Unit operation mode rapid optimization method, system and equipment integrating space-time diagram convolutional network and direct current delivery optimization, and medium

Abstract

The invention discloses a unit operation mode rapid optimization method, system and device fusing a space-time diagram convolutional network and direct current delivery optimization and a medium, and belongs to the technical field of power system optimization scheduling, and the method comprises the steps: constructing and training a space-time diagram convolutional network model fusing the physical characteristics and operation rules of a power system, predicting the optimal operation mode of the unit; predicting a current power grid time sequence section by using the space-time diagram convolutional network model, and outputting a preliminary unit operation mode scheme; taking the prediction result as an initial solution set, and carrying out rapid optimization verification and correction on the prediction result; safety and stability verification is carried out on the corrected operation mode scheme; and outputting a final unit operation mode optimization scheme. According to the method, preliminary prediction is carried out by introducing the space-time diagram convolutional network, the unit operation mode close to the optimal solution can be instantly given, and the bottleneck that iterative calculation of a traditional optimization algorithm is time-consuming is overcome.

Description

Technical Field

[0001] This invention relates to the field of power system optimization and dispatching technology, specifically to a method, system, equipment, and medium for rapid optimization of unit operation modes by integrating spatiotemporal graph convolutional networks and DC transmission optimization. Background Technology

[0002] The integration of a high proportion of renewable energy sources poses a challenge to the stability of the power system. For DC transmission systems using asynchronous interconnection, it is of great significance to study how to ensure the large-capacity and stable transmission of DC channels under the condition of a high proportion of renewable energy sources.

[0003] The increasing scale and complexity of the power grid, coupled with the rapid and frequent fluctuations in the output of renewable energy units, places higher demands on the daily adjustment of unit operation modes in the sending-end system. Existing DC-OCO optimization methods that take into account the adjustment of the DC sending-end system's support capacity are suitable for adjusting unit operation modes for static aspects of system operation, but cannot yet meet the needs of optimizing the grid's time-series operation under a high proportion of renewable energy integration.

[0004] The DC-OCO method for optimizing DC transmission capacity adjustment at the DC sending end mainly adopts an optimization method that combines intelligent optimization algorithms (such as the multi-objective particle swarm optimization algorithm IMOPSO) with power system simulation software (such as PSD). This method establishes an optimization model with the imbalance of DC transmission capacity, renewable energy absorption capacity and voltage support capacity as objectives. Based on the constraints of system voltage and frequency support capacity, it performs static section optimization of the unit operation mode.

[0005] In addition, existing research has also attempted to introduce artificial intelligence technologies (such as neural networks, graph convolutional networks (GCN), long short-term memory networks (LSTM), etc.) to assist in unit combination optimization in order to improve computational efficiency. These methods typically reduce the complexity of the optimization problem by building data-driven models to predict unit start-up and shutdown states or output plans.

[0006] Existing technologies have the following objective drawbacks: 1. While existing intelligent algorithm-based optimization methods (such as IMOPSO-PSD joint optimization) can achieve good optimization results under static cross-sections, their computational complexity is high, the number of iterations is large, and the solution speed is slow, making it difficult to meet the real-time requirements of frequent adjustments to the power grid's time-series operation mode. 2. Intelligent optimization algorithms are highly dependent on the quality of the initial population. If the quality of the initial solution is poor, it is easy to get trapped in local optima, resulting in unsatisfactory optimization results or even failure to converge. 3. Although some intelligent optimization methods can accelerate the optimization process to a certain extent, most models do not fully consider the temporal characteristics and spatial topology of power grid operation data, making it difficult to accurately capture the dynamic correlation between new energy output fluctuations and load changes, resulting in limited prediction accuracy. 4. Existing intelligent prediction models often focus on economic or single-objective optimization, failing to incorporate system voltage and frequency support capabilities as hard constraints into the prediction process, which may lead to stability problems in practical applications. 5. The models have poor generalization ability and rely on scenario classification. Some studies use clustering methods to classify load scenarios and train models separately, which not only increases model complexity and training cost but also reduces the generality and practicality of the method. 6. Existing data-driven models often operate independently of physical models, failing to effectively integrate power system operating rules and physical constraints, leading to risks in the physical feasibility of the prediction results. Summary of the Invention

[0007] In view of the above-mentioned problems, the present invention is proposed.

[0008] Therefore, the technical problem solved by this invention is: how to construct an optimization method for the operation mode of DC sending-end system units that has both rapid response capability and strict safety and reliability, overcome the shortcomings of low computational efficiency of traditional optimization algorithms and insufficient security of existing artificial intelligence models, so as to directly support the real-time scheduling decision of the power grid in the environment of high proportion of new energy access.

[0009] To address the aforementioned technical problems, this invention provides the following technical solution: a rapid optimization method for unit operation modes that integrates spatiotemporal graph convolutional networks and DC transmission optimization, comprising, Construct and train a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, and use it to predict the optimal operating mode of the generating units based on the time-series operation data of the power grid; The spatiotemporal graph convolutional network model is used to predict the current power grid time series section and output a preliminary unit operation mode scheme; Using the prediction results as the initial solution set, the DC transmission optimization model is used to quickly optimize, verify, and correct the prediction results. The safety and stability of the modified operation mode scheme were verified using a DC power transmission optimization model. The output is the final optimized unit operation mode scheme that satisfies all safety constraints, and the spatiotemporal graph convolutional network model is updated based on the newly added optimized samples.

[0010] As a preferred embodiment of the rapid optimization method for unit operation modes that integrates spatiotemporal graph convolutional networks and DC transmission optimization described in this invention, the method involves: constructing and training a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, used to predict the optimal operating mode of the units based on power grid time-series operating data. A spatiotemporal graph convolutional network model is constructed to learn the spatial correlation of power grid structure and the temporal evolution of operational data through spatiotemporal graph convolutional layers; Supervised training of the spatiotemporal graph convolutional network model was conducted using power grid time-series operation data and a sample set of corresponding optimization target values. In the training process of the spatiotemporal graph convolutional network model, physical characteristics and constraints reflecting the operating laws of the power system are incorporated.

[0011] Furthermore, when learning from historical data, the model can internalize actual operating rules such as unit output limits and power balance into its prediction logic. Therefore, when the model of this invention makes rapid predictions for new grid sections, its initial output solution spontaneously meets basic physical feasibility and operational compliance, providing a higher-quality starting point for the optimization and verification steps that is closer to the final feasible solution, and reducing the iterative adjustment time required for the overall optimization process.

[0012] As a preferred embodiment of the rapid optimization method for unit operation modes that integrates spatiotemporal graph convolutional networks and DC transmission optimization as described in this invention, the step of using a spatiotemporal graph convolutional network model to predict the current power grid time series profile and outputting a preliminary unit operation mode scheme includes: The time-series cross-sectional data representing the current system operating state are input into the trained spatiotemporal graph convolutional network model; The spatiotemporal graph convolutional network model, after being trained, generates corresponding unit output prediction results based on the learned mapping relationships; The prediction results will be output as a preliminary unit operation mode plan.

[0013] As a preferred embodiment of the rapid optimization method for unit operation modes that integrates spatiotemporal graph convolutional networks and DC transmission optimization as described in this invention, the step of using the prediction results as the initial solution set and employing a DC transmission optimization model to rapidly optimize, verify, and correct the prediction results includes... Based on the preliminary unit operation mode scheme, an initial solution set for iterative optimization is constructed; An optimization model for DC power transmission is adopted, and iterative calculations are performed based on the initial solution set. The preliminary scheme is revised and its feasibility is verified through iterative calculations, and an optimized unit operation mode scheme is output.

[0014] As a preferred embodiment of the rapid optimization method for unit operation modes that integrates spatiotemporal graph convolutional networks and DC transmission optimization as described in this invention, the step of verifying the safety and stability of the modified operation mode scheme through the DC transmission optimization model includes: Calculate and verify the multi-site short-circuit ratio of each new energy power station, and require it to be no less than the critical value; Evaluate and verify that the maximum transient frequency deviation of the system does not exceed the safety limit under DC blockage fault scenarios; Verify and confirm that the power flow distribution of each line in the system does not exceed the load capacity constraint.

[0015] This invention specifies that safety verification must include concrete verification of short-circuit ratios, transient frequency deviations, and power flow loads at multiple substations. This technical feature clarifies the general requirement of "safety and stability" as the verification of compliance with the three most critical and quantifiable rigid safety indicators in power system operation. By calculating and controlling these three indicators one by one in the optimized scheme, it ensures that the final optimized scheme not only pursues economic goals but also directly meets the hard technical requirements of the power grid in terms of voltage support, frequency stability, and equipment safety, thus enabling the scheme to have the reliability and security to directly support real-time dispatch decisions.

[0016] As a preferred embodiment of the rapid optimization method for unit operation modes that integrates spatiotemporal graph convolutional networks and DC transmission optimization described in this invention, the output is a final optimized unit operation mode scheme that satisfies all safety constraints, and the spatiotemporal graph convolutional network model is updated based on newly added optimization samples. The unit operation mode schemes that have passed safety and stability verification will be output as the final optimized schemes that can be used for scheduling and execution. Save the power grid time-series section data, the final optimization scheme and the verification results corresponding to the current optimization as new samples; When the number of new samples accumulates to a preset size, the spatiotemporal graph convolutional network model is retrained and continuously updated using the updated sample set containing the new samples.

[0017] As a preferred embodiment of the rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional network and DC transmission optimization described in this invention, wherein: the verification and validation of the power flow distribution of each line of the system not exceeding the load capacity constraint includes unit output constraint, ramping constraint and power balance constraint; The unit output constraint is expressed as: in, , These represent the lower and upper limits of the generator unit's output at time t, respectively, with the lower limit of the new energy unit's output being 0. Constrained by maximum capacity active power fluctuation faults and DC blocking faults, the lower limit of output for each synchronous generator unit participating in frequency regulation is set based on the current DC transmission situation, expressed as: in, , These represent the minimum output of different hydropower and thermal power synchronous generating units participating in frequency regulation at time t. and These refer to the number of hydropower and thermal power units participating in frequency regulation. The active power surplus is filled by the load participating in frequency regulation after the fault occurs at time t. The ramp constraint for the synchronous machine is set as follows: in, and These represent the upper and lower limits of the unit's operating ramp rate, respectively. , and These represent the unit outputs of the synchronous generator at time t, time t-1, and time 0, respectively. The system power balance constraint is expressed as follows: in, 、 、 These represent the generator active power output, load level, and line active power loss at time t, respectively. Let t be the DC transmission capacity of the system at time t. 、 、 These represent the number of generators, the number of loads, and the number of lines, respectively.

[0018] This invention provides a rapid optimization system for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization.

[0019] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a rapid optimization system for unit operation mode that integrates spatiotemporal graph convolutional network and DC transmission optimization, comprising: a model building module, a preliminary solution output module, a verification and correction module, and a final solution output module; The model building module is used to build and train a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, and is used to predict the optimal operating mode of the units based on the time series operation data of the power grid. The preliminary output scheme module uses a spatiotemporal graph convolutional network model to predict the current power grid time series section and output a preliminary unit operation mode scheme. The verification and correction module uses the prediction results as the initial solution set and employs a DC transmission optimization model to quickly optimize, verify, and correct the prediction results; the DC transmission optimization model is used to verify the safety and stability of the corrected operation mode scheme. The output final scheme module outputs the final unit operation mode optimization scheme that satisfies all safety constraints, and updates the spatiotemporal graph convolutional network model based on the newly added optimization samples.

[0020] This invention provides a computer device, including a memory and a processor. The memory stores a computer program. The processor executes the computer program to implement the steps of the method for rapid optimization of unit operation mode by fusing spatiotemporal graph convolutional networks and DC transmission optimization.

[0021] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the aforementioned method for rapid optimization of unit operation mode by fusing spatiotemporal graph convolutional networks and DC transmission optimization.

[0022] The beneficial effects of this invention are as follows: By introducing a spatiotemporal graph convolutional network (SGC) for initial prediction, this invention can instantly provide a near-optimal unit operation mode, overcoming the bottleneck of time-consuming iterative calculations in traditional optimization algorithms. Subsequently, only a simplified DC-OCO model needs to be started for fine-tuning and verification, enabling rapid response and optimization of the power grid's time-series operation mode, meeting the stringent timeliness requirements of power grid dispatching in a high-proportion renewable energy environment. The STGCN model can simultaneously learn the spatial topology of the power grid (through the graph convolutional layer GCN) and the temporal evolution of operating parameters (through a gated temporal convolutional layer), thereby gaining a deeper understanding of the complex nonlinear mapping relationship between renewable energy output fluctuations, load changes, and the optimal unit operation mode. This makes the prediction results not only fast but also accurate in trend and with small errors. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization, as provided in one embodiment of the present invention.

[0025] Figure 2 The diagram shows the structure of a STGCN-based prediction model for optimal unit operation mode, which is a method for rapid optimization of unit operation mode by integrating spatiotemporal graph convolutional networks and DC transmission optimization, as provided in an embodiment of the present invention.

[0026] Figure 3 This is a diagram of the STGCN network model structure, which integrates the physical characteristics and grid operation rules of a rapid optimization method for unit operation mode that combines spatiotemporal graph convolutional networks and DC transmission optimization, as provided in an embodiment of the present invention.

[0027] Figure 4 The flowchart below shows the STGCN model sample generation method for a rapid optimization method of unit operation mode that integrates spatiotemporal graph convolutional network and DC transmission optimization, as provided in one embodiment of the present invention.

[0028] Figure 5 The spatiotemporal graph convolutional layer result diagram is provided as an embodiment of the present invention for a rapid optimization method of unit operation mode that integrates spatiotemporal graph convolutional network and DC transmission optimization. Detailed Implementation

[0029] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0030] Example 1 is an embodiment of the present invention, which provides a rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization, including: In DC transmission systems with a high proportion of renewable energy integration, grid dispatch faces the dual challenge of balancing the real-time requirements for frequent adjustments to operating modes with the need to ensure the safety and reliability of optimization results.

[0031] Both existing mainstream technical solutions have significant limitations: On the one hand, traditional methods based on the joint solution of intelligent optimization algorithms and detailed physical simulation (such as DC-OCO) can obtain optimization solutions with high security, but their iterative calculation process is complex and time-consuming, making it difficult to meet the needs of rapid decision-making on a minute-level or even shorter time scale.

[0032] On the other hand, while AI prediction models that rely solely on historical data have an advantage in computational speed, they often overlook the inherent physical operating laws and safety and stability constraints of the power system. Their output results pose risks in terms of power balance, voltage support, and frequency stability, and cannot be directly used for actual dispatching.

[0033] Therefore, this invention proposes a rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization.

[0034] S1. Construct and train a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, for predicting the optimal operating mode of the generating units based on the time-series operating data of the power grid; S2. Use the spatiotemporal graph convolutional network model to predict the current power grid time series section and output a preliminary unit operation mode scheme; S3 uses the prediction results as the initial solution set and employs a DC transmission optimization model to quickly optimize, verify, and correct the prediction results. S4. Verify the safety and stability of the modified operation mode scheme using the DC transmission optimization model; S5. Output the final optimized unit operation mode scheme that satisfies all safety constraints, and update the spatiotemporal graph convolutional network model based on the newly added optimized samples.

[0035] Example 2, refer to Figures 1-5 As an embodiment of the present invention, based on the previous embodiment, a rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization is provided, including: Data-driven deep learning models utilize historical data on state and target variables to fit the mapping relationship between system cross-sections and optimized operation objectives from a data perspective. Integrating algebraic equations from the physical model into deep learning can yield superior predictive performance compared to single physical or data model algorithms in specific scenarios. Guided fusion methods, with data-driven computation at their core, leverage explicit physical relationships within the physical model to guide the data-driven process, ensuring results align better with prior knowledge and experience. This approach also offers advantages such as speed and interpretability. The proposed prediction model requires both speed and the ability to leverage the physical model to improve the accuracy of the data model and provide interpretability for the prediction results. Therefore, a guided fusion method is employed. This method embeds mathematical models of DC transmission and renewable energy consumption into the loss function of the STGCN model, introducing the physical model of unit optimal operation into deep learning. The STGCN-based unit optimal operation prediction model, constructed using guided fusion and incorporating the physical characteristics and operating rules of the power system, is shown below. Figure 2 As shown The STGCN model comprises two spatio-temporal graph convolutional blocks (ST-Conv) connected by a residual structure, one pooling layer, five fully connected layers, one physical constraint layer, and a feedback mechanism that incorporates power grid operation characteristics. The internal structure of this model is described in detail below.

[0036] Step S1: Construct and train a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, used to predict the optimal operating mode of generating units based on power grid time-series operating data. This includes the following steps: S11. Construct a spatiotemporal graph convolutional network model to learn the spatial correlation of the power grid structure and the temporal evolution of operational data through spatiotemporal graph convolutional layers.

[0037] S12. Supervised training of the spatiotemporal graph convolutional network model is performed using power grid time-series operation data and a sample set of corresponding optimization target values.

[0038] use Figure 3 After the framework completes the construction of the STGCN model integrating grid operation characteristics, in order for the model to predict the optimal operating mode of the units, it is necessary to construct a labeled sample set for training. The main steps of the construction process are as follows: (1) Determine the cross sections to be optimized. Time series data of wind power, photovoltaic, and synchronous generator units such as hydropower and thermal power in areas with a high proportion of new energy sources are obtained. Combined with the power grid structure model, a sample to be optimized can be constructed. However, due to the large number of samples, in order to improve the efficiency of the optimization sample, typical cross sections are selected from the sample based on Euclidean distance to form the cross sections to be optimized, which will be optimized by the DC-OCO model in the future.

[0039] (2) Constructing a unit operation mode database. During the optimization of the DC-OCO model, numerous unit operation modes will be generated. These operation modes are deduplicated and stored to form a unit operation mode database. This database can provide initial values ​​for the DC-OCO model to accelerate optimization. Furthermore, after matching with atypical sections, optimization samples for atypical sections can be directly generated after minor adjustments and feasibility verification.

[0040] (3) Generate unit operation mode optimization samples. The time-series samples are obtained by arranging the optimized typical cross-section samples and the matched atypical cross-section samples in time sequence. The structure is a four-dimensional matrix (NS, NN, NT, 2). NS represents the number of samples, NN represents the number of nodes, NT represents the time-series data length, and 2 represents the number of features. Each sample contains active power and node load data for all nodes in the power grid. The grid structure data is G=(V, E, A), where V represents the set of nodes in the power grid diagram (e.g., power plants, load nodes); E represents the set of edges in the power grid diagram (e.g., transmission lines); and A is the adjacency matrix, representing the connection relationships between nodes in the power grid. By combining the time-series samples with the grid structure data, the unit operation mode optimization samples can be obtained, such as... Figure 4 As shown.

[0041] S13. In the training process of the spatiotemporal graph convolutional network model, physical characteristics and constraints that reflect the operating laws of the power system are incorporated.

[0042] The physical constraint layer in the model improves the model's learning efficiency in solving power system operation optimization problems by adding power system operation constraints to the electrical quantities in the STGCN model. The expressions for the physical constraint layer include: (a) Unit output constraints, i.e. (4) in, , These represent the lower and upper limits of the generator unit output at time t, respectively. The upper and lower limits differ for different units; the lower limit for new energy units is 0. To ensure sufficient synchronous frequency regulation capacity, the lower limit of the output of each synchronous generator unit participating in frequency regulation is set based on the current DC transmission situation, constrained by the maximum capacity active power fluctuation fault—DC blocking fault. (5) in, , These represent the minimum output of different hydropower and thermal power synchronous generating units participating in frequency regulation at time t. and These refer to the number of hydropower and thermal power units participating in frequency regulation. This represents the active power surplus that is replenished by the load participating in frequency regulation after the fault occurs at time t.

[0043] (b) Synchronous Generator Climbing Constraint. The rotor of a synchronous generator has significant mechanical inertia, and changes in output require a certain amount of time. If the output changes too rapidly, it may cause excessive mechanical stress on the rotor, thereby damaging the equipment. Therefore, a climbing constraint is set for the synchronous generator in the model, i.e. (6) in, and These represent the upper and lower limits of the unit's operating ramp rate, respectively. , and These represent the unit outputs of the synchronous generator at time t, time t-1, and time 0, respectively. (c) System power balance constraints, i.e. (7) in, 、 、 These represent the generator active power output, load level, and line active power loss at time t, respectively. Let t be the DC transmission capacity of the system at time t. 、 、 These represent the number of generators, the number of loads, and the number of lines, respectively.

[0044] The feedback mechanism integrating grid operation characteristics is achieved by introducing DC transmission capacity and renewable energy absorption capacity indicators into the model's loss function. This approach improves the model's training speed and effectively uncovers the characteristics in the unit optimization task. Loss function The model is shown below: (8) Where xi and yi are the i-th actual value and the predicted value, respectively; and These are hyperparameters that weigh the importance of two error terms: DC transmission loss and renewable energy consumption loss. Their values ​​need to be adjusted appropriately based on the training results. The DC transmission loss and renewable energy consumption loss are shown below: (9) (10) in, and It is a constant that ensures the non-negativity of the loss function calculation result; For the active power output of the i-th unit; and These represent the active power outputs of the i-th wind turbine and the j-th photovoltaic unit, respectively. Let be the size of the j-th active load.

[0045] In the implementation of this application, the construction of the spatiotemporal graph convolutional network model in S11 is based on the spatiotemporal convolutional layer to construct a spatiotemporal graph convolutional network STGCN model that integrates the physical characteristics and operating rules of the power system.

[0046] Construction of Spatial Convolutional Layers: Convolutional Neural Networks (CNNs) and Graph Convolutional Networks (GCNs) are classic neural network models commonly used for feature extraction. Since traditional CNNs cannot handle non-Euclidean space data, GCNs, which are more suitable for processing graph-structured data, are used as the core tool for processing graph-structured data in power systems.

[0047] For a power system, the graph structure of the power grid can be defined as G = (V, E, A), where: V represents the set of nodes in the power grid graph (e.g., power plants, load nodes); E represents the set of edges in the power grid graph (e.g., transmission lines); and A is the adjacency matrix, representing the connection relationships between the nodes in the power grid. The basic idea of ​​graph convolution is to aggregate information about neighboring nodes through the adjacency matrix and the node feature matrix. The inter-layer propagation method of GCN can be defined as: In the formula, It is an activation function (such as ReLU). Indicates the first The input to the layer graph convolution is the feature matrix of the nodes in layer l. The adjacency matrix represents the structure of the power grid graph. The degree matrix representing the structure of the power grid diagram. Represents the identity matrix. This represents the weight parameter matrix.

[0048] Construction of temporal convolutional layers: To address issues such as time consumption of serial iterations and complexity of gating mechanisms, a gