A navigation situation estimation method for unmanned surface vehicle

Abstract

The application discloses a kind of unmanned surface vehicle's sailing situation estimation method, comprising: monitoring the high-dimensional time-varying data generated when unmanned surface vehicle sails, using entropy weight method to measure sailing situation estimation index set, based on entropy value judges the discrete degree of each index.Based on ST-LSTM network, the sailing situation of unmanned vehicle under complex sea conditions is estimated, by extending the forget gate of neural network to time domain and space domain, to process the information related to sailing situation of input data in real time, learn the time sequence correlation and reliability of sailing data, update the weight of information stored in memory cell.Based on ST-LSTM network, the time-varying data of unmanned surface vehicle is processed and analyzed, to estimate the sailing situation of unmanned surface vehicle in real time.The application has improved the task execution efficiency and sailing safety of unmanned surface vehicle and the like advantages.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of intelligent information technology, and particularly relates to a navigation situation estimation method for an unmanned surface vehicle. BACKGROUND

[0002] With the continuous exploration of marine resources in China, the unmanned surface vehicle technology has attracted extensive attention from researchers, and has shown important advantages in performing tasks such as coastal defense, sea rescue, exploration and development. However, in a dynamic and complex sea area, there are often unpredictable dangers. How to realize high-precision navigation situation estimation of the unmanned vehicle and ensure its safe execution of tasks has become one of the research hotspots of unmanned driving technology. However, the existing situation estimation method is difficult to process the time-varying data of the unmanned vehicle in real time, cannot accurately identify the complex and changeable navigation environment, and the situation estimation model constructed has the characteristics of large calculation amount and high time delay, which greatly reduces the maneuverability and flexibility of the unmanned surface vehicle. Therefore, in the face of complex sea conditions and task targets, a navigation situation estimation method for an unmanned surface vehicle is needed to improve the efficiency and navigation safety of the unmanned surface vehicle. SUMMARY

[0003] The purpose of the present application is to provide a navigation situation estimation method for an unmanned surface vehicle to solve the problems existing in the prior art.

[0004] To achieve the above purpose, the present application provides a navigation situation estimation method for an unmanned surface vehicle, comprising:

[0005] S1. Obtain an unmanned vehicle navigation situation estimation index set; the index set is high-dimensional time-varying data generated when the unmanned vehicle navigates; normalize the index set based on a threshold method; use an entropy weight method to measure the normalized index set to obtain the entropy weight value of each navigation index in the index set;

[0006] S2. Extend the forgetting gate of the ST-LSTM network to the time domain and the space domain, process the input data and the information related to the navigation situation in the time domain and the space domain, learn the time sequence correlation and reliability of the navigation data based on the entropy weight, and update the weight value of the information stored in the memory unit;

[0007] The S2 comprises the following steps:

[0008] S21. Take the hidden state at the time and the input information at the current time t as the input of the forgetting gate, and forget the redundant information;

[0009] S22. Input the hidden state information and the input information into the input gate, store the information with high importance degree, and update the state of the memory cell;

[0010] S23. The output hidden state should carry the information of the input information u

[0011] S3. Based on the ST-LSTM network processing the high-dimensional time-varying data of the unmanned ship, a sufficient representation quantity of the navigation situation estimation index set is outputted, and the navigation situation of the unmanned ship is estimated.

[0012] Optionally, the indexes in the index set in S1 include:

[0013] Average endurance, average endurance time, average navigation speed, average horizontal positioning accuracy, average inclination accuracy, average obstacle detection distance, average automatic obstacle avoidance number, average energy consumption.

[0014] Optionally, the normalization processing is specifically:

[0015] The navigation index f it of the i-th unmanned ship at time t is normalized to obtain an index g it : Wherein, a and b are the maximum and minimum values of f i , fi is the navigation index of the i-th unmanned ship, i=1, 2, 3…m.

[0016] Optionally, the entropy value of the navigation index is calculated by:

[0017]

[0018] Wherein, E i is the entropy value of the i-th navigation index, p it is the proportion of the i-th navigation index at time t in the index set.

[0019] Optionally, the entropy weight is calculated by:

[0020]

[0021] Wherein, g i is the difference coefficient of the navigation index of the i-th unmanned ship.

