A Control Method, Device, Computer Equipment and Medium for a Static Synchronous Compensator Based on the BERT-Transformer Cooperative Architecture DDPG

Abstract

This paper proposes a static synchronous compensator (STATCOM) control method, apparatus, computer device, and medium based on the BERT-Transformer collaborative architecture (DDPG). This method uses a Transformer-based bidirectional encoder representation (BERT) network as the action network of a deep deterministic policy gradient (DDPG) to generate control policies. The Transformer network also serves as a critic network to evaluate the value of state-action pairs. Through its unique network structure design, this method effectively addresses the complex control issues faced by STATCOMs when connected to the power grid, enabling precise control of the STATCOM, optimizing the reactive power compensation of the power grid, and improving the grid's stability and power quality.

Description

Technical Field

[0001] The present invention belongs to the fields of new energy, large models, power system control, power electronics technology, and smart grids, relates to a reinforcement learning control method, and is suitable for controlling reactive power compensation of a static synchronous compensator in a power system. Background Art

[0002] The existing reinforcement learning control method has the problem of poor reactive power compensation control effect of the static synchronous compensator when the power grid changes dynamically.

[0003] In addition, the action network and critic network of existing reinforcement learning control methods suffer from underfitting problems when facing dynamic control problems. Summary of the Invention

[0004] Based on this, it is necessary to provide a static synchronous compensator control method, device, computer equipment, computer-readable storage medium and computer program product based on the BERT-Transformer collaborative architecture DDPG to address the above technical problems.

[0005] In a first aspect, the present application provides a static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG. The method comprises:

[0006] collecting voltage and current data and cumulative value data from the static synchronous compensator, and normalizing the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the cumulative value data include the cumulative value of the AC side positive-sequence voltage error, the cumulative value of the DC voltage error, the cumulative value of the d-axis current error, and the cumulative value of the q-axis current error;

[0007] Inputting the accumulated value data and the normalized voltage and current data as a state vector into a current action network to obtain an action vector, and applying the action vector output by the current action network to a static synchronous compensator, wherein the current action network is a BERT network, which is a bidirectional encoder representation based on a Transformer;

[0008] Input the state vector and action vector together into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network;

[0009] Deep deterministic policy gradient DDPG is used to update the weights and bias parameters of the current critic network and the current action network.

[0010] In a second aspect, the present application also provides a static synchronous compensator control device based on the BERT-Transformer collaborative architecture DDPG. The device includes:

[0011] a data preprocessing module, configured to collect voltage and current data and accumulated value data from the static synchronous compensator, and normalize the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the accumulated value data include the accumulated value of the AC side positive-sequence voltage error, the accumulated value of the DC voltage error, the accumulated value of the d-axis current error, and the accumulated value of the q-axis current error;

[0012] An action network output action module is used to input the accumulated value data and the normalized voltage and current data as a state vector into the current action network to obtain an action vector, and apply the action vector output by the current action network to the static synchronous compensator. The current action network is a BERT network, which is a bidirectional encoder representation based on Transformer;

[0013] A critic network evaluation module is used to input the state vector and the action vector into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network;

[0014] The network update and optimization module is used to update the weights and bias parameters of the current critic network and the current action network using deep deterministic policy gradient DDPG.

[0015] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are performed:

[0016] collecting voltage and current data and cumulative value data from the static synchronous compensator, and normalizing the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the cumulative value data include the cumulative value of the AC side positive-sequence voltage error, the cumulative value of the DC voltage error, the cumulative value of the d-axis current error, and the cumulative value of the q-axis current error;

[0017] Inputting the accumulated value data and the normalized voltage and current data as a state vector into a current action network to obtain an action vector, and applying the action vector output by the current action network to a static synchronous compensator, wherein the current action network is a BERT network, which is a bidirectional encoder representation based on a Transformer;

[0018] Input the state vector and action vector together into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network;

[0019] Deep deterministic policy gradient DDPG is used to update the weights and bias parameters of the current critic network and the current action network.

