Ship main and auxiliary power rotating speed control system based on two-channel neural network

Abstract

The invention relates to the technical field of ship main and auxiliary power rotating speed control, in particular to a ship main and auxiliary power rotating speed control system based on a two-channel neural network, which comprises the following steps: collecting operation parameters of a ship main power system and an auxiliary power system, and preprocessing the operation parameters; a two-channel convolutional neural network is constructed, and each channel uses multiple convolutional layers, a pooling layer and an activation function to extract the characteristics of operation parameters. According to the ship main and auxiliary power rotating speed control system based on the double-channel neural network, the double-channel neural network is adopted for main and auxiliary power switching control, and main power system parameters such as steam flow, steam pressure, steam temperature and steam turbine rotating speed of a steam turbine are input into one channel; auxiliary power system parameters such as voltage, current, frequency and rotating speed of the propulsion motor are input into the other channel, and synchronous control over the two channels in the switching process is guaranteed by considering information of the main power and the auxiliary power at the same time.

Description

Technical Field

[0001] The present invention relates to the technical field of ship main and auxiliary power speed control, and specifically to a ship main and auxiliary power speed control system based on a dual-channel neural network. Background Art

[0002] Ship main and auxiliary power speed control refers to the process of precisely managing the speeds of the main propulsion device (usually a steam turbine) and the auxiliary power system (such as a generator set or other power sources for providing non-propulsion functions) on a ship. This process aims to ensure that the ship can operate efficiently and safely under various working conditions, while meeting the requirements of navigation performance and energy conservation and emission reduction standards.

[0003] Current ship power system switching schemes usually rely on simple on-off control logic. Although this control method is simple and direct, it has significant limitations in practical applications, easily leading to sudden changes in power output, thus affecting the stability of the entire system. Therefore, a ship main and auxiliary power speed control system based on a dual-channel neural network is proposed to solve the above problems. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides a ship main and auxiliary power speed control system based on a dual-channel neural network, which has the advantages of synchronous control of the main and auxiliary power systems, etc., and solves the problem that current ship power system switching schemes usually rely on simple on-off control logic. Although this control method is simple and direct, it has significant limitations in practical applications, easily leading to sudden changes in power output, thus affecting the stability of the entire system.

[0005] To achieve the above object, the present invention provides the following technical solution: A ship main and auxiliary power speed control system based on a dual-channel neural network, including the following steps:

[0006] S1. Collect the operating parameters of the ship's main power system and auxiliary power system, and perform preprocessing of the operating parameters;

[0007] S2. Construct a dual-channel convolutional neural network, and each channel uses multiple convolutional layers, pooling layers, and activation functions to extract the features of the operating parameters;

[0008] S3. Calculate the attention scores of each channel, and adjust the weights of the feature maps according to these scores to obtain a fused feature map;

[0009] S4. Flatten the fused feature map into a one-dimensional vector in row-major order;

[0010] S5. Input the flattened one-dimensional vector into a fully connected layer, map it to the control signal space, and obtain a control signal;

[0011] S6. Design a smooth transition mechanism based on the control signal output by the fully connected layer to enable smooth transition and synchronous control of the main and auxiliary power systems during the switching process.

[0012] Preferably, the step S1 further includes the following sub-steps:

[0013] S1.1. Use a multi-channel data acquisition card to trigger all sensors to sample through a unified clock signal, and simultaneously collect the operating parameters of the main and auxiliary power systems at time t;

[0014] S1.2. Use limit filtering to process the outliers of the operating parameters;

[0015] S1.3. Perform standardization processing on the operating parameters:

[0016] S1.3.1. Define the minimum value min op and the maximum value max op ;

[0017] S1.3.2. Perform standardization processing on the operating parameters using min-max standardization:

[0018]

[0019] Preferably, the step S2 further includes the following sub-steps:

[0020] S2.1. Define the input layer and initialize the tensors of the main and auxiliary power systems:

[0021] Input of the main power system: Shape: (N, T, 4),

[0022] Input of the auxiliary power system: Shape: (N, T, 4),

[0023] where N is the number of samples, T is the time step, and 4 is the number of features per time step;

[0024] S2.2. Apply convolution operations to capture spatio-temporal features:

[0025] Convolution of the main power system:

