to obtain the residual χ 2 Fault information statistics q of the test method k ; S6: Based on the covariance matrix P generated during the filtering process of the integrated navigation system k and the residual χ 2 Fault information statistics q of the test method k , the fault information of the combined navigation systems of SINS / GNSS and SINS / DVL / COMPASS of surface ships is obtained through cascade neural networks; The cascade neural network in S6 is composed of two parts: a time domain feature fault detection network and a STFT frequency domain fault detection network. The time domain feature fault detection network can simultaneously extract the residual χ 2 Fault information statistics q of the test method k The statistical characteristics of the integrated navigation system and the covariance matrix P generated during the filtering process k The slow features are input into the BP neural network; The frequency domain fault detection network of the STFT can extract the residual χ 2 Failure information statistics q of the test method k The frequency domain fault features are input into the CNN neural network; Finally, the outputs of the time domain characteristic fault detection network and the frequency domain fault detection network of STFT are used through the decision network to obtain the fault information of the combined navigation systems of SINS / GNSS and SINS / DVL / COMPASS of surface ships.

2. A ship integrated navigation fault detection method based on cascade neural network according to claim 1, characterized in that: In S1, the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for surface ships are established as follows: In the formula: x k is the estimated error of the integrated navigation state at time k; k is the combined navigation measurement error at time k; f(.) is the nonlinear state transfer equation; H k is the measurement equation at time k; D k is the coefficient matrix of the system measurement noise at time k; w k is the system process noise at time k; v k is the system measurement noise at time k; k represents the time when the system is working.

3. A ship integrated navigation fault detection method based on cascade neural network according to claim 2, characterized in that: In S2, the variables of the integrated navigation state estimation error are selected as follows: Where: is the pitch angle error; is the roll angle error; is the heading angle error; δv E is the eastward velocity error; δv N is the north velocity error; δv U is the celestial velocity error; δL is the latitude error; δλ is the longitude error; δh is the height error; ε E is the gyro zero bias in the east direction; N is the gyro bias in the north direction; U The gyro zero bias for the celestial direction; is the accelerometer zero bias in the east direction; is the accelerometer bias in the north direction; is the accelerometer zero bias in the celestial direction; ε and They are the gyroscope bias and accelerometer bias respectively.

4. A ship integrated navigation fault detection method based on cascade neural network according to claim 3, characterized in that: In S3, the variables of the integrated navigation measurement error are selected as follows: Where: is the measurement error of GNSS; is the measurement error of DVL and COMPASS; is the position error of the SINS system at time k; is the position error of the GNSS system at time k; is the velocity error of the SINS system at time k; is the velocity error of the GNSS system at time k; is the velocity error of the DVL system at time k; is the heading angle error of the SINS system at time k; is the heading angle error of the COMPASS system at time k.

5. A ship integrated navigation fault detection method based on cascade neural network according to claim 4, characterized in that: In S4, the combined navigation system equations of SINS / GNSS and SINS / DVL / COMPASS for discretized surface ships are obtained as follows: In the formula, is the one-step prediction of the state error; P k|k-1 is the state error covariance matrix one-step prediction; Φ k is the state transfer matrix of the discretization of f(.); B k is the coefficient matrix of the discretized system process noise; Q k is the system noise covariance matrix at time k; K k is the filter gain at time k; I is the unit matrix; where, E(.) is the expectation equation; is the one-step prediction of the state error; P k|k-1 is the state error covariance matrix for one-step prediction; R k is the measurement noise covariance matrix; P k is the covariance matrix generated during the filtering process of the integrated navigation system.

6. A ship integrated navigation fault detection method based on cascade neural network according to claim 5, characterized in that: In S5, the residual χ is obtained. 2 The fault information statistics method of the inspection method is as follows: S51: Get the residuals of the surface ship SINS / GNSS and SINS / DVL / COMPASS integrated navigation systems at time k: Where: γ k represents the residual error of the integrated navigation system at time k; One-step prediction for state error; S52: Get the theoretical value of the innovation variance: Where: S k is the theoretical value of the innovation variance; P k|k-1 is the state error covariance matrix for one-step prediction; R k is the measurement noise covariance matrix; S53: Get the residual χ 2 The fault information statistics of the test method is Where: q k χ represents the residual 2 Failure information statistics of the test method.

Images