A method and device for optimizing the control of time delay in wireless communication network of ultra-high voltage system

Abstract

The invention discloses a method and device for optimizing the control of time delay of a wireless communication network of an ultra-high voltage system. The method comprises: collecting time delay data of the ultra-high voltage system; normalizing the time delay data, and constructing a training set and a test set; constructing an LSTM neural network prediction model based on an attention mechanism; training the LSTM neural network prediction model; selecting a control increment at the previous moment according to a delay prediction value output by the LSTM neural network prediction model at moment k, deriving a control input at the current moment according to the control increment and the control input at the previous moment, and using the control input at the current moment as an input of the ultra-high voltage system; the invention has the advantages that the time delay prediction value has high accuracy and strong robustness, can output optimal information, and meets the actual requirements of the integrated operation and maintenance of ultra-high voltage network communications for safety and intelligence.

Description

Technical Field

[0001] The present invention relates to the field of communication transmission, and more specifically to a method and device for optimizing control of time delay in a wireless communication network of an ultra-high voltage system. Background Art

[0002] With the completion and operation of UHV transmission channels, the scale of grid interconnection continues to increase, the electrical connections within the grid are closer, the safety and stability issues of UHV grids are becoming more and more prominent, and the technical difficulty and safety risks of operation and management have increased significantly. After the completion of the UHV synchronous grid, it will play the role of the skeleton of the national unified power grid. Its safe and reliable operation plays an important role in the safe and stable operation of the national power grid. The UHV system operation and maintenance business has high requirements for real-time transmission. Real-time transmission is the first line of defense for the safe and stable operation of the UHV system, and can most directly and effectively ensure the transient stability of the power system. Therefore, how to optimize the control of wireless network communication delay to further improve the communication quality of heterogeneous networks in the UHV system is the long-term work focus and work goal of the majority of UHV system practitioners.

[0003] Chinese patent publication number CN110556841A discloses a method for designing an isolated microgrid frequency controller taking into account wireless communication delay, including the following steps: establishing a cascaded time-varying delay model taking into account changes in network communication load based on M / M / 1 queuing theory; establishing a continuous-time frequency control model for a microgrid distributed generation system based on the continuous-time characteristics of the power system; using piecewise integration to process the continuous-time frequency control model, and establishing a discrete-time distributed generation system frequency control model taking into account the cascaded time-varying delay model; based on the discrete-time distributed generation system frequency control model, for the time-varying delay caused by changes in network load, combined with the discrete random control principle in control theory, a linear quadratic Gaussian control method is used to design a controller to reduce the impact of communication delay in wireless networks on the frequency stability of microgrids. This patent application reduces the impact of time-varying communication delay in networked control on the frequency stability of microgrids. However, this patent application does not involve a detailed delay optimization control process, the delay prediction value is not accurate, the robustness is not strong, it is difficult to output optimal information, and it cannot meet the actual needs of safety and intelligence for the integrated operation and maintenance of ultra-high voltage network communications. Summary of the invention

[0004] The technical problem to be solved by the present invention is that the prior art power grid wireless communication delay optimization control method has low accuracy of delay prediction value and weak robustness, making it difficult to output optimal information and unable to meet the actual requirements of safety and intelligence for integrated operation and maintenance of UHV network communications.

[0005] The present invention solves the above technical problems through the following technical means: a method for optimizing the delay control of a wireless communication network of a UHV system, the method comprising:

[0006] Step 1: Collect UHV system delay data;

[0007] Step 2: Normalize the delay data and construct training and test sets;

[0008] Step 3: Construct an LSTM neural network prediction model based on the attention mechanism;

[0009] Step 4: Train the LSTM neural network prediction model;

[0010] Step 5: At time k, select the control increment of the previous moment according to the delay prediction value output by the LSTM neural network prediction model, obtain the control input of the current moment according to the control increment and the control input of the previous moment, and use the control input of the current moment as the input of the UHV system.