[0020] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0021] collecting voltage and current data and cumulative value data from the static synchronous compensator, and normalizing the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the cumulative value data include the cumulative value of the AC side positive-sequence voltage error, the cumulative value of the DC voltage error, the cumulative value of the d-axis current error, and the cumulative value of the q-axis current error;

[0022] Inputting the accumulated value data and the normalized voltage and current data as a state vector into a current action network to obtain an action vector, and applying the action vector output by the current action network to a static synchronous compensator, wherein the current action network is a BERT network, which is a bidirectional encoder representation based on a Transformer;

[0023] Input the state vector and action vector together into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network;

[0024] Deep deterministic policy gradient DDPG is used to update the weights and bias parameters of the current critic network and the current action network.

[0025] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the following steps:

[0026] collecting voltage and current data and cumulative value data from the static synchronous compensator, and normalizing the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the cumulative value data include the cumulative value of the AC side positive-sequence voltage error, the cumulative value of the DC voltage error, the cumulative value of the d-axis current error, and the cumulative value of the q-axis current error;

[0027] Inputting the accumulated value data and the normalized voltage and current data as a state vector into a current action network to obtain an action vector, and applying the action vector output by the current action network to a static synchronous compensator, wherein the current action network is a BERT network, which is a bidirectional encoder representation based on a Transformer;

[0028] Input the state vector and action vector together into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network;

[0029] Deep deterministic policy gradient DDPG is used to update the weights and bias parameters of the current critic network and the current action network.

[0030] The above-mentioned static synchronous compensator control method, device, computer equipment, storage medium and computer program product based on the BERT-Transformer collaborative architecture DDPG collects voltage and current data and cumulative value data from the static synchronous compensator, normalizes the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive sequence voltage, AC side positive sequence voltage error, DC voltage error, d-axis current error and q-axis current error, and the cumulative value data includes the cumulative value of the AC side positive sequence voltage error, the cumulative value of the DC voltage error, the cumulative value of the d-axis current error and the cumulative value of the q-axis current error; the cumulative value data and the normalized voltage and current data are input into the current action network as the state vector to obtain the action vector, and the current The action vector output by the previous action network is applied to the static synchronous compensator, and the current action network is a BERT network, which is a bidirectional encoder representation based on Transformer. The state vector and the action vector are input together into the current critic network to evaluate the value of the current state-action pair. The current critic network is a Transformer network. The deep deterministic policy gradient (DDPG) is used to update the weights and bias parameters of the current critic network and the current action network. The Transformer-based bidirectional encoder representation BERT network and the Transformer network work together to achieve precise control of the static synchronous compensator, have the function of optimizing the reactive power compensation effect of the power grid, can improve the stability and power quality of the power grid, and reduce the energy consumption of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of the method of the present invention.

[0032] Figure 2 This is a control framework diagram of the static synchronous compensator of the method of the present invention.

[0033] Figure 3 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0035] In one embodiment, Figure 1As shown, a flowchart of a static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0036] The present invention proposes a static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG. The deep deterministic policy gradient DDPG has an action network and a critic network. The action network includes a current action network and a target action network, and the critic network includes a current critic network and a target critic network. The Transformer-based bidirectional encoder representation BERT network is used as the action network of the deep deterministic policy gradient DDPG, and the Transformer network is used as the critic network of the deep deterministic policy gradient DDPG. The Transformer-based bidirectional encoder representation BERT network and the Transformer network work together to achieve precise control of the static synchronous compensator, have the function of optimizing the reactive power compensation effect of the power grid, can improve the stability and power quality of the power grid, and reduce the energy consumption of the power grid. The steps during use are:

[0037] Step (1): Data preprocessing;

[0038] collecting voltage and current data and cumulative value data from the static synchronous compensator, and normalizing the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the cumulative value data include the cumulative value of the AC side positive-sequence voltage error, the cumulative value of the DC voltage error, the cumulative value of the d-axis current error, and the cumulative value of the q-axis current error;

[0039] The voltage and current data collected from the static synchronous compensator include 、 、 、 、 、 、 、 、 、 、 and ; is the d-axis current, is the q-axis current, is the DC voltage, is the positive sequence voltage on the AC side, is the AC side positive sequence voltage error, is the DC voltage error, is the d-axis current error, is the q-axis current error, is the cumulative value of the AC side positive sequence voltage error, is the cumulative value of the DC voltage error, is the accumulated value of the d-axis current error, is the accumulated value of the q-axis current error;

[0040] The normalization operation normalizes the non-normalized input by rescaling it to the interval between 0 and 1 to obtain the normalized input; the non-normalized input is , that is, normalize the voltage and current data;

[0041] and 、 、 and No normalization is required, the cumulative value data includes 、 、 and ;

[0042] The process of normalizing non-normalized input is: .

