CN114911159A - Simulated bat aircraft depth control method based on T-S fuzzy neural network - Google Patents

Abstract

The invention relates to a depth control method of a simulated manta ray aircraft based on a T-S fuzzy neural network, which is characterized in that depth deviation and depth change rate in the navigation process of the aircraft are collected as input of neural network training, and corresponding tail fin angles are used as output of the neural network training; then, a T-S type fuzzy neural network is established, the neural network is trained by utilizing a training set, and then the trained neural network is tested by utilizing a testing set; and finally, the tested T-S neural network controller is used for controlling the tail fin angle of the simulated bat ray aircraft in the swimming process, so that the tail fin angle of the aircraft is adjusted in real time according to a control value, and the purpose of depth control of the aircraft is finally achieved. Experiments show that the method for controlling the depth of the simulated manta ray aircraft based on the T-S fuzzy neural network has high accuracy and reliability, and the method has practical value in the aspect of aircraft depth control.

Description

Depth control method of manta ray-like vehicle based on T-S fuzzy neural network

technical field

The invention relates to the field of underwater vehicle motion control, and relates to a method for depth control of a manta ray-imitation vehicle, in particular to a manta ray-imitation vehicle depth control method based on a T-S fuzzy neural network.

Background technique

As an indispensable part of marine science and technology, underwater unmanned vehicles play an important role in many fields such as military, civil, science and technology. Traditional underwater vehicles mainly use propellers for propulsion, but in the face of increasingly complex mission requirements, traditional propellers for propulsion have exposed many shortcomings such as high energy consumption, large volume, and low efficiency.

Biomimetic scientific research provides a new idea for developing new underwater vehicles to solve the problems of traditional vehicles. The manta-like vehicle is a new type of bionic vehicle that adopts the central fin/opposite fin mode for propulsion, and has many advantages such as small size, low noise and high maneuverability.

The depth control of the manta-like vehicle is of great significance for the underwater operation of the vehicle. The existing methods for depth control of underwater vehicles are mainly aimed at traditional vehicles that use propellers for propulsion. However, due to the difference in structure and propulsion mechanism between the manta ray-like vehicle and the traditional vehicle, the depth control method suitable for the traditional vehicle is difficult to be directly used for the depth control of the manta-ray vehicle.

The underwater vehicle depth control method mentioned in the invention patent CN113050666A relies on the dynamic model of the underwater vehicle. However, since the manta-like vehicle is a complex system with strong nonlinearity and strong coupling, it is very difficult to establish its dynamic model, so this kind of method is not suitable for the depth control of the manta-like vehicle.

SUMMARY OF THE INVENTION

technical problem to be solved

In order to avoid the deficiencies of the prior art, the present invention proposes a depth control method based on the T-S fuzzy neural network for the imitation manta ray vehicle, and the problem to be solved is the depth control of the imitation manta ray vehicle.

Since the change of the deflection angle and direction of the tail fin of the manta-like vehicle can change the magnitude and direction of the pitching moment acting on the manta-like vehicle, the depth control of the manta-like vehicle can be realized by using this pitching moment. In the invention, depth control is achieved by changing the angle of the tail fin of the manta-like vehicle.

Technical solutions

A method for depth control of imitation manta ray vehicle based on T-S fuzzy neural network is characterized in that the steps are as follows:

Step 1: Construct the training set and test set for training the TS fuzzy neural network by using the depth deviation _{ed , the depth deviation rate ec d ,} and the corresponding tail fin angle γ at different moments during the swimming process of the imitation manta ray vehicle;

Step 2: The T-S fuzzy neural network includes an input layer, a fuzzification layer, a fuzzy inference layer, a normalization layer and an output layer;

The first layer is the input layer and is connected with the input vector, and the input value is transmitted to the node. The number of nodes in this layer is equal to the dimension of the input vector;

The second layer is the fuzzification layer, in which each node represents a fuzzy linguistic variable value, and the membership degree of each input belonging to the fuzzy set of each linguistic variable value is calculated by the membership function;

The membership function:

Among them, n is the number of inputs; m _i is the number of fuzzy divisions of _xi ; c _ij is the central value of the membership function; σ _ij is the width of the membership function, and both c _ij and σ _ij are adjustable parameters;

The third layer is the fuzzy inference layer, in which each node represents a fuzzy rule, and the fitness of each rule is calculated; the weight α _j of each rule adopts the product method:

where i ₁ ∈{1,2,…,m ₁ },i ₂ ∈{1,2,…,m ₂ },…,in ∈{1,2,…, _{m n }} ; j=1, 2,…,m;

Use the weighted average method to calculate the output value y of the fuzzy model:

In the formula, α _i represents the proportion of the weight of the i-th rule in the total output, that is, the weight, and y _i represents the output of the i-th rule;

The fourth layer is a normalized layer, in which each node is correspondingly connected to the nodes of the third layer, and the number of nodes is the same as that of the third layer, and normalized calculation is performed;

The fifth layer is the input layer, in which each node represents a clear output for clear calculation;

Step 3: Use the training set of Step 1 to learn and train the TS neural network. During learning and training, the center value c _ij of the membership function in the second layer, the width value σ _ij and the connection weight sum in the fifth layer are continuously updated. ω _i , the steps are as follows:

First take the error cost function as:

In the formula, y _di represents the expected output, and y _i represents the actual output;

Then according to the gradient descent method, the parameters are modified according to the following formula:

In the formula, β>0 is the learning rate;

Step 4: Use the constructed test set to test the trained T-S neural network, and use the tested T-S-based fuzzy neural network as a depth controller to control the angle of the tail fin of the imitation manta ray vehicle; The depth deviation and depth deviation rate collected by the sensor during the moving process are used as the input, and the caudal fin angle is used as the output, so that the steering gear rotates the corresponding angle according to the output caudal fin angle, changes the pitching moment of the vehicle, and realizes the depth control of the manta-like vehicle.

beneficial effect

The invention proposes a depth control method based on T-S fuzzy neural network for imitation manta ray vehicle. First, the depth deviation and depth change rate during the navigation process of the vehicle are collected as the input of neural network training, and the corresponding tail fin angle is used as the neural network. The output of network training is used to construct the training set and test set for T-S fuzzy network training; then a T-S fuzzy neural network is established, and the training set is used to train the neural network, and then the test set is used to train the neural network obtained. Carry out the test; finally, the tested T-S neural network controller is used to control the angle of the tail fin during the swimming process of the imitation manta ray vehicle, so that the angle of the tail fin of the vehicle can be adjusted in real time according to the control value, and finally the depth control of the vehicle is achieved. Purpose. Experiments show that the proposed method for depth control of manta ray-like vehicle based on T-S fuzzy neural network has high accuracy and reliability, and the method has practical value in the depth control of vehicle.

The present invention has the following beneficial effects:

1. The depth control method of the imitation manta ray vehicle based on the T-S fuzzy neural network provided in the present invention provides an effective method for the depth control of the imitation manta ray vehicle.

2. The method provided by the present invention realizes the depth control of the aircraft by establishing a T-S fuzzy neural network to calculate the control angle of the tail fin, has high efficiency and prediction accuracy, and improves the stability and reliability of the depth control of the manta ray aircraft. .

3. The present invention uses the prototype to carry out experiments, realizes the effective control of the depth of the imitation manta ray aircraft, and obtains the depth-fixing curve of the aircraft. The feasibility and authenticity of the method provided by the present invention are verified in the real working environment.

Description of drawings

Fig. 1 is the control block diagram of the present invention;

Fig. 2 is the structure diagram of T-S fuzzy neural network of the present invention;

Fig. 3 is the depth determination curve diagram of the aircraft obtained from the experiment;

Detailed ways

The present invention will now be further described in conjunction with the embodiments and accompanying drawings:

The specific implementation steps of this method are as follows:

The depth deviation _{ed and the depth deviation rate ec d collected} when the imitation manta ray vehicle swims in the pool are used as the training input of the model, and the corresponding caudal fin angle γ is used as the training output of the model. Therefore, the present invention has two input layer nodes and one output layer node. The TS fuzzy neural network in the present invention has a total of five layers, as shown in FIG. 2 .