[0022] Optionally, the ST-LSTM network selects the navigation index with the entropy weight greater than the average entropy weight as the input information of the ST-LSTM network at the current time.

[0023] Optionally, the average entropy weight is:

[0024] Optionally, the ST-LSTM network input gate controls the modulation of the input information u j,tThe forgetting gate controls the influence of the memory cell at the t-1 moment on the state at the current moment.

[0025] Optionally, the update formula of the memory cell of the ST-LSTM network is as follows:

[0026]

[0027] wherein c j,t represents the memory cell of the jth cell at the time t, represents the element-wise product, c j,t-1 and c j-1,t respectively represent the memory cell of the previous neuron at the previous moment and the current neuron at the current moment.

[0028] Optionally, the update formula of the output gate of the ST-LSTM network is as follows:

[0029] h j,t = o j,t ⊙ tanh(c j,t )

[0030] wherein tanh(c j,t ) represents the hyperbolic tangent transformation of the memory cell.

[0031] The technical effects of the present application are as follows:

[0032] 1. The present application is aimed at the high-dimensional data generated by the navigation of the unmanned surface vehicle in a complex environment, the entropy weight of each situation estimation index is calculated by adopting the entropy weight method, the dispersion degree between indexes is judged, the noise of the high-dimensional data is reduced, the indexes with high entropy weight values are extracted, and the extracted indexes are used as the input information of the ST-LSTM network.

[0033] 2. The present application estimates the navigation situation of the unmanned surface vehicle based on the ST-LSTM neural network, the input of the neuron is processed in real time by setting the time domain and space domain forgetting gates, the time sequence correlation between the high-dimensional nonlinear data is learned in combination with the navigation information at the previous moment, the internal coupling relationship between the data is mined, the navigation situation of the unmanned surface vehicle is accurately estimated, and the efficiency and safety of the task execution are improved. DETAILED DESCRIPTION

[0034] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the present application illustrated in the drawings and their descriptions are used to explain the present application and do not limit the present application. In the drawings:

[0035] Figure 1 The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the present application illustrated in the drawings and their descriptions are used to explain the present application and do not limit the present application. In the drawings: DETAILED DESCRIPTION

[0036] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other in the case of no conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0037] A navigation situation estimation method of an unmanned surface vehicle, comprising the following steps:

[0038] S1, monitoring high-dimensional time-varying data generated when the unmanned surface vehicle navigates to form an unmanned vehicle navigation situation estimation index set. The threshold method is used to normalize the index set to unify the dimensions of different indexes. The entropy weight method is used to measure the navigation situation estimation index set, the dispersion degree of each index is judged based on the entropy value, and the index with higher entropy weight value is selected to provide data support for the navigation situation estimation of the unmanned vehicle.

[0039] S2, estimating the navigation situation of the unmanned vehicle under complex sea conditions based on the ST-LSTM network, extending the forgetting gate of the neural network to the time domain and the spatial domain to process the information related to the input data and the navigation situation in the two domains, learning the time sequence correlation and reliability of the navigation data, and updating the weight value of the information stored in the memory cell.

[0040] S2 comprises the following steps:

[0041] S21, taking the hidden state at t-1 time and the input information at the current time t as the input of the forgetting gate, and forgetting the redundant information.

[0042] S22, inputting the hidden state information and the input information into the input gate, storing the information with high importance degree, and updating the state of the memory cell.

[0043] S23, using Tanh and Sigmoid activation functions in the output gate to output the information carried by the hidden state.

[0044] S3, processing and analyzing the time-varying data of the unmanned surface vehicle based on the ST-LSTM network, outputting a sufficient representation of the navigation situation estimation index set, and estimating the navigation situation of the unmanned surface vehicle in real time.

[0045] The navigation situation estimation index set of the unmanned surface vehicle includes: average endurance mileage of the unmanned vehicle, average endurance time, average navigation speed, average horizontal positioning accuracy, average inclination accuracy, average obstacle detection distance, average number of automatic obstacle avoidance, and average energy consumption.

[0046] The threshold method is used to process the i-th unmanned vehicle navigation index f it The normalized processing index g it .

[0047] The smaller the information entropy value of the situation estimation index is, the greater the influence on the unmanned ship navigation situation estimation is, and the greater the discrete degree is. In particular, if the information entropy values of the indexes are all equal, the index is negligible in situation estimation.

[0048] The ST-LSTM network selects the index with an entropy weight value greater than the average entropy weight as the input information at the current time.

[0049] The ST-LSTM network input gate controls the modulation of the input information u j,t The degree of updating the memory unit, and the forget gate controls the influence of the memory unit at t-1 time on the state at the current time.

[0050] Embodiment one

[0051] As Figure 1 shown, the embodiment provides a navigation situation estimation method for an unmanned surface vehicle, referring to Figure 1 .

[0052] The navigation situation estimation method for the unmanned surface vehicle comprises the following steps:

[0053] S1, monitoring the high-dimensional time-varying data generated by the unmanned surface vehicle during task execution, obtaining an unmanned ship navigation situation estimation index set, including: average endurance, average endurance time, average navigation speed, average horizontal positioning accuracy, average inclination accuracy, average obstacle detection distance, average number of automatic obstacle avoidance, and average energy consumption.

[0054] Based on the threshold method, the i-th unmanned ship navigation index f it is normalized to obtain the index g it : Where a and b are the maximum and minimum values of f i , fi is the i-th unmanned ship navigation index, and i = 1, 2, 3…m.

[0055] By using the entropy weight method to measure the unmanned ship navigation situation estimation index set, the discrete degree of each index is judged based on the entropy value. The smaller the information entropy value is, the greater the influence of the index on the unmanned ship navigation situation estimation is, and the greater the discrete degree is. In particular, if the information entropy values of the indexes are all equal, the index is negligible in situation estimation. By calculating the entropy weight of each index in the index set, data support is provided for the unmanned ship navigation situation estimation. The steps of the entropy weight method are as follows:

[0056] (1) Calculate the proportion p it of the i-th index in the index set at time t:

[0057]

[0058] (2) Calculate the entropy value E of the i-th unmanned ship navigation situation estimation index i :

[0059]

[0060] (3) Calculate the difference coefficient g of the i-th unmanned ship navigation index i : g i = 1-E i

[0061] (4) Calculate the entropy weight w of the i-th unmanned ship navigation index i :

[0062] By using ST-LSTM to estimate the navigation situation of unmanned ship in complex environment, based on the entropy weight of each unmanned ship navigation index, the index with entropy weight greater than the average entropy weight is selected as the input information of ST-LSTM (Spatio-Temporal LSTM) at the current time. The average entropy weight of unmanned ship index is:

[0063] S2, in order to process the noise of unmanned ship navigation data in complex environment, the forgetting gate of traditional LSTM is expanded to two concurrent domains, namely time domain and space domain, to analyze the input data related to navigation situation in two domains.

[0064] S21, the hidden state at t-1 time and the input information at current time t are taken as the input of the forgetting gate, the forgetting gate controls the influence of the memory unit at t-1 time on the current time state, learns the time sequence correlation and reliability of the data, and updates the weight of the information stored in the memory unit. The formula of ST-LSTM network is:

[0065]

[0066] Where, j∈{1,...,}, t∈{1,...,} represent the index of ST-LSTM network neuron and time respectively, x j,t ,i j,t ,o j,t ,u j,t is the input information, input gate, output gate and modulated input of the network, and correspond to the two forgetting gates of the context information of ST-LSTM network, namely space domain and time domain. h j,t-1 and h j-1,t represent the hidden state of the previous neuron at the previous time and the current time respectively. σ is Sigmoid activation function, tanh is hyperbolic tangent function. M: is a function transformation composed of model parameters, wherein, is a matrix of input information, D is the dimension of input information x j,t , d is the number of LSTM cell states.

[0067] S22, input the hidden state information and the input information into the input gate, the input gate controls the modulation of the input information u j,t at time t, stores information with high importance degree, and updates the state of the memory cell. The memory cell update formula is as follows:

[0068]

[0069] wherein, c j,t represents the memory cell of the jth unit at time t. ⊙ represents element-wise multiplication, c j,t-1 and c j-1,t represent the memory cells of the previous neuron at the previous time and the current time, respectively.