[0026] where, is the output feature map at the t-th time step, B pp is the bias term, is the input data of the main power system at the t-th time step, ReLU() is the activation function, W pp is the weight matrix of the convolutional layer of the main power system,

[0027] Convolution of the auxiliary power system:

[0028] Among them, is the output feature map at the t-th time step, and B ap is the bias term, is the input data of the auxiliary power system at the t-th time step, ReLU() is the activation function, and W ap is the weight matrix of the convolutional layer of the main power system;

[0029] S2.3. Use a pooling layer to reduce the spatial dimension of the output feature map:

[0030] Main power system pooling:

[0031]

[0032] Among them, is the feature map after pooling at the t-th time step, z pp (t·s:t·s+k) is the feature map within the sliding window range, s is the stride, and k is the pooling window size,

[0033] Auxiliary power system pooling:

[0034]

[0035] Among them, is the feature map after pooling at the t-th time step, z ap (t·s:t·s+k) is the feature map within the sliding window range, s is the stride, and k is the pooling window size;

[0036] S2.4. Apply the ReLU activation function to introduce a non-linear transformation.

[0037] Preferably, the step S3 further includes the following sub-steps:

[0038] S3.1. Define the query vector Q, and transform and the feature maps into the key vector K and the value vector V respectively;

[0039]

[0040] Among them, W k and W V are the weight matrices for linear transformation;

[0041] S3.2. Use dot product similarity to calculate the attention score:

[0042] Main power system score z = Q T K z

[0043] Auxiliary power system score f = Q T K f ;

[0044] S3.3. Obtain the attention weights through the Softmax function:

[0045] α f = Softmax(score f )

[0046] α z = Softmax(score z );

[0047] S3.4. Reload according to the obtained attention weights and feature maps to obtain the weighted representation:

[0048] z f = α f V f

[0049]

[0050] S3.5. Concatenate the weighted feature maps z f and z z along the last dimension to form the comprehensive feature representation CHAR:

[0051] CHAR = Concat(z f , z z , dim = -1)

[0052] where dim = -1 indicates concatenation along the last dimension, with the shape (N, T, 8).

[0053] Preferably, the step S4 further includes the following sub-steps:

[0054] S4.1. Represent the i-th matrix in CHAR as CHAR i , then the elements of CHAR i can be represented as CHAR ijk :

[0055] where i is the sample index, i ∈ [1, N], j is the time step index, j ∈ [1, T], and k is the feature index, k ∈ [1, 8];

[0056] S4.2. List all the elements in CHAR i in row-major order:

[0057] CHAR flat,i=[CHAR i11 , CHAR i12 , ···, CHAR i18 , CHAR i21 , CHAR i22 , ···,CHAR i28 , ···, CHAR iT1 , CHAR iT2 , ···, CHAR iT8 ;

[0058] S4.3. For the comprehensive feature CHAR, the flattened vector CHAR flat The expression is:

[0059] CHAR flat = CHAR flat,1 ; CHAR flat,2 ; ···; CHAR flat,N .

[0060] Preferably, the step S5 includes the following sub-steps:

[0061] S5.1. Input CHAR flat into the first fully connected layer, and map the feature vector of this layer to a hidden space H;

[0062] H = ReLU(W fc1 · CHAR flat + b fc1 )

[0063] where W fc1 is the hidden layer weight matrix, and b fc1 is the hidden layer bias term;

[0064] S5.2. Input the hidden space H into the second connection layer, and this layer maps the feature vector to the control signal space to obtain the control signal y:

[0065] y = σ(W fc2 · H + b fc2 )

[0066] where W fc2 is the output layer weight matrix, b fc2 is the output layer bias term, σ(·) is the activation function, C PPS is the number of control signals of the main power system, C APS is the number of control signals of the auxiliary power system, d h is the dimension of the hidden layer,

[0067] y includes two parts: the control signal y zThe control signal y of the auxiliary power system f :

[0068] y z = σ(W fc2,PPS ·H + b fc2,PPS )

[0069] y f = σ(W fc2,APS ·H + b fc2,APS )。

[0070] Among them,

[0071] Preferably, the step S6 includes the following sub-steps:

[0072] S6.1. Map the output values of the control signal y of the main power system z and the control signal y of the auxiliary power system f back to the actual physical values:

[0073] S6.2. Design a smooth transition mechanism based on the physical values of the output values of the main and auxiliary power systems.