The specific steps of using the T-S fuzzy neural network for the depth control of the manta-like vehicle are as follows:

1. Create a training set.

The swimming pool experiment was carried out using the manta ray-like vehicle, and the depth deviation _{ed , the depth deviation rate ec d and} the corresponding tail fin angle γ at different times during the swimming process of the vehicle were collected by the sensors on the vehicle. The depth deviation _{ed and the depth deviation rate ec d obtained from the experiment are used to form X=[ed , ec d ] as the} network input, and the tail fin angle γ is used as the network expected output y _di . The training set and test set for training the TS fuzzy neural network are constructed.

2. Establish T-S fuzzy neural network and conduct training and testing.

(1) Select membership function and weight calculation

Suppose there is input X=[x ₁ , x ₂ ,...,x _n ], first calculate the membership degree of each input variable according to the membership degree function

The membership function of the present invention adopts a Gaussian function, as shown in the following formula:

Among them, n is the number of inputs; m _i is the number of fuzzy divisions of _xi ; c _ij is the central value of the membership function; σ _ij is the width of the membership function, and both c _ij and σ _ij are adjustable parameters.

The weight α _j of each rule is calculated by the product method, as shown in the following formula:

where i ₁ ∈{1,2,…,m ₁ },i ₂ ∈{1,2,…,m ₂ },…,in ∈{1,2,…, _{m n }} ; j=1, 2,…,m;

Use the weighted average method to calculate the output value y of the fuzzy model:

In the formula, α _i represents the proportion (weight) of the ith rule in the total output, and y _i represents the output of the ith rule.

(2) T-S fuzzy neural network construction

The T-S neural network constructed according to the T-S fuzzy algorithm is shown in Figure 2. T-S fuzzy neural network has 5 layers including: input layer, fuzzification layer, fuzzy inference layer, normalization layer and output layer

The first layer is the input layer, which is directly connected to the input vector. Its function is to transmit the input value to the node. The number of nodes in this layer is equal to the dimension of the input vector.

The second layer is the fuzzification layer, each node of this layer represents a fuzzy linguistic variable value, and its function is to calculate the membership degree of each input belonging to the fuzzy set of each linguistic variable value through the membership degree function.

The third layer is the fuzzy inference layer, each node represents a fuzzy rule, and the function of this layer is to calculate the fitness of each rule.

The fourth layer is the normalized layer, each node of this layer is connected to the node of the third layer correspondingly, and the number of nodes is the same as that of the third layer. Its role is to perform normalization calculations.

The fifth layer is the input layer, and each node represents a clear output. Its role is to perform clearing calculations.

(3) Selection of learning algorithm

Since the number of fuzzy divisions of the input value in the TS fuzzy neural network is predetermined, it is only necessary to continuously update the center value c _ij , the width value σ _ij and the width value σ ij of the membership function in the second layer during the training process. Connect the weights and ω _i . In the present invention, the parameters of the neural network are determined by the gradient descent method to obtain the ideal mapping relationship between the input and the output, and the specific steps are as follows.

First take the error cost function as:

In the formula, y _di represents the expected output, and y _i represents the actual output.

Then according to the gradient descent method, the parameters are modified according to the following formula:

In the formula, β>0 is the learning rate.

(4) Training and testing of T-S fuzzy neural network

First, use the constructed training set to train the T-S neural network, and then use the constructed test set to test the trained T-S neural network to verify its effectiveness.

T-S fuzzy neural network training mainly includes the following steps:

(a) Initialize the basic parameters of the neural network, including the network input layer, output layer, the number of nodes in the middle layer, the maximum number of iterations of the algorithm, the learning rate β, the central value of the membership function c _ij and the width value σ _ij .

(b) For each set of input parameters in the training set, the corresponding actual output _yi is calculated by the neural network.

(c) Update the center value c _ij , the width value σ _ij of the membership function in the second layer, and the connection weight and ω _i in the fifth layer according to the error between the actual output _yi and the expected output y _di .

(d) Judging the end condition, if the number of iterations reaches the set maximum number of iterations of the algorithm, the training ends; otherwise, return to step (b) to continue training.

The T-S neural network obtained after training is validated using the test set to determine the effectiveness of the resulting T-S neural network.

3. The tested depth controller based on T-S fuzzy neural network is used to control the caudal fin angle of the manta ray vehicle. The depth control of the manta ray-like vehicle is carried out by using the final controller based on T-S fuzzy neural network. During the swimming process of the vehicle, the sensor collects the depth data of the manta-like vehicle, and the depth deviation and the depth deviation rate are calculated by the lower computer and used as the input of the controller to obtain the corresponding control value of the tail fin angle. Send the control value to the tail fin servo to rotate the corresponding angle to control the depth of the imitation manta ray vehicle.

Experiments are carried out in the pool by using the prototype to verify the depth control method of the imitation manta ray vehicle based on the T-S neural network proposed by the present invention. The depth variation curve shown in Figure 3 is obtained through experiments. The experimental results show that the controller based on the T-S fuzzy neural network can effectively control the depth of the manta-like vehicle.

Claims (1)

1. a kind of imitation manta ray vehicle depth control method based on T-S fuzzy neural network, it is characterized in that step is as follows: Step 1: Construct the training set and test set for training the TS fuzzy neural network by using the depth deviation _{ed , the depth deviation rate ec d ,} and the corresponding tail fin angle γ at different moments during the swimming process of the imitation manta ray vehicle; Step 2: The T-S fuzzy neural network includes an input layer, a fuzzification layer, a fuzzy inference layer, a normalization layer and an output layer; The first layer is the input layer and is connected to the input vector, and the input value is transmitted to the node. The number of nodes in this layer is equal to the dimension of the input vector; The second layer is the fuzzification layer, in which each node represents a fuzzy linguistic variable value, and the membership degree of each input belonging to the fuzzy set of each linguistic variable value is calculated by the membership function; The membership function:

Among them, n is the number of inputs; m _i is the number of fuzzy divisions of _xi ; c _ij is the central value of the membership function; σ _ij is the width of the membership function, and both c _ij and σ _ij are adjustable parameters; The third layer is the fuzzy inference layer, in which each node represents a fuzzy rule, and the fitness of each rule is calculated; The weight α _j of each rule adopts the product method:

where i ₁ ∈{1,2,…,m ₁ },i ₂ ∈{1,2,…,m ₂ },…,in ∈{1,2,…, _{m n }} ; j=1, 2,…,m;

Use the weighted average method to calculate the output value y of the fuzzy model:

In the formula, α _i represents the proportion of the weight of the i-th rule in the total output, that is, the weight, and y _i represents the output of the i-th rule; The fourth layer is a normalized layer, in which each node is correspondingly connected to the nodes of the third layer, and the number of nodes is the same as that of the third layer, and normalized calculation is performed; The fifth layer is the input layer, in which each node represents a clear output for clear calculation; Step 3: Use the training set of Step 1 to learn and train the TS neural network. During learning and training, the center value c _ij of the membership function in the second layer, the width value σ _ij and the connection weight sum in the fifth layer are continuously updated. ω _i , the steps are as follows: First take the error cost function as:

In the formula, y _di represents the expected output, and y _i represents the actual output; Then according to the gradient descent method, the parameters are modified according to the following formula:

In the formula, β>0 is the learning rate; Step 4: Use the constructed test set to test the trained T-S neural network, and use the tested T-S fuzzy neural network based on the test as a depth controller to control the angle of the tail fin of the imitation manta ray vehicle; The depth deviation and depth deviation rate collected by the sensor during the moving process are used as the input, and the caudal fin angle is used as the output, so that the steering gear rotates the corresponding angle according to the output caudal fin angle, changes the pitching moment of the vehicle, and realizes the depth control of the manta-like vehicle.

Images