[0070] S23, use Tanh and Sigmoid activation functions in the output gate, control the amount of information output by the memory cell, output the information carried by the hidden state, and generate the output of the ST-LSTM network. The output gate update formula is as follows:

[0071] h j,t = o j,t ⊙ tanh(c j,t )

[0072] wherein, tanh(c j,t ) represents a hyperbolic tangent transformation of the memory cell.

[0073] S3, based on the effective analysis of the high-dimensional data set of the unmanned ship by the ST-LSTM network, output a sufficient representation of the navigation situation estimation index set, and then accurately estimate the navigation situation of the unmanned ship, and improve the operation efficiency and safety of the unmanned ship.

[0074] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of estimating a navigation situation of an unmanned surface vehicle, characterized in that, The method comprises the following steps: S1. Obtain an unmanned ship navigation situation estimation index set; the index set is high-dimensional time-varying data generated when the unmanned ship navigates; The index set is normalized based on a threshold method; the normalized index set is measured by using an entropy weight method, and entropy weight values of each navigation index in the index set are obtained; S2. Extend the forgetting gate of the ST-LSTM network to the time domain and the space domain, process information related to the navigation situation of input data in the time domain and the space domain, learn the time sequence correlation and reliability of navigation data based on the entropy weight, and update the weight of information stored in the memory unit; The S2 comprises the following steps: S21. Take the hidden state at t-1 time and the input information at the current time t as the input of the forgetting gate, and forget the redundant information; S22. Input the hidden state information and the input information into the input gate, store the information with high importance, and update the state of the memory cell; S23. Use Tanh and Sigmoid activation functions in the output gate to output the information that the hidden state should carry; S3. Process and analyze the high-dimensional time-varying data of the unmanned ship based on the ST-LSTM network, output a sufficient representation of the navigation situation estimation index set, and estimate the navigation situation of the unmanned ship.

2. The method of claim 1, wherein The indexes in the index set in S1 include: Average endurance, average endurance time, average navigation speed, average horizontal positioning accuracy, average inclination accuracy, average obstacle detection distance, average automatic obstacle avoidance number, and average energy consumption.

3. The method of claim 1, wherein The normalization process is as follows: The navigation index f of the i-th unmanned surface vessel at time t is based on the threshold method. it Normalization is performed to obtain the index g it : Where a and b are f i The maximum and minimum values ​​of f i Let i be the navigation index of the i-th unmanned surface vessel, where i = 1, 2, 3…m.

4. The sailing situation estimation method of the unmanned surface vehicle according to claim 3, characterized in that, The entropy value of the navigation index is calculated by the following formula: wherein E i is the entropy value of the i-th navigation indicator, p it is the proportion of the i-th navigation indicator in the indicator set at time t.

5. The method of claim 4, wherein, The entropy weight is calculated by the following formula: wherein g i is the difference coefficient of the navigation index of the ith unmanned ship, the difference coefficient g i = 1 - E i .

6. The method of claim 5, wherein, The ST-LSTM network selects the navigation index with the entropy weight greater than the average entropy weight as the input information of the current time of the ST-LSTM network.

7. The method of claim 6, wherein, The average entropy weight is:

8. The method of claim 1, wherein, The ST-LSTM network input gate controls modulation input information u at time t j,t The degree of updating the memory unit, the forget gate controls the influence of the memory unit at time t-1 on the state at the current time.

9. The method of claim 1, wherein, The update formula of the memory unit of the ST-LSTM network is as follows: where c j,t denotes the memory cell of the jth unit at time t, denotes the element-wise product, c j,t-1 and c j-1,t denote the memory cell of the previous neuron at the previous time and the current neuron at the current time, respectively, j e {1,..., J} and t e {1,..., T} denote the indices of the ST-LSTM network neuron and time, respectively, i j,t , u j,t are the input gate and the modulation input of the network, and correspond to the two forget gates of the context information of the ST-LSTM network, i.e., the spatial domain and the temporal domain, respectively.

10. The method of claim 9, wherein, The update formula of the output gate of the ST-LSTM network is as follows: h j,t = o j,t ⊙ tanh(c j,t ) wherein tanh(c j,t ) denotes a hyperbolic tangent transformation of the memory cell, o j,t is an output gate of the network.

Images