[0043] in, The set of minimum values of each element of the unnormalized input, The set of maximum values of each element of the unnormalized input, is the normalized input, that is, the normalized 、 、 、 、 、 、 and , that is, the normalized voltage and current data;

[0044] Step (2): The action network outputs the action;

[0045] Inputting the accumulated value data and the normalized voltage and current data as a state vector into a current action network to obtain an action vector, and applying the action vector output by the current action network to a static synchronous compensator, wherein the current action network is a BERT network, which is a bidirectional encoder representation based on a Transformer;

[0046] 、 、 、 and normalized input After inputting into the deep deterministic policy gradient DDPG, the state vector is constructed , the state vector for: , the state vector Input to the current action network, the current action network is a Transformer-based bidirectional encoder representation BERT network; in the Transformer-based bidirectional encoder representation BERT network, the state vector After the neural network transformation, the state vector The information is abstracted and transformed layer by layer to extract feature representation, which provides a basis for subsequent action decision-making; after being processed by the Transformer-based bidirectional encoder representation BERT network, the action vector is output. , motion vector for: , is the q-axis current reference value, is the d-axis current reference value, is the d-axis voltage, is the q-axis voltage; that is, the previous action network outputs the q-axis current reference value, d-axis current reference value, d-axis voltage and q-axis voltage, that is, the Transformer-based bidirectional encoder represents that the BERT network outputs the q-axis current reference value, d-axis current reference value, d-axis voltage and q-axis voltage, that is, the action vector obtained after the accumulated value data and the normalized voltage and current data are input into the current action network is the q-axis current reference value, d-axis current reference value, d-axis voltage and q-axis voltage.

[0047] The Transformer-based bidirectional encoder representation BERT network includes:

[0048] An input layer configured to receive raw feature input and map it into a feature embedding vector, wherein the raw feature includes at least one of a numerical feature, a categorical feature, or a sequence feature;

[0049] a position encoding module configured to generate a position encoding vector that matches the dimension of the feature embedding vector and add the position encoding vector to the feature embedding vector to represent the position information of the feature;

[0050] A multi-layer bidirectional Transformer encoder, where each layer contains a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism allows each feature to interact bidirectionally with all other features and calculates attention weights using a query matrix, a key matrix, and a value matrix. The feedforward neural network performs a nonlinear transformation on the attention output.

[0051] A layer normalization and residual connection module is configured to perform layer normalization operations before and after the multi-head self-attention mechanism and the feedforward neural network, and introduce residual connections to enhance feature transfer;

[0052] The output layer is configured to aggregate the outputs of the multi-layer bidirectional Transformer encoder to generate processed feature outputs, wherein the aggregation method includes at least one of global average pooling, global maximum pooling, or weighted summation.

[0053] The action vector output by the current action network Applied to static synchronous compensator, control the operation of static synchronous compensator, static synchronous compensator adjusts reactive power output according to control action, thereby controlling the voltage level and power flow of the power grid; in the current action network output action vector After the power grid's operating status changes, the status information is updated. and the corresponding reward signal; the updated state information is collected again and the target state vector is obtained after data preprocessing in step (1) As the input of the current action network in the next control cycle; the reward signal is set according to the control target and the grid operation index, and is used to measure the contribution of the current control action to the realization of the control target. For example, in the implementation, the reward signal Can be designed as: or wait.

[0054] Step (3): Critic network evaluation;

[0055] Input the state vector and action vector together into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network;

[0056] The state vector and motion vector Input together into the current critic network, the current critic network is a Transformer network, which can process state vectors at the same time and motion vector The information is used to evaluate the value of the current state-action pair. In the Transformer network, the state-action pair is jointly analyzed and evaluated, and the Transformer network outputs a Value, that is, the value of the current state-action pair, the The value is reflected in the current state vector The following Transformer-based bidirectional encoder represents the action vector output by the BERT network The long-term expected return that can be obtained is the action vector output by the BERT network for the Transformer-based bidirectional encoder. The quality of the Transformer network is quantitatively evaluated; The value is used as a feedback signal to guide the update and optimization of the Transformer-based bidirectional encoder representation BERT network; The higher the value, the more the current motion vector The more conducive it is to achieving the control goal under the Transformer network evaluation;

[0057] Transformer networks include but are not limited to:

[0058] An input encoding module configured to receive raw feature data and generate a feature embedding vector, wherein the raw feature data includes at least one of numerical, categorical, or temporal features, and the input encoding module includes a feature embedding layer and a position encoding layer, wherein the position encoding layer generates a position encoding vector using a sine or cosine function and adds the position encoding vector to the feature embedding vector;

[0059] The encoder stack consists of multiple encoder layers with the same structure. Each encoder layer contains:

[0060] a multi-head self-attention sublayer configured to calculate association weights between input features, wherein the multi-head self-attention sublayer linearly transforms the input features to generate query, key, and value matrices, and calculates attention scores using a scaled dot product attention mechanism;

[0061] Multiple residual connections and layer normalization sublayers, configured to apply residual connections and layer normalization operations before and after the multi-head self-attention sublayer respectively;

[0062] The feedforward neural network sublayer consists of two linear transformation layers and a nonlinear activation function;

[0063] The decoder stack consists of multiple decoder layers with the same structure. Each decoder layer contains:

[0064] A masked multi-head self-attention sub-layer, configured to prevent attention to features at future positions;

[0065] The encoder-decoder attention sublayer is configured to associate the encoder output with the current state of the decoder;

[0066] Feedforward neural network sublayer;

[0067] An output decoding module is configured to convert the output of the decoder stack into a target feature representation, wherein the output decoding module includes a linear transformation layer and an optional activation function layer.

[0068] Step (4): Network update and optimization;

[0069] Deep deterministic policy gradient DDPG is used to update the weights and bias parameters of the current critic network and the current action network.

[0070] According to the output of the current critic network reward signals generated by changes in value and grid operating status , the weight and bias parameters of the current action network Optimize the weights and bias parameters of the current action network after deep deterministic policy gradient DDPG optimization for: ,in, is the learning rate, is the gradient operator, is the number of samples;

[0071] The goal of optimization is to make the optimized current action network output a better action vector than the current action network before optimization, so as to maximize the long-term cumulative reward and thus achieve precise control of the static synchronous compensator;

[0072] Reward signal generated based on changes in grid operating status and the target critic network output value , using the loss function Weight and bias parameters for the current critic network To update, the loss function Can be designed as: ,in, is the discount factor; the Transformer network output of the target critic network and the current critic network The value update process is .

[0073] Update the current critic network weights and bias parameters The process is: .

[0074] The updated current critic network estimates the state-action pair more accurately than the current critic network before updating. value;

[0075] The weights and bias parameters of the target action network and the target critic network are smoothed by the factor Gradually approach the weights and bias parameters of the current action network and the current critic network to achieve smooth synchronization of the target action network and the target critic network with the current action network and the current critic network. The synchronization process is: and ,in, are the weight and bias parameters of the target critic network, are the weight and bias parameters of the target action network;

[0076] By synchronizing the target action network and the target critic network, the control strategy is optimized to make the control of the static synchronous compensator stable and reliable. In the subsequent control cycle, the target action network and the target critic network provide accurate target values for the update of the current action network and the current critic network, promoting the control performance of the entire static synchronous compensator control system.

[0077] The present invention has the following advantages and effects compared to the prior art:

[0078] (1) The existing traditional control method for controlling the static synchronous compensator has the problems of poor control effect and insufficient precision of control action when facing dynamic problems of the power grid. The present invention proposes a static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG, which can solve the control problem of the static synchronous compensator when facing dynamic problems of the power grid and realize precise control of the static synchronous compensator.

[0079] (2) The existing traditional control method for controlling the static synchronous compensator has the problems of poor reactive power compensation effect, poor grid stability and low power quality. The present invention proposes a static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG, which can optimize the reactive power compensation effect of the grid when controlling the static synchronous compensator, and improve the stability and power quality of the grid.

[0080] (3) The existing reinforcement learning methods have the problem of underfitting of the action network and critic network when controlling the static synchronous compensator. The Transformer-based bidirectional encoder representation BERT network and Transformer network can solve the problem of underfitting of the reinforcement learning action network and critic network.

[0081] Figure 2 The input of the DDPG control method based on the BERT-Transformer collaborative architecture is the AC side positive sequence voltage error obtained from the static synchronous compensator. , DC voltage error , d-axis current error and q-axis current error , output q-axis current reference value , d-axis current reference value , d-axis voltage and q-axis voltage to the static synchronous compensator to realize the control of the static synchronous compensator.

[0082] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means can be implemented via Wi-Fi, a mobile cellular network, NFC (near-field communication), or other technologies. When executed by the processor, the computer program implements a static synchronous compensator control method based on the BERT-Transformer collaborative architecture (DDPG). The display unit of the computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0083] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0084] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0085] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0086] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0087] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0088] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0089] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0090] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG, characterized in that: collecting voltage and current data and cumulative value data from the static synchronous compensator, and normalizing the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the cumulative value data include the cumulative value of the AC side positive-sequence voltage error, the cumulative value of the DC voltage error, the cumulative value of the d-axis current error, and the cumulative value of the q-axis current error; Inputting the accumulated value data and the normalized voltage and current data as a state vector into a current action network to obtain an action vector, and applying the action vector output by the current action network to a static synchronous compensator, wherein the current action network is a BERT network, which is a bidirectional encoder representation based on a Transformer; Input the state vector and action vector together into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network; Deep deterministic policy gradient DDPG is used to update the weights and bias parameters of the current critic network and the current action network.

2. The static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG according to claim 1, characterized in that: The action vectors are a q-axis current reference value, a d-axis current reference value, a d-axis voltage, and a q-axis voltage.

3. The static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG according to claim 1, characterized in that: The BERT network includes an input layer, a position encoding module, a multi-layer bidirectional Transformer encoder, a layer normalization and residual connection module, and an output layer.

4. The static synchronous compensator control method based on the BERT-Transformer collaborative architecture DDPG according to claim 1, characterized in that: The Transformer network includes an input encoding module, an encoder stack, a multi-head self-attention sublayer, multiple residual connections and layer normalization sublayers, a feedforward neural network sublayer, a decoder stack, a masked multi-head self-attention sublayer, an encoder-decoder attention sublayer and an output decoding module.

5. A static synchronous compensator control device based on the BERT-Transformer collaborative architecture DDPG, characterized in that: The device comprises: a data preprocessing module, configured to collect voltage and current data and accumulated value data from the static synchronous compensator, and normalize the voltage and current data to obtain normalized voltage and current data, wherein the voltage and current data include d-axis current, q-axis current, DC voltage, AC side positive-sequence voltage, AC side positive-sequence voltage error, DC voltage error, d-axis current error, and q-axis current error; and the accumulated value data include the accumulated value of the AC side positive-sequence voltage error, the accumulated value of the DC voltage error, the accumulated value of the d-axis current error, and the accumulated value of the q-axis current error; An action network output action module is used to input the accumulated value data and the normalized voltage and current data as a state vector into the current action network to obtain an action vector, and apply the action vector output by the current action network to the static synchronous compensator. The current action network is a BERT network, which is a bidirectional encoder representation based on Transformer; A critic network evaluation module is used to input the state vector and the action vector into the current critic network to evaluate the value of the current state-action pair, wherein the current critic network is a Transformer network; The network update and optimization module is used to update the weights and bias parameters of the current critic network and the current action network using deep deterministic policy gradient DDPG.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a static synchronous compensator control method based on a BERT-Transformer collaborative architecture DDPG as described in any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a static synchronous compensator control method based on a BERT-Transformer collaborative architecture DDPG as described in any one of claims 1 to 4 are implemented.

Images