[0011] The present invention is based on LSTM and attention mechanism to extract important timing information, and can predict time delay more accurately, so that the wireless communication network of the UHV system can operate more stably. At time k, the control increment of the previous moment is selected according to the delay prediction value output by the LSTM neural network prediction model, and the control input of the current moment is obtained according to the control increment and the control input of the previous moment, so as to realize active compensation of time delay and output optimal information, thus meeting the actual requirements of safety and intelligence for the integrated operation and maintenance of UHV network communication.

[0012] Furthermore, the step 1 comprises:

[0013] Wireless communication equipment is used to collect and save the UHV system delay data, and the collected first n+1 time delay sequence data are exported as the original experimental data X n =[x1,x2,...,x n+1 ].

[0014] Furthermore, the step 2 includes:

[0015] By formula The original experimental data is normalized and mapped to [0, 1]. The data at the first n moments are selected as the training set, and the data at the last moment k is selected as the test set.

[0016] Furthermore, the step three includes:

[0017] Integrate the attention mechanism into the LSTM network, through formula e t =tanh(W·h t-1 +b·h t) Calculate the hidden layer state h at time t-1 t-1 and the hidden layer state h at time t t The correlation index between

[0018] According to the relevant index through the formula Calculate the hidden layer state h at time t t The attention weight of

[0019] By formula Calculate the output features of the LSTM network at time t;

[0020] Constructing feature mapping function based on SCNs one-dimensional random configuration network Optimize the output features of the LSTM prediction network, where f0=0; L max is the maximum number of nodes in the network layer; y=[y1,y2,...,y i ] T is the feature matrix output in the attention-based LSTM prediction network structure, i is the dimension of the feature matrix; w j ,b j are the input weight and bias of the jth node in the network layer; g j (·) is the sigmoid activation function of the jth node in the network layer; β j =[β j,1 ,β j,2 ,...,β j,m ] T is the output weight matrix of the jth node in the network layer, and m is the dimension of the output weight matrix.

[0021] Furthermore, the step 4 includes:

[0022] The difference between the model output and the true value is defined as e L-1 =ff L-1 =[e L-1,1 ,...,e L-1,m ],

[0023] Update the parameters of the LSTM network and calculate the difference between the model output value and the true value. When the difference does not meet the set value, randomly generate the parameter g L ,b g , and calculate the output weight β by solving the global least squares method L , add a network layer node and update the model output to f L =f L-1 +β L g L , so that

[0024] Furthermore, the updating of the parameters of the LSTM network includes:

[0025] Assume that the LSTM network weight parameter W and the bias term b follow a standard normal distribution with a mean of 0 and a variance of 1, the output data initial parameter h0 = 0, and the state data initial parameter c0 = 0.

[0026] Input the training set into the LSTM network, and save the weight parameter W={W f , W i , W c , W o} and the bias term b={b f , b i , b c , b o}, where W f , W i , W c , W o They are the weight parameters of the input gate, forget gate, output gate and unit state of the LSTM network, b f , b i , b c , b o They are the bias parameters of the input gate, forget gate, output gate and unit state of the LSTM network respectively.

[0027] Furthermore, the step five includes:

[0028] According to the output feature y of the LSTM network at time k k , select the control increment ΔU(k)=(b T b+1) -1 b T (y k -y k-1 ), the control increment M generated at time k is ΔU M =[ΔU(k),ΔU(k+1),...,ΔU(k+M-1)];

[0029] According to the control increment and the control input at the previous moment, the control input at the current moment is obtained as u(k)=u(k-1)+ΔU(n), where: T is the sampling period;

[0030] The control input u(k) at the current moment is used as the input of the UHV system.

[0031] The present invention also provides a time delay optimization control device for a wireless communication network of a UHV system, the device comprising:

[0032] Data acquisition module, used to collect UHV system delay data;

[0033] Normalization module, used to normalize the delay data and construct training and test sets;

[0034] Model building module, used to build LSTM neural network prediction model based on attention mechanism;

[0035] Model training module, used to train LSTM neural network prediction model;

[0036] The model application module is used to select the control increment at the previous moment according to the delay prediction value output by the LSTM neural network prediction model at time k, derive the control input at the current moment according to the control increment and the control input at the previous moment, and use the control input at the current moment as the input of the UHV system.

[0037] Furthermore, the data acquisition module is used for:

[0038] Wireless communication equipment is used to collect and save the UHV system delay data, and the collected first n+1 time delay sequence data are exported as the original experimental data X n =[x1,x2,...,x n+1 ].

[0039] Furthermore, the normalization module is used to:

[0040] By formula The original experimental data is normalized and mapped to [0, 1]. The data at the first n moments are selected as the training set, and the data at the last moment k is selected as the test set.

[0041] Furthermore, the model building module is used to:

[0042] Integrate the attention mechanism into the LSTM network, through formula e t =tanh(W·h t-1 +b·h t ) Calculate the hidden layer state h at time t-1 t-1 and the hidden layer state h at time t t The correlation index between

[0043] According to the relevant index through the formula Calculate the hidden layer state h at time t t The attention weight of

[0044] By formula Calculate the output features of the LSTM network at time t;

[0045] Constructing feature mapping function based on SCNs one-dimensional random configuration network Optimize the output features of the LSTM prediction network, where f0=0; L max is the maximum number of nodes in the network layer; y=[y1,y2,...,y i ] T is the feature matrix output in the attention-based LSTM prediction network structure, i is the dimension of the feature matrix; w j ,b j are the input weight and bias of the jth node in the network layer; g j (·) is the sigmoid activation function of the jth node in the network layer; β j =[β j,1 ,β j,2 ,...,β j,m ] T is the output weight matrix of the jth node in the network layer, and m is the dimension of the output weight matrix.

[0046] Furthermore, the model training module is used to:

[0047] The difference between the model output and the true value is defined as e L-1 =ff L-1 =[e L-1,1 ,...,e L-1,m ],

[0048] Update the parameters of the LSTM network and calculate the difference between the model output value and the true value. When the difference does not meet the set value, randomly generate the parameter g L ,b g , and calculate the output weight β by solving the global least squares method L , add a network layer node and update the model output to f L =f L-1 +β L g L , so that

[0049] Furthermore, the updating of the parameters of the LSTM network includes:

[0050] Assume that the LSTM network weight parameter W and the bias term b follow a standard normal distribution with a mean of 0 and a variance of 1, the output data initial parameter h0 = 0, and the state data initial parameter c0 = 0.

[0051] Input the training set into the LSTM network, and save the weight parameter W={W f , W i , W c , W o} and the bias term b={b f , b i , bc , b o}, where W f , W i , W c , W o They are the weight parameters of the input gate, forget gate, output gate and unit state of the LSTM network, b f , b i , b c , b o They are the bias parameters of the input gate, forget gate, output gate and unit state of the LSTM network respectively.

[0052] Furthermore, the model application module is used to:

[0053] According to the output feature y of the LSTM network at time k k , select the control increment ΔU(k)=(b T b+1) -1 b T (y k -y k-1 ), the control increment M generated at time k is ΔU M =[ΔU(k),ΔU(k+1),...,ΔU(k+M-1)];

[0054] According to the control increment and the control input at the previous moment, the control input at the current moment is obtained as u(k)=u(k-1)+ΔU(n), where: T is the sampling period;

[0055] The control input u(k) at the current moment is used as the input of the UHV system.

[0056] The advantages of the present invention are: based on LSTM and attention mechanism to extract important timing information, the present invention can predict the delay more accurately, so that the wireless communication network of the ultra-high voltage system can operate more stably, and the control increment of the previous moment is selected at time k according to the delay prediction value output by the LSTM neural network prediction model, and the control input of the current moment is obtained according to the control increment and the control input of the previous moment, so as to realize active compensation of delay and output the optimal information, so as to meet the actual requirements of the integrated operation and maintenance of ultra-high voltage network communication for safety and intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The present invention discloses a flow chart of a method for optimizing the delay control of a wireless communication network in a UHV system. DETAILED DESCRIPTION

[0058] 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 combination with 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.

[0059] Example 1

[0060] like Figure 1 As shown, a method for optimizing the delay control of a wireless communication network of an ultra-high voltage system, the method comprising:

[0061] S1: Collecting UHV system delay data; the specific process is as follows:

[0062] Wireless communication equipment is used to collect and save the UHV system delay data, and the collected first n+1 time delay sequence data are exported as the original experimental data X n =[x1,x2,...,x n+1 ].

[0063] S2: Normalize the delay data and construct the training set and test set. The specific process is as follows:

[0064] By formula The original experimental data is normalized and mapped to [0, 1]. The data at the first n moments are selected as the training set, and the data at the last moment k is selected as the test set.

[0065] S3: Construct an LSTM neural network prediction model based on the attention mechanism; the specific process is:

[0066] Assume that the current training LSTM neural network model has an input layer, a hidden layer, an attention layer, and an output layer. Assume that the input data is x t , the output data is h t , the status data is C t , the output data of the previous training is h t-1 , the state data of the previous training is C t-1 , the expression of the LSTM neural network is as follows:

[0067]

[0068] Among them, t represents the current time, t-1 represents the previous time, Represents candidate memory state data, the expression is:

[0069]

[0070] f t Represents the data that needs to be forgotten, and the expression is:

[0071] f t =σ(w f ·[h t-1 ,x t ]+b f )

[0072] i t Indicates the data that needs to be remembered, the expression is:

[0073] i t =σ(w i ·[h t-1 ,x t ]+b i )

[0074] Among them, W f , W i , W c , W o They are the weight parameters of the input gate, forget gate, output gate and unit state, b f , b i , b c , b o are the bias parameters of the input gate, forget gate, output gate and unit state, respectively, h t-1 is the output of the previous gate, expressed as:

[0075] h t =o t tanh(C t )

[0076] o t =σ(W o ·[h t-1 ,x t ]+b c )

[0077] Where σ is the standard sigmoid function, defined as:

[0078]

[0079] The attention mechanism is integrated into the LSTM prediction network. The attention mechanism is expressed by t =tanh(W·h t-1 +b·h t ) Calculate the hidden layer state h at time t-1 t-1 and the hidden layer state h at time t t The correlation index between t And according to the relevant index e t Utilization Calculate h t The attention weight a t , according to the formula Calculate the output feature y of the LSTM network t ;

[0080] In order to improve the generalization ability of the model, a feature mapping function is constructed based on the SCNs one-dimensional random configuration network to optimize the information output by the LSTM prediction network to ensure classification accuracy. Specifically, a feature mapping function is constructed based on the SCNs one-dimensional random configuration network Optimize the output features of the LSTM prediction network, where f0=0; L max is the maximum number of nodes in the network layer; y=[y1,y2,...,y i ] T is the feature matrix output in the attention-based LSTM prediction network structure, i is the dimension of the feature matrix; w j ,b j are the input weight and bias of the jth node in the network layer; g j (·) is the sigmoid activation function of the jth node in the network layer; β j =[β j,1 ,β j,2 ,...,β j,m ] T is the output weight matrix of the jth node in the network layer, and m is the dimension of the output weight matrix.

[0081] S4: Train the LSTM neural network prediction model; the specific process is:

[0082] The difference between the model output and the true value is defined as e L-1 =ff L-1 =[e L-1,1 ,...,e L-1,m ],

[0083] Update the parameters of the LSTM network and calculate the difference between the model output value and the true value. When the difference does not meet the set value, randomly generate the parameter g L ,b g , and calculate the output weight β by solving the global least squares method L , add a network layer node and update the model output to f L =f L-1 +β L g L , so that

[0084] Among them, the parameters of the LSTM network are updated, including:

[0085] Assume that the LSTM network weight parameter W and the bias term b follow a standard normal distribution with a mean of 0 and a variance of 1, the output data initial parameter h0 = 0, and the state data initial parameter c0 = 0.

[0086] Input the training set into the LSTM network, and save the weight parameter W={W f , W i , W c , W o} and the bias term b={b f , b i , b c , b o}, where W f , W i , W c , W o They are the weight parameters of the input gate, forget gate, output gate and unit state of the LSTM network, b f , b i , b c , b o They are the bias parameters of the input gate, forget gate, output gate and unit state of the LSTM network respectively.

[0087] S5: At time k, the control increment of the previous moment is selected according to the delay prediction value output by the LSTM neural network prediction model, and the control input of the current moment is obtained according to the control increment and the control input of the previous moment, and the control input of the current moment is used as the input of the UHV system; the specific process is as follows:

[0088] According to the output feature y of the LSTM network at time k k , select the control increment ΔU(k)=(b T b+1) -1 b T (y k -y k-1 ), the control increment M generated at time k is ΔU M =[ΔU(k),ΔU(k+1),...,ΔU(k+M-1)];

[0089] According to the control increment and the control input at the previous moment, the control input at the current moment is obtained as u(k)=u(k-1)+ΔU(n), where: n≤m, T is the sampling period;

[0090] The control input u(k) at the current moment is used as the input of the UHV system.

[0091] Through the above technical scheme, the present invention is based on LSTM and attention mechanism to extract important timing information, which can predict the delay more accurately, so that the wireless communication network of the UHV system can operate more stably. At time k, the control increment of the previous moment is selected according to the delay prediction value output by the LSTM neural network prediction model, and the control input of the current moment is obtained according to the control increment and the control input of the previous moment, so as to realize active compensation of delay and output the optimal information, so as to meet the actual requirements of safety and intelligence for the integrated operation and maintenance of UHV network communication.

[0092] Example 2

[0093] The present invention also provides a time delay optimization control device for a wireless communication network of a UHV system, the device comprising:

[0094] Data acquisition module, used to collect UHV system delay data;

[0095] Normalization module, used to normalize the delay data and construct training and test sets;

[0096] Model building module, used to build LSTM neural network prediction model based on attention mechanism;

[0097] Model training module, used to train LSTM neural network prediction model;

[0098] The model application module is used to select the control increment at the previous moment according to the delay prediction value output by the LSTM neural network prediction model at time k, derive the control input at the current moment according to the control increment and the control input at the previous moment, and use the control input at the current moment as the input of the UHV system.

[0099] Specifically, the data acquisition module is used for:

[0100] Wireless communication equipment is used to collect and save the UHV system delay data, and the collected first n+1 time delay sequence data are exported as the original experimental data X n =[x1,x2,...,x n+1 ].

[0101] More specifically, the normalization module is used to:

[0102] By formula The original experimental data is normalized and mapped to [0, 1]. The data at the first n moments are selected as the training set, and the data at the last moment k is selected as the test set.

[0103] More specifically, the model building module is used to:

[0104] Integrate the attention mechanism into the LSTM network, through formula et =tanh(W·h t-1 +b·h t ) Calculate the hidden layer state h at time t-1 t-1 and the hidden layer state h at time t t The correlation index between

[0105] According to the relevant index through the formula Calculate the hidden layer state h at time t t The attention weight of

[0106] By formula Calculate the output features of the LSTM network at time t;

[0107] Constructing feature mapping function based on SCNs one-dimensional random configuration network Optimize the output features of the LSTM prediction network, where f0=0; L max is the maximum number of nodes in the network layer; y=[y1,y2,...,y i ] T is the feature matrix output in the attention-based LSTM prediction network structure, i is the dimension of the feature matrix; w j ,b j are the input weight and bias of the jth node in the network layer; g j (·) is the sigmoid activation function of the jth node in the network layer; β j =[β j,1 ,β j,2 ,...,β j,m ] T is the output weight matrix of the jth node in the network layer, and m is the dimension of the output weight matrix.

[0108] More specifically, the model training module is used to:

[0109] The difference between the model output and the true value is defined as e L-1 =ff L-1 =[e L-1,1 ,...,e L-1,m ],

[0110] Update the parameters of the LSTM network and calculate the difference between the model output value and the true value. When the difference does not meet the set value, randomly generate the parameter g L ,b g , and calculate the output weight β by solving the global least squares method L , add a network layer node and update the model output to f L =f L-1 +β L g L , so that

[0111] More specifically, the updating of the parameters of the LSTM network includes:

[0112] Assume that the LSTM network weight parameter W and the bias term b follow a standard normal distribution with a mean of 0 and a variance of 1, the output data initial parameter h0 = 0, and the state data initial parameter c0 = 0.

[0113] Input the training set into the LSTM network, and save the weight parameter W={W f , W i , W c , W o} and the bias term b={b f , b i , b c , b o}, where W f , W i , W c , W o They are the weight parameters of the input gate, forget gate, output gate and unit state of the LSTM network, b f , b i , b c , b o They are the bias parameters of the input gate, forget gate, output gate and unit state of the LSTM network respectively.

[0114] More specifically, the model application module is used to:

[0115] According to the output feature y of the LSTM network at time k k , select the control increment ΔU(k)=(b T b+1) -1 b T (y k -y k-1 ), the control increment M generated at time k is ΔU M =[ΔU(k),ΔU(k+1),...,ΔU(k+M-1)];

[0116] According to the control increment and the control input at the previous moment, the control input at the current moment is obtained as u(k)=u(k-1)+ΔU(n), where: T is the sampling period;

[0117] The control input u(k) at the current moment is used as the input of the UHV system.

[0118] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the delay control of a wireless communication network in a UHV system, characterized in that: The method comprises: Step 1: Collect the UHV system delay data; use wireless communication equipment to collect and save the UHV system delay data, and export the collected first n+1 time delay sequence data as the original experimental data X n =[x1,x2,...,x n+1 ]; Step 2: Normalize the delay data and construct the training set and test set; Normalize the original experimental data, map the normalized data to [0, 1], select the data of the first n moments as the training set, and the data of the last moment k as the test set; Step 3: Construct an LSTM neural network prediction model based on the attention mechanism; integrate the attention mechanism into the LSTM network, and use formula e t =tanh(W·h t-1 +b·h t ) Calculate the hidden layer state h at time t-1 t-1 and the hidden layer state h at time t t The correlation index between According to the relevant index through the formula Calculate the hidden layer state h at time t t The attention weight of By formula Calculate the output features of the LSTM network at time t; Constructing feature mapping function based on SCNs one-dimensional random configuration network Optimize the output features of the LSTM prediction network, where f0=0; L max is the maximum number of nodes in the network layer; y=[y1,y2,...,y i ] T is the feature matrix output in the attention-based LSTM prediction network structure, i is the dimension of the feature matrix; w j ,b j are the input weight and bias of the jth node in the network layer; g j (·) is the sigmoid activation function of the jth node in the network layer; β j =[β j,1 ,β j,2 ,...,β j,m ] T is the output weight matrix of the jth node in the network layer, and m is the dimension of the output weight matrix; Step 4: Train the LSTM neural network prediction model; Step 5: At time k, select the control increment of the previous moment according to the delay prediction value output by the LSTM neural network prediction model, obtain the control input of the current moment according to the control increment and the control input of the previous moment, and use the control input of the current moment as the input of the UHV system.

2. The method for optimizing the delay control of a wireless communication network in a UHV system according to claim 1, characterized in that: The fourth step comprises: The difference between the model output and the true value is defined as e L-1 =ff L-1 =[e L-1,1 ,...,e L-1,m ], Update the parameters of the LSTM network and calculate the difference between the model output value and the true value. When the difference does not meet the set value, randomly generate the parameter g L ,b g , and calculate the output weight β by solving the global least squares method L , add a network layer node and update the model output to f L =f L-1 +β L g L , so that 3. The method for optimizing the delay control of a wireless communication network in a UHV system according to claim 2, characterized in that: The updating of the parameters of the LSTM network includes: Assume that the LSTM network weight parameter W and the bias term b follow a standard normal distribution with a mean of 0 and a variance of 1, the output data initial parameter h0 = 0, and the state data initial parameter c0 = 0. Input the training set into the LSTM network, and save the weight parameter W={W f , W i , W c , W o } and the bias term b={b f , b i , b c , b o }, where W f , W i , W c , W o They are the weight parameters of the input gate, forget gate, output gate and unit state of the LSTM network, b f , b i , b c , b o They are the bias parameters of the input gate, forget gate, output gate and unit state of the LSTM network respectively.

4. The method for optimizing the delay control of a wireless communication network in a UHV system according to claim 3, characterized in that: The step five comprises: According to the output feature y of the LSTM network at time k k , select the control increment ΔU(k)=(b T b+1) -1 b T (y k -y k-1 ), the control increment M generated at time k is ΔU M =[ΔU(k),ΔU(k+1),...,ΔU(k+M-1)]; According to the control increment and the control input at the previous moment, the control input at the current moment is obtained as u(k)=u(k-1)+ΔU(n), where: T is the sampling period; The control input u(k) at the current moment is used as the input of the UHV system.

5. A time delay optimization control device for a wireless communication network of an ultra-high voltage system, characterized in that: The device comprises: The data acquisition module is used to collect the UHV system delay data; wireless communication equipment is used to collect and save the UHV system delay data, and the collected first n+1 time delay sequence data are exported as the original experimental data X n =[x1,x2,...,x n+1 ]; The normalization module is used to normalize the delay data and construct the training set and test set. Normalize the original experimental data, map the normalized data to [0, 1], select the data of the first n moments as the training set, and the data of the last moment k as the test set; The model building module is used to build an LSTM neural network prediction model based on the attention mechanism; the attention mechanism is integrated into the LSTM network, and the formula e t =tanh(W·h t-1 +b·h t ) Calculate the hidden layer state h at time t-1 t-1 and the hidden layer state h at time t t The correlation index between According to the relevant index through the formula Calculate the hidden layer state h at time t t The attention weight of By formula Calculate the output features of the LSTM network at time t; Constructing feature mapping function based on SCNs one-dimensional random configuration network Optimize the output features of the LSTM prediction network, where f0=0; L max is the maximum number of nodes in the network layer; y=[y1,y2,...,y i ] T is the feature matrix output in the attention-based LSTM prediction network structure, i is the dimension of the feature matrix; w j ,b j are the input weight and bias of the jth node in the network layer; g j (·) is the sigmoid activation function of the jth node in the network layer; β j =[β j,1 ,β j,2 ,...,β j,m ] T is the output weight matrix of the jth node in the network layer, and m is the dimension of the output weight matrix; Model training module, used to train LSTM neural network prediction model; The model application module is used to select the control increment at the previous moment according to the delay prediction value output by the LSTM neural network prediction model at moment k, derive the control input at the current moment according to the control increment and the control input at the previous moment, and use the control input at the current moment as the input of the UHV system.

6. The delay optimization control device for a wireless communication network of a UHV system according to claim 5, characterized in that: The model training module is used to: The difference between the model output and the true value is defined as e L-1 =ff L-1 =[e L-1,1 ,...,e L-1,m ], Update the parameters of the LSTM network and calculate the difference between the model output value and the true value. When the difference does not meet the set value, randomly generate the parameter g L ,b g , and calculate the output weight β by solving the global least squares method L , add a network layer node and update the model output to f L =f L-1 +β L g L , so that 7. The delay optimization control device for a wireless communication network of a UHV system according to claim 6, characterized in that: The updating of the parameters of the LSTM network includes: Assume that the LSTM network weight parameter W and the bias term b follow a standard normal distribution with a mean of 0 and a variance of 1, the output data initial parameter h0 = 0, and the state data initial parameter c0 = 0. Input the training set into the LSTM network, and save the weight parameter W={W f , W i , W c , W o } and the bias term b={b f , b i , b c , b o }, where W f , W i , W c , W o They are the weight parameters of the input gate, forget gate, output gate and unit state of the LSTM network, b f , b i , b c , b o They are the bias parameters of the input gate, forget gate, output gate and unit state of the LSTM network respectively.

8. The delay optimization control device for a wireless communication network of a UHV system according to claim 7, characterized in that: The model application module is used to: According to the output feature y of the LSTM network at time k k , select the control increment ΔU(k)=(b T b+1) -1 b T (y k -y k-1 ), the control increment M generated at time k is ΔU M =[ΔU(k),ΔU(k+1),...,ΔU(k+M-1)]; According to the control increment and the control input at the previous moment, the control input at the current moment is obtained as u(k)=u(k-1)+ΔU(n), where: T is the sampling period; The control input u(k) at the current moment is used as the input of the UHV system.

Images