Asynchronous power grid panoramic intelligent self-driving control system and method based on deep learning

Abstract

The invention discloses an asynchronous power grid panoramic intelligent self-driving control system and method based on deep learning, and belongs to the technical field of power grid panoramic self-driving, and the system comprises a power grid data obtaining module which is used for obtaining a power grid topological structure diagram; the power grid data imaging module is used for calculating an electrical distance matrix through an impedance matrix on the basis of the power grid topological structure diagram and performing dimension reduction conversion to form an input tensor; the power grid power stability control calculation module takes the input tensor as the input of the CNN model to obtain an output matrix of power grid power stability control; and the self-driving control module performs power control on each node of the power grid according to the output matrix. Through the application of the deep learning algorithm, the method can autonomously learn the operation law and characteristics of the power grid, and achieves the intelligent control of the power grid. Compared with a traditional manual control mode and a control mode based on a preset rule, the method has higher flexibility and adaptability, and can better deal with the complexity and uncertainty of the power grid.

Description

Technical Field

[0001] The present invention relates to the field of power grid panoramic self-driving technology, and in particular to a deep learning-based asynchronous power grid panoramic intelligent self-driving control system and method. Background Art

[0002] In recent years, deep learning technology has made significant progress in fields such as image recognition and natural language processing. Its powerful feature extraction and pattern recognition capabilities offer new insights into power grid control. Furthermore, the introduction of asynchronous training methods allows multiple neural network training threads to run simultaneously, significantly improving training speed and efficiency. These technologies have opened the door to the construction of asynchronous, panoramic, intelligent, and autonomous power grid control systems based on deep learning.

[0003] With the rapid development of smart grid technology, the complexity, dynamics, and uncertainty of power grids are increasing, placing higher demands on their monitoring, dispatching, and control. Traditional power grid control systems often rely on manual experience and preset rules, making them difficult to cope with the complex and changing power grid environment. Therefore, developing intelligent control systems capable of autonomous learning and adaptive adjustment has become a hot topic of research.

[0004] Therefore, a panoramic intelligent self-driving control system and method for asynchronous power grid based on deep learning came into being. Summary of the Invention

[0005] To solve the above technical problems, a deep learning-based panoramic intelligent self-driving control system for asynchronous power grids is proposed, including a power grid data acquisition module for obtaining the power grid topology diagram; a power grid data visualization module, which calculates the electrical distance matrix through the impedance matrix based on the power grid topology diagram and performs dimensionality reduction transformation to form an input tensor; a power grid power stability control calculation module, which uses the input tensor as the input of the CNN model to obtain the output matrix of the power grid power stability control; and a self-driving control module, which controls the power of each node in the power grid according to the output matrix.

[0006] As a preferred solution of the asynchronous power grid panoramic intelligent self-driving control system based on deep learning described in the present invention, wherein: the calculation of the electrical distance matrix through the impedance matrix includes obtaining a node admittance matrix based on the power grid topology diagram, wherein the node admittance matrix represents the relationship between the currents of each node in the power grid topology diagram when voltage is used as the excitation; obtaining an inverse matrix based on the node admittance matrix, the inverse matrix is ​​an impedance matrix; obtaining a node connection impedance matrix based on the impedance matrix through the superposition principle; calculating the modulus of the elements in the connection impedance matrix, the modulus of the element corresponding to the element of the electrical distance matrix between the nodes.

[0007] As a preferred solution of the asynchronous power grid panoramic intelligent self-driving control system based on deep learning described in the present invention, the dimensionality reduction conversion includes converting the electrical distance matrix into a high-dimensional space, defining the row vector of the multidimensional vector X in the space to represent the spatial distribution of each node; defining the conditional probability, indicating that the node x corresponding to the jth row of the row vector of the multidimensional vector X j Appears at node x i The probability of being near the matrix; obtain the joint probability distribution of the high-dimensional space and the space where the matrix is ​​located; introduce the KL divergence to measure the deviation of the two spaces, and set the KL divergence as the objective function; calculate the gradient of the objective function for each two-dimensional node; then obtain the minimum objective function through the gradient descent algorithm, and finally obtain the coordinates of the two-dimensional node after dimensionality reduction.

[0008] As a preferred solution of the asynchronous power grid panoramic intelligent self-driving control system based on deep learning described in the present invention, the dimensionality reduction conversion also includes placing the two-dimensional coordinate Y in a sub-two-dimensional plane, forming a channel of the power grid operation characteristic information corresponding to each node and assigning it to the corresponding node coordinate, and superimposing the channels composed of various characteristic quantities into an input tensor INPUT.

[0009] As a preferred solution of the asynchronous power grid panoramic intelligent self-driving control system based on deep learning described in the present invention, wherein: the input tensor INPUT includes: the value range of the node coordinate Y is [0,1], the coordinate Y is denormalized and amplified into an integer, and the amplification multiple is determined according to the needs of CNN and the actual situation of the power system node; select an amplification integer z, and the node coordinate y with a value range of [0,1] is amplified into an integer. i By using the normalized inverse operation method to enlarge it to [0, h], and then rounding the enlarged node coordinates to get the integer form of the node coordinate Y INT , normalized by the normalization equation; the grid operation characteristic information is separately composed of a channel and assigned to each node in the grid, and the value is assigned to the corresponding node coordinate. The channels composed of various characteristic quantities are superimposed and input to form the input tensor INPUT, where the grid operation characteristic information includes bus active load and reactive load, bus voltage amplitude and phase angle, line active power and reactive power, bus phase angle difference at both ends of the line, total system active load, and total system reactive load.

[0010] As a preferred solution of the asynchronous power grid panoramic intelligent self-driving control system based on deep learning described in the present invention, the CNN model includes designing an input layer according to the preprocessed data format. If the data is a two-dimensional image, a two-dimensional input layer is used; if the data is a one-dimensional sequence, a one-dimensional convolutional layer is used.

[0011] Design multiple convolutional layers to extract spatial or temporal features from the input data through convolution operations. Activation functions are added after the convolutional layers to introduce nonlinearity. Pooling layers are used to reduce the dimension of the feature map and extract features. A fully connected layer is added after the convolutional and pooling layers to merge the feature information extracted by the convolutional layers and pass it to the next layer.

[0012] Based on the requirements of the control task, the output layer converts the output of the fully connected layer into grid control parameters. The output matrix contains control signals for regulating grid power stability. The output matrix contains multiple control parameters, which are used to adjust the output of the generator, adjust the load distribution, and control the reactive compensation equipment to maintain voltage stability, frequency stability, and power balance of the grid.

[0013] As a preferred solution of the deep learning-based asynchronous power grid panoramic intelligent self-driving control system described in the present invention, the power control of each node of the power grid according to the output matrix includes: selecting a control strategy based on the output matrix, including linear control, nonlinear control, optimal control or intelligent control; designing a controller according to the control strategy, and the controller should be able to receive real-time data from the power grid and calculate the corresponding control input.

[0014] Another object of the present invention is to provide a panoramic intelligent self-driving control method for asynchronous power grids based on deep learning. The present invention solves the technical problem that traditional power grid control systems often rely on manual experience and preset rules and are difficult to cope with complex and changeable power grid environments.

[0015] As a preferred solution of the deep learning-based panoramic intelligent self-driving control method for an asynchronous power grid described in the present invention, it is characterized by including: obtaining a power grid topology diagram; based on the power grid topology diagram, calculating an electrical distance matrix through an impedance matrix, and performing dimensionality reduction conversion to form an input tensor; using the input tensor as input to a CNN model to obtain an output matrix for power grid power stability control; and performing power control on each node of the power grid according to the output matrix.

[0016] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the asynchronous power grid panoramic intelligent self-driving control method based on deep learning are implemented.

[0017] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for controlling a panoramic intelligent self-driving system of an asynchronous power grid based on deep learning.

[0018] The present invention has the following beneficial effects: It improves the level of intelligent power grid control. By applying deep learning algorithms, this method can autonomously learn the operating patterns and characteristics of the power grid, thus achieving intelligent control of the grid. Compared with traditional manual control and control methods based on preset rules, this method has greater flexibility and adaptability, and can better cope with the complexity and uncertainty of the power grid.

[0019] This method achieves comprehensive grid awareness. This approach integrates multiple sensors and monitoring devices to monitor grid parameters such as voltage, current, and power in real time, as well as external environmental factors such as weather and load. This comprehensive awareness enables the control system to fully understand the grid's operating status, providing comprehensive data support for subsequent decision-making and control.

[0020] Improve grid stability and operational efficiency. Through continuous learning and optimization, deep learning-based control models can accurately predict dynamic grid changes and proactively implement appropriate control measures. This helps reduce grid fluctuations and failures, improving grid stability. Furthermore, by optimizing grid dispatch and control strategies, this approach can improve grid operational efficiency and reduce energy consumption and costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 This is an overall structural diagram of a deep learning-based asynchronous power grid panoramic intelligent self-driving control system provided by one embodiment of the present invention;

[0023] Figure 2 This is a workflow diagram of a deep learning-based asynchronous power grid panoramic intelligent self-driving control method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0025] Example 1, with reference to Figure 1, which is the first embodiment of the present invention, provides an asynchronous power grid panoramic intelligent self-driving control system based on deep learning, including:

[0026] A power grid data acquisition module is used to obtain a power grid topology diagram;

[0027] The power grid data visualization module calculates the electrical distance matrix based on the power grid topology diagram through the impedance matrix and performs dimensionality reduction transformation to form the input tensor;

[0028] The grid power stability control calculation module takes the input tensor as the input of the CNN model and obtains the output matrix of the grid power stability control;

[0029] The self-driving control module controls the power of each node in the power grid according to the output matrix.

[0030] It should be noted that the system in this embodiment, through the grid data acquisition module, can fully capture the grid topology and connect key elements such as generators, buses, and loads as nodes. Each node contains rich information about the grid's operating characteristics. This comprehensive data integration approach enables the system to more accurately understand the current state of the grid, providing a solid foundation for subsequent intelligent control.

[0031] Advanced power grid data visualization technology: The power grid data visualization module utilizes an impedance matrix and an electrical distance matrix, employing a nonlinear dimensionality reduction algorithm to transform complex power grid topologies into intuitive two-dimensional coordinate representations. This visualization approach not only simplifies data complexity but also makes the power grid topology and the electrical relationships between nodes clearer and easier to understand. Furthermore, grid operational characteristics are assigned as channels to corresponding node coordinates, forming a multi-dimensional input tensor that provides rich input information for subsequent deep learning models.

[0032] Deep Learning-Driven Grid Power Stability Control: The grid power stability control calculation module uses a CNN model to process input tensors and derive the output matrix of the grid power stability control method. Deep learning models possess powerful feature extraction and pattern recognition capabilities, automatically learning complex patterns and regularities in grid operation data. This allows for more accurate predictions of the grid's power stability state and the development of appropriate control strategies. This deep learning-based control method offers greater accuracy and robustness than traditional control methods.

[0033] Intelligent self-driving control enables autonomous grid optimization: The self-driving control module controls the power output of each grid node based on the output matrix, enabling intelligent self-driving of the grid. This control method automatically adjusts the power output of each node based on the real-time status of the grid, ensuring stable grid operation. Furthermore, intelligent self-driving control can automatically adjust control strategies based on load changes and fault conditions, enabling autonomous grid optimization and self-healing.

[0034] Improving the safety and economic efficiency of grid operations: Through comprehensive data integration, advanced image processing, deep learning-driven control strategies, and intelligent self-driving control, the system significantly improves the safety and economic efficiency of grid operations. On the one hand, the system monitors the grid's operating status in real time, promptly identifying and addressing potential faults and hidden dangers to ensure stable grid operation. On the other hand, the system automatically adjusts power output and control strategies based on grid load changes and energy distribution, achieving efficient energy utilization and minimizing costs.

[0035] Embodiment 2, which is the second embodiment of the present invention, provides an asynchronous power grid panoramic intelligent self-driving control system based on deep learning, including:

[0036] The power grid data acquisition module is used to obtain the power grid topology diagram.

[0037] In this embodiment, the generator, busbar, and load are interconnected as nodes to form a power grid topology diagram, and each node contains power grid operation characteristic information.

[0038] The power grid data visualization module calculates the electrical distance matrix through the impedance matrix based on the power grid topology diagram, and performs dimensionality reduction transformation to form the input tensor.

[0039] It should be noted that the power grid data visualization module is used to obtain the impedance matrix based on the power grid topology diagram, and calculate the electrical distance matrix D between the nodes of the power grid topology diagram based on the impedance matrix. ij , the matrix D is reduced by nonlinear dimensionality reduction algorithm ij Convert it into the two-dimensional coordinate Y corresponding to the node, and place the two-dimensional coordinate Y in the sub-two-dimensional plane. The grid operation characteristic information corresponding to each node is composed of a channel and assigned to the corresponding node coordinate. The channels composed of various characteristic quantities are superimposed and input to form the input tensor INPUT.

[0040] In a preferred embodiment of the present invention, an impedance matrix is ​​obtained based on the power grid topology diagram, and an electrical distance matrix D between nodes in the power grid topology diagram is calculated based on the impedance matrix. ij The specific process includes the following:

[0041] Obtaining a node admittance matrix based on a power grid topology diagram, wherein the node admittance matrix represents the relationship between currents of each node in the power grid topology diagram when voltage is used as an excitation;

[0042] Calculate the inverse matrix based on the node admittance matrix, which is the impedance matrix Z;

[0043] Based on the impedance matrix, the node connection impedance matrix is ​​obtained by superposition principle;

[0044] ZL ij =(Z ii -Z ij )-(Z ii -Z jj )

[0045] Among them, ZL ij is the connection impedance matrix, Z ij is the impedance value of the i-th row and j-th column in the impedance matrix Z, Z ii is the impedance value of the i-th row and i-th column in the impedance matrix Z, Z jj is the impedance value of the jth row and jth column in the impedance matrix Z;

[0046] Calculate the contact impedance matrix ZL ij The modulus of the element in the corresponding electrical distance matrix D between nodes ij Elements of the matrix D ij is an n-order square matrix, where n is the number of nodes in the power grid topology diagram. The elements in the n-th column of the i-th row of the matrix represent the electrical distances between the node i corresponding to the row and the other n-1 nodes, and the elements on the main diagonal are 0.

[0047] In a preferred embodiment of the present invention, the dimensionality reduction transformation is to transform the matrix D into ij The conversion to the two-dimensional coordinate Y corresponding to the node specifically includes the following process:

[0048] The matrix D ij Convert to a high-dimensional space, where the row vectors of the multidimensional vector X represent the spatial distribution of each node;

[0049] Define the conditional probability P j|i , which represents the node x corresponding to the jth row of the multidimensional vector X j Appears at node x i The probability of being nearby, P j|i The formula is as follows:

[0050]

[0051] Among them, ||x i -x j ||2 Represents node x j With node x i The straight-line distance between two vectors in Euclidean space, Represents other nodes x k With node x i The sum of straight-line distances in Euclidean space;

[0052] Obtain the joint probability distribution of the high-dimensional space and the space where the matrix is ​​located;

[0053]

[0054] Among them, (1+||y i -y j || 2 ) -1 Represents the matrix D in high-dimensional space ij The corresponding node y j With node y i The straight-line distance between two vectors in Euclidean space, Σ k≠i (1+||y k -y i || 2 ) -1 Represents the matrix D in high-dimensional space ij The corresponding node y k With node y i The straight-line distance between two vectors in Euclidean space;

[0055] KL divergence is introduced to measure the deviation between the two spaces, and KL divergence is set as the objective function C:

[0056]

[0057] Calculate the objective function C for each two-dimensional node y i Gradient:

[0058]

[0059] Then, the minimum objective function is obtained through the gradient descent algorithm, and finally the two-dimensional node coordinate Y after dimensionality reduction can be obtained.

[0060] In an optional embodiment, the two-dimensional coordinate Y is placed in a sub-two-dimensional plane, the grid operation characteristic information corresponding to each node is composed into a channel and assigned to the corresponding node coordinate, and the channels composed of the various characteristic quantities are superimposed and input to form the input tensor INPUT, which specifically includes the following process:

[0061] The value range of the node coordinate Y is [0, 1]. In order to denormalize the coordinate Y and amplify it into an integer, the amplification factor is determined by the needs of CNN and the actual situation of the power system node.

[0062] Choose a magnification integer z and set the node coordinate y to be in the range [0,1] i By using the normalized inverse operation method to enlarge it to [0, h], and then rounding the enlarged node coordinates to get the integer form of the node coordinate Y INT , the normalized equation is:

[0063]

[0064] Among them, GYH represents the normalization operation, Y max is the maximum value of the node coordinate, Y min is the minimum coordinate value of the node;

[0065] The grid operation characteristic information is individually composed of a channel and assigned to each node in the grid. The value is assigned to the corresponding node coordinates, and the channels composed of various characteristic quantities are superimposed and input to form the input tensor INPUT. The grid operation characteristic information includes bus active load and reactive load, bus voltage amplitude and phase angle, line active power and reactive power, bus phase angle difference at both ends of the line, total system active load, and total system reactive load.

[0066] It should be noted that power grid topology diagrams typically contain a large number of nodes and complex connection relationships, making direct analysis of these relationships difficult. Using a nonlinear dimensionality reduction algorithm, the high-dimensional electrical distance matrix can be converted into two-dimensional coordinates, allowing the grid topology to be intuitively displayed on a plane. Compared to the high-dimensional electrical distance matrix, the two-dimensional coordinate representation significantly reduces computational complexity, making subsequent analysis and processing more efficient. Through this intuitive two-dimensional coordinate representation, engineers or researchers can more easily identify potential problems or anomalies in the power grid, thereby improving the accuracy and reliability of analysis.

[0067] The grid power stability control calculation module takes the input tensor as the input of the CNN model and obtains the output matrix of the grid power stability control;

[0068] In an optional embodiment of the present invention, data encoding is performed before input into the CNN model, and the extracted power grid operation characteristic information is encoded into a format suitable for CNN processing; the data is standardized or normalized to ensure stability and convergence during the model training process.

[0069] In a preferred embodiment of the present invention, in the CNN model, the input layer is designed according to the preprocessed data format. If the data is encoded as a two-dimensional image, the input layer will be two-dimensional; if it remains as a one-dimensional sequence, a one-dimensional convolutional layer is used;

[0070] It should be noted that in the CNN (Convolutional Neural Network) model, it is very critical to design the input layer according to the preprocessed data format, because different data formats require different types of convolutional layers to process them. The following is a detailed explanation of this design idea:

[0071] When the data is encoded as a two-dimensional image, such as common RGB images or grayscale images, the input layer is usually two-dimensional and a two-dimensional convolutional layer is used.

[0072] The 2D convolutional layer performs convolution operations across the spatial dimensions (height and width) of an image, thereby capturing local features and patterns in the image.

[0073] This design is particularly suitable for image processing tasks such as image classification, object detection, image segmentation, etc.

[0074] If the data remains a one-dimensional sequence, such as time series data, text data (in some encoding method, such as a sequence after word embedding), or audio signals, the input layer will use a one-dimensional convolutional layer.

[0075] The one-dimensional convolutional layer can perform convolution operations on the temporal dimension or order dimension of the sequence, thereby capturing local features and patterns in the sequence.

[0076] This design is particularly suitable for sequence processing tasks such as text classification, sentiment analysis, and time series forecasting in natural language processing (NLP).

[0077] Data format matching: The design of the input layer must match the format of the preprocessed data. If the data is two-dimensional, use a two-dimensional convolutional layer; if the data is one-dimensional, use a one-dimensional convolutional layer.

[0078] Feature extraction: The main function of the convolutional layer is to extract local features from the data. The two-dimensional convolutional layer focuses on the spatial features in the image, while the one-dimensional convolutional layer focuses on the temporal or sequential features in the sequence.

[0079] Parameter sharing and sparse interaction: The convolutional layer reduces the number of model parameters through parameter sharing and sparse interaction, improves computational efficiency, and enhances the model's ability to capture local features.

[0080] Design multiple convolutional layers to extract spatial or temporal features of the data, and each convolutional layer is usually followed by an activation function to increase nonlinearity;

[0081] Use pooling layers to reduce the dimension of feature maps, reduce computational effort, and extract the most important features;

[0082] After the convolutional and pooling layers, a fully connected layer is added to transform the extracted features into the final output;

[0083] The design of the output layer depends on the specific control task. For power stabilization control, the output may be one or more control parameters, which constitute the output matrix.

[0084] During model training, a suitable loss function is selected to evaluate the difference between the model's predicted value and the actual value; an optimization algorithm is selected to update the model's weights to minimize the loss function.

[0085] The output matrix contains multiple control parameters, which are used to adjust the output of the generator, adjust the load distribution, and control the reactive compensation equipment to maintain the voltage stability, frequency stability and power balance of the power grid.

[0086] The self-driving control module controls the power of each node in the power grid according to the output matrix.

[0087] In a preferred embodiment of the present invention, power control of each node in the power grid based on the output matrix specifically includes the following process:

[0088] Select control strategy: Based on the output matrix, select the appropriate control strategy, which includes linear control, nonlinear control, optimal control or intelligent control;

[0089] Design a controller: Based on the control strategy, design a specific controller that can receive real-time data from the power grid and calculate the corresponding control inputs;

[0090] Implement controls:

[0091] Data acquisition: Real-time collection of grid operation data, including voltage, current, and power;

[0092] Calculate control input: Use the controller and output matrix to calculate the control input based on real-time data;

[0093] Execute control actions: convert control inputs into specific control actions, adjust generator output, and switch reactive compensation devices;

[0094] Monitor the status of the power grid: Continuously monitor the operating status of the power grid to ensure the achievement of control objectives;

[0095] Adjust control strategy: According to the actual operation of the power grid, adjust the control strategy and controller parameters in a timely manner to cope with changes in the power grid.

[0096] Example 3 is the third embodiment of the present invention, which differs from the first two embodiments in that:

[0097] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0098] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0099] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0100] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0101] Example 4, reference Figure 2 , the fourth embodiment of the present invention provides an asynchronous power grid panoramic intelligent self-driving control method based on deep learning, such as Figure 2 As shown, the method includes:

[0102] Step 1: obtaining a power grid topology diagram, wherein the generator, busbar, and load are connected to each other as nodes to form a power grid topology diagram, and each node contains power grid operation characteristic information;

[0103] Step 2: Based on the grid topology diagram, the electrical distance matrix is ​​calculated through the impedance matrix, and dimensionality reduction is performed to form the input tensor;

[0104] Step 3: The input tensor is used as the input of the CNN model to obtain the output matrix of the power grid stability control;

[0105] Step 4: Perform power control on each node of the power grid according to the output matrix.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A deep learning-based, panoramic intelligent self-driving control system for asynchronous power grids, characterized by: include, A power grid data acquisition module is used to obtain a power grid topology diagram; The power grid data visualization module calculates the electrical distance matrix based on the power grid topology diagram through the impedance matrix and performs dimensionality reduction transformation to form the input tensor; The grid power stability control calculation module takes the input tensor as the input of the CNN model and obtains the output matrix of the grid power stability control; The self-driving control module controls the power of each node in the power grid according to the output matrix.

2. The deep learning-based asynchronous power grid panoramic intelligent self-driving control system according to claim 1, characterized in that: Calculating the electrical distance matrix using the impedance matrix includes: Obtaining a node admittance matrix based on a power grid topology diagram, wherein the node admittance matrix represents the relationship between currents of each node in the power grid topology diagram when voltage is used as an excitation; Obtain an inverse matrix based on the node admittance matrix, which is an impedance matrix; Based on the impedance matrix, the node connection impedance matrix is ​​obtained by superposition principle; Computes the magnitude of the elements in the contact impedance matrix that correspond to the elements of the electrical distance matrix between nodes.

3. The deep learning-based asynchronous power grid panoramic intelligent self-driving control system according to claim 2, characterized in that: The dimensionality reduction conversion includes: The electrical distance matrix is ​​converted into a high-dimensional space, in which the row vectors of the multidimensional vector X are defined to represent the spatial distribution of each node; Define the conditional probability, which represents the node x corresponding to the jth row of the multidimensional vector X j Appears at node x i Probability of nearby; Obtain the joint probability distribution of the high-dimensional space and the space where the matrix is ​​located; KL divergence is introduced to measure the deviation between the two spaces, and KL divergence is set as the objective function; Calculate the gradient of the objective function for each two-dimensional node; Then, the minimum objective function is obtained through the gradient descent algorithm, and finally the two-dimensional node coordinates after dimensionality reduction can be obtained.

4. The deep learning-based asynchronous power grid panoramic intelligent self-driving control system according to claim 3, characterized in that: The dimensionality reduction conversion further includes: Place the two-dimensional coordinates in the sub-two-dimensional plane, organize the grid operation characteristic information corresponding to each node into a channel and assign it to the corresponding node coordinates, and superimpose the channels composed of various characteristic quantities into the input tensor INPUT.

5. The deep learning-based asynchronous power grid panoramic intelligent self-driving control system according to claim 4, characterized in that: The input tensor INPUT includes: The value range of the node coordinate Y is [0, 1]. In order to denormalize the coordinate Y and amplify it into an integer, the amplification factor is determined by the needs of CNN and the actual situation of the power system node. Choose a magnification integer z and set the node coordinate y to be in the range [0,1] i By using the normalized inverse operation method to enlarge it to [0, h], and then rounding the enlarged node coordinates to get the integer form of the node coordinate Y INT , normalized by the normalization equation; The grid operation characteristic information is individually composed of a channel and assigned to each node in the grid. The value is assigned to the corresponding node coordinates, and the channels composed of various characteristic quantities are superimposed and input to form the input tensor INPUT. The grid operation characteristic information includes bus active load and reactive load, bus voltage amplitude and phase angle, line active power and reactive power, bus phase angle difference at both ends of the line, total system active load, and total system reactive load.

6. The deep learning-based asynchronous power grid panoramic intelligent self-driving control system according to claim 5, characterized in that: The CNN model includes, Design the input layer according to the preprocessed data format. If the data is a two-dimensional image, use a two-dimensional input layer. If the data is a one-dimensional sequence, use a one-dimensional convolutional layer. Design multiple convolutional layers to extract spatial or temporal features from the input data through convolution operations, and equip the convolutional layers with activation functions to introduce nonlinearity; Use the pooling layer to reduce the dimension of the feature map and extract features; A fully connected layer is added after the convolutional layer and the pooling layer to merge the feature information extracted by the convolutional layer and pass it to the next layer. According to the requirements of the control task, the output layer converts the output of the fully connected layer into grid control parameters. The output matrix contains the control signal used to adjust the grid power stability. The output matrix includes multiple control parameters, which are used to adjust the output of the generator, adjust the load distribution, and control the reactive compensation equipment to maintain the voltage stability, frequency stability and power balance of the power grid.

7. The deep learning-based asynchronous power grid panoramic intelligent self-driving control system according to claim 6, characterized in that: The power control of each node of the power grid according to the output matrix includes: Based on the output matrix, select the control strategy, including linear control, nonlinear control, optimal control or intelligent control; According to the control strategy, the controller is designed. The controller should be able to receive real-time data from the power grid and calculate the corresponding control input.

8. A method for controlling an asynchronous power grid with panoramic intelligent self-driving technology based on deep learning, using the method of controlling an asynchronous power grid with panoramic intelligent self-driving technology based on deep learning according to any one of claims 1 to 7, characterized in that: Methods include: Obtain the power grid topology diagram; Based on the grid topology diagram, the electrical distance matrix is ​​calculated through the impedance matrix, and dimensionality reduction is performed to form the input tensor; The input tensor is used as the input of the CNN model to obtain the output matrix of the power grid stability control; The power of each node in the power grid is controlled according to the output matrix.

9. 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 the asynchronous power grid panoramic intelligent self-driving control method based on deep learning according to claim 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the asynchronous power grid panoramic intelligent self-driving control method based on deep learning according to claim 7 are implemented.

Images