[0074] S6.2.1. Set the total switching time C time ;

[0075] S6.2.2. Mark the control signal of the main power system during the current ship operation as y z _old and the control signal of the auxiliary power system as y f _old;

[0076] S6.2.3. Use the linear interpolation formula to achieve smooth transition. The formula is as follows:

[0077]

[0078] Among them, y i,old is the i-th control signal of the current ship's main and auxiliary power systems, y i,target is the i-th control signal of the control system, and C is the current time step.

[0079] Compared with the prior art, the technical solution of this application has the following beneficial effects:

[0080] The ship's main and auxiliary power speed control system based on a dual-channel neural network uses a dual-channel neural network for main and auxiliary power switching control. One channel inputs the main power system parameters such as the steam flow rate, steam pressure, steam temperature, and steam turbine speed of the steam turbine, and the other channel inputs the auxiliary power system parameters such as the voltage, current, frequency, and speed of the propulsion motor. By considering the information of both the main and auxiliary powers simultaneously, synchronous control of the two during the switching process is ensured. Description of the Drawings

[0081] Figure 1 This is a schematic flow diagram of the main and auxiliary power speed control system of a ship based on a dual-channel neural network. Detailed Embodiments

[0082] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0083] Please refer to Figure 1 , the main and auxiliary power speed control system of a ship based on a dual-channel neural network in this embodiment includes the following steps:

[0084] S1. Collect the operation parameters of the ship's main power system and auxiliary power system, and perform preprocessing on the operation parameters;

[0085] S2. Construct a dual-channel convolutional neural network, and each channel uses multiple convolutional layers, pooling layers, and activation functions to extract the features of the operation parameters;

[0086] S3. Calculate the attention scores of each channel, and adjust the weights of the feature maps according to these scores to obtain a fused feature map;

[0087] S4. Flatten the fused feature map into a one-dimensional vector in row-major order;

[0088] S5. Input the flattened one-dimensional vector into a fully connected layer, map it to the control signal space, and obtain a control signal;

[0089] S6. Based on the control signal output by the fully connected layer, design a smooth transition mechanism to make the main and auxiliary power systems smoothly transition and synchronously control during the switching process.

[0090] In this embodiment, step S1 further includes the following sub-steps:

[0091] S1.1. Use a multi-channel data acquisition card to trigger all sensors to sample through a unified clock signal, and simultaneously collect the operation parameters of the main and auxiliary power systems at time t;

[0092] The operation parameters of the main power system include the steam flow rate Q of the steam turbine S , steam pressure P S , steam temperature T S and the rotational speed N of the steam turbine t , and the operation parameters of the auxiliary power system include the voltage V of the propulsion motorm , propulsion motor current I m , propulsion motor frequency f m and propulsion motor speed N m ;

[0093] S1.2. Use limit filtering to process outliers of operating parameters;

[0094] S1.2.1. Define the value of the current operating parameter as Num op , mean as μ op and standard deviation as σ op ;

[0095] S1.2.2. Calculate the magnitude of Num op and μ op ±3σ op . When Num op >μ op ±3σ op or Num op <μ op ±3σ op , reject all operating parameters at the current moment.

[0096] S1.3. Perform standardization processing on operating parameters:

[0097] S1.3.1. Define the minimum value min op and the maximum value max op of the current operating parameter;

[0098] S1.3.2. Use min - max standardization to perform standardization processing on operating parameters:

[0099]

[0100] In this embodiment, step S2 further includes the following sub - steps:

[0101] S2.1. Define the input layers of the main and auxiliary power systems and initialize the tensors:

[0102] Main power system input: Shape is (N, T, 4),

[0103] Auxiliary power system input: Shape is (N, T, 4),

[0104] where N is the number of samples, T is the time step, and 4 is the number of features per time step;

[0105] S2.2. Apply convolution operations to capture spatio - temporal features:

[0106] Main power system convolution:

[0107] Among them, is the output feature map at the t-th time step, and B pp is the bias term. is the input data of the main power system at the t-th time step, ReLU() is the activation function, and W pp is the weight matrix of the convolutional layer of the main power system.

[0108] Convolution of the auxiliary power system:

[0109] Among them, is the output feature map at the t-th time step, and B ap is the bias term. is the input data of the auxiliary power system at the t-th time step, ReLU() is the activation function, and W ap is the weight matrix of the convolutional layer of the main power system;

[0110] S2.3. Use the pooling layer to reduce the spatial dimension of the output feature map:

[0111] Pooling of the main power system:

[0112]

[0113] Among them, is the feature map after pooling at the t-th time step, and z pp (t·s:t·s+k) is the feature map within the sliding window range, s is the stride, and k is the pooling window size.

[0114] Pooling of the auxiliary power system:

[0115]

[0116] Among them, is the feature map after pooling at the t-th time step, and z ap (t·s:t·s+k) is the feature map within the sliding window range, s is the stride, and k is the pooling window size;

[0117] S2.4. Apply the ReLU activation function to introduce non-linear transformation.

[0118] f(x) = max(0, χ)

[0119] Among them, χ is the element in the feature map after pooling. When χ ≥ 0, then χ is output; otherwise, 0 is output.

[0120] In this embodiment, step S3 further includes the following sub-steps:

[0121] S3.1. Define the query vector Q, and take and The feature maps are respectively transformed into a key vector K and a value vector V;

[0122]

[0123] where, W k and W V are weight matrices for linear transformation;

[0124] S3.2. Calculate the attention scores using dot product similarity:

[0125] Main power system score z = Q T K z

[0126] Auxiliary power system score f = Q T K f ;

[0127] S3.3. Obtain the attention weights through the Softmax function:

[0128] α f = Softmax(score f )

[0129] α z = Softmax(score z );

[0130] S3.4. Reload the and feature maps according to the obtained attention weights to obtain the weighted representation:

[0131] z f = α f V f

[0132]

[0133] S3.5. Concatenate the weighted feature maps z f and z z along the last dimension to form the comprehensive feature representation CHAR:

[0134] CHAR = Concat(z f , z z , dim = -1)

[0135] where, dim = -1 indicates concatenation along the last dimension, with the shape of (N, T, 8).

[0136] In this embodiment, step S4 further includes the following sub-steps:

[0137] S4.1. Represent the i-th matrix in CHAR as CHAR i , then the elements of CHAR i can be represented as CHAR ijk :

[0138] where i is the sample index, i ∈ [1, N], j is the time step index, j ∈ [1, T], and k is the feature index, k ∈ [1, 8];

[0139] S4.2. List all the elements in CHAR i in row-major order:

[0140] CHAR flat,i = [CHAR i11 , CHAR i12 , ···, CHAR i18 , CHAR i21 , CHAR i22 , ···, CHAR i28 , ···, CHAR iT1 , CHAR iT2 , ···, CHAR iT8 ;

[0141] S4.3. For the comprehensive feature CHAR, the flattened vector CHAR flat is expressed as:

[0142] CHAR flat = CHAR flat,1 ; CHAR flat,2 ; ···; CHAR flat,N .

[0143] In this embodiment, step S5 includes the following sub-steps:

[0144] S5.1. Input CHAR flat into the first fully connected layer, and map the feature vector of this layer to a hidden space H;

[0145] H = ReLU(W fc1 · CHAR flat + b fc1 )

[0146] where W fc1 is the hidden layer weight matrix, and b fc1 is the hidden layer bias term;

[0147] S5.1.1. For each sample i, calculate the hidden layer output h i :

[0148] h i = W fc1 ·CHAR flat,i + b fc1

[0149] where CHAR flat,i is the i-th row vector of CHAR flat .

[0150] S5.1.2. Apply the activation function:

[0151] h i = max(0, W fc1 ·CHAR flat,i + b fc1 );

[0152] S5.1.3. For the entire batch N, write the above operations in matrix form:

[0153]

[0154] B fc1 is the broadcast form of the bias term.

[0155] S5.2. Input the hidden space H into the second connection layer, which maps the feature vector to the control signal space to obtain the control signal y:

[0156] y = σ(W fc2 ·H + b fc2 )

[0157] where W fc2 is the output layer weight matrix, b fc2 is the output layer bias term, σ(·) is the activation function, C PPS is the number of control signals of the main power system, C APS is the number of control signals of the auxiliary power system, d h is the dimension of the hidden layer,

[0158] y includes two parts: the control signal y z of the main power system and the control signal y f of the auxiliary power system:

[0159] y z = σ(W fc2,PPS ·H + b fc2,PPS )

[0160] y f = σ(W fc2,APS ·H + b fc2,APS ).

[0161] where,

[0162] In this embodiment, step S6 includes the following sub-steps:

[0163] S6.1 Map the output values of the control signal y of the main power system z and the control signal y of the auxiliary power system f back to the actual physical values:

[0164] The output value of the control signal y of the main power system z is the steam valve opening V Valve and the worm gear speed N turbine ; the output value of the control signal y of the auxiliary power system f is the motor voltage V motor and the frequency conversion frequency F inverter , and the mapping formula is as follows:

[0165] V Valve = y z,1 ·(V Valve,max - V Valve,min ) + V Valve,min

[0166] N turbine = y z,2 ·(N turbine,max - N turbine,min ) + N turbine,min

[0167] V motor = y f,1 ·(V motor,max - V motor,min ) + V motor,min

[0168] F inverter = y f,2 ·(F inverter,max - F inverter,min ) + F inverter,min

[0169] Among them, V Valve,max and V Valve,min are the maximum and minimum openings of the steam valve respectively, N turbine,max and N turbine,min are the maximum and minimum speeds of the turbine respectively, V motor,max and V motor,min are the maximum and minimum operating voltages of the motor respectively, and F inverter,max and F inverter,min are the maximum and minimum output frequencies of the frequency converter respectively.

[0170] S6.2. Design a smooth transition mechanism based on the physical values of the output values of the main and auxiliary power systems.

[0171] S6.2.1. Set the total switching time C time ;

[0172] S6.2.2. Mark the control signal of the main power system during the current ship operation as y z _old and the control signal of the auxiliary power system as y f _old;

[0173] y z _old includes the current steam valve opening V Valve _old and the current worm gear speed N turbine _old,

[0174] y f _old includes the current motor voltage V motor _old and the current frequency conversion frequency F inverter _old.

[0175] S6.2.3. Use the linear interpolation formula to achieve smooth transition. The formula is as follows:

[0176]

[0177] Where y i,old is the i-th control signal of the current ship's main and auxiliary power systems, y i,target is the i-th control signal of the control system, and C is the current time step.

[0178] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0179] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A main and auxiliary power speed control system for ships based on a dual-channel neural network, characterized in that, It includes the following steps: S1. Collect the operation parameters of the main power system and auxiliary power system of the ship, and perform preprocessing of the operation parameters; S2. Construct a dual-channel convolutional neural network, and each channel uses multiple convolutional layers, pooling layers, and activation functions to extract the features of the operation parameters; S3. Calculate the attention scores of each channel, and adjust the weights of the feature maps according to these scores to obtain a fused feature map; S4. Flatten the fused feature map into a one-dimensional vector in row-major order; S5. Input the flattened one-dimensional vector into a fully connected layer, map it to the control signal space, and obtain a control signal; S6. Based on the control signal output by the fully connected layer, design a smooth transition mechanism to make the main and auxiliary power systems smoothly transition and synchronously control during the switching process.

2. The main and auxiliary power speed control system for ships based on a dual-channel neural network according to claim 1, characterized in that The step S1 further includes the following sub-steps: S1.

1. Use a multi-channel data acquisition card to trigger all sensors to sample through a unified clock signal, and simultaneously collect the operation parameters of the main and auxiliary power systems at time t; S1.

2. Use limit filtering to process the outliers of the operation parameters; S1.

3. Perform standardization processing on the operation parameters: S1.3.

1. Define the minimum value min of the current operating parameters op and the maximum value max op ; S1.3.

2. Perform standardization processing on the operation parameters using min-max standardization:

3. The main and auxiliary power speed control system for ships based on a dual-channel neural network according to claim 1, wherein The step S2 further includes the following sub-steps: S2.

1. Define the input layers of the main and auxiliary power systems and initialize tensors: Active power system input: The shape is (N, T, 4), Auxiliary power system input: The shape is (N, T, 4), where N is the number of samples, T is the time step, and 4 is the number of features at each time step; S2.

2. Apply convolution operations to capture spatio-temporal features; Active power system convolution: Among them, is the output feature map at the t-th time step, B pp is the bias term, is the input data of the active power system at the t-th time step, ReLU() is the activation function, W pp is the weight matrix of the convolutional layer of the active power system, Auxiliary power system convolution: Among them, is the output feature map at the t-th time step, and B ap is the bias term, is the input data of the auxiliary power system at the t-th time step, ReLU() is the activation function, and W ap is the weight matrix of the convolutional layer of the main power system; S2.

3. Use pooling layers to reduce the spatial dimension of the output feature maps: Main power system pooling: Among them, is the feature map after pooling at the t-th time step, z pp (t·s:t·s+k) is the feature map within the sliding window range, s is the stride, and k is the pooling window size. Auxiliary power system pooling: Among them, is the feature map after pooling at the t-th time step, z ap (t·s:t·s+k) is the feature map within the sliding window range, s is the stride, and k is the pooling window size; S2.

4. Apply the ReLU activation function to introduce non-linear transformation.

4. The ship main and auxiliary power speed control system based on a dual-channel neural network according to claim 3, characterized in that, The step S3 further includes the following sub-steps: S3.

1. Define the query vector Q, and transform the and feature maps into the key vector K and the value vector V respectively; Among them, W k and W V are weight matrices for linear transformation; S3.

2. Use dot product similarity to calculate the attention scores; Active power system score z = Q T K z Auxiliary power system score f = Q T K f ; S3.

3. Obtain the attention weights through the Softmax function; α f = Softmax(score f ) α z = Softmax(score z ); S3.

4. Reload according to the obtained attention weights and obtain the weighted representation of the feature map: z f = α f V f z z ＝α zVz ； S3.

5. Concatenate the weighted feature maps z f and z z along the last dimension to form the comprehensive feature representation CHAR: CHAR = Concat(z f , z z , dim=-1) where dim=-1 indicates concatenation on the last dimension, and the shape is (N, T, 8).

5. The main and auxiliary power speed control system for ships based on a dual-channel neural network according to claim 4, characterized in that, The step S4 further includes the following sub-steps: S4.

1. Represent the i-th matrix in CHAR as CHAR i , then the elements of CHAR i can be represented as CHAR ijk : where i is the sample index, i∈[1,N], j is the time step index, j∈[1,T], and k is the feature index, k∈[1,8]; S4.

2. List all elements in CHAR in row-major order i as follows: CHAR flat,i = [CHAR i11 , CHAR i12 , ···, CHAR i18 , CHAR i21 , CHAR i22 , ···,CHAR i28 , ···, CHAR iT1 ,CHAR iT2 ,···,CHAR iT8 ; S4.

3. For the comprehensive feature CHAR, the flattened vector CHAR flat The expression is: CHAR flat = CHAR flat,1 ; CHAR flat,2 ; ···; CHAR flat,N 。 6. The main and auxiliary power speed control system for ships based on a dual-channel neural network according to claim 5, characterized in that The step S5 includes the following sub-steps: S5.

1. Input CHAR flat into the first fully connected layer, and map the feature vector of this layer to a hidden space H; H = ReLU(W fc1 ·CHAR flat + b fc1 ) Among them, W fc1 is the weight matrix of the hidden layer, and b fc1 is the bias term of the hidden layer; S5.

2. Input the hidden space H into the second connection layer, which maps the feature vector to the control signal space to obtain the control signal y; y = σ(W fc2 ·H + b fc2 ) where, W fc2 is the output layer weight matrix, b fc2 is the output layer bias term, and σ(·) is the activation function, C PPS is the number of control signals of the main power system, C APS is the number of control signals of the auxiliary power system, d h is the dimension of the hidden layer, y consists of two parts: the control signal y of the main power system z and the control signal y of the auxiliary power system f : y z = σ(W fc2,PPS ·H + b fc2,PPS ) y f = σ(W fc2,APS ·H + b fc2,APS )。 Among them, 7. The main and auxiliary power speed control system for ships based on a dual-channel neural network according to claim 6, characterized in that, The step S6 includes the following sub-steps: S6.

1. Map the output values of the control signal y of the main power system z and the control signal y of the auxiliary power system f back to the actual physical values: S6.

2. Design a smooth transition mechanism based on the physical values of the output values of the main and auxiliary power systems. S6.2.

1. Set the total switching time C time ; S6.2.

2. Mark the control signal of the main power system during the current ship operation as y z _old and the control signal y f _old; S6.2.

3. Use the linear interpolation formula to achieve smooth transition, and the formula is as follows: Among them, y i,old is the i-th control signal of the current main and auxiliary power systems of the ship, and y i,target is the i-th control signal of the control system, and C is the current time step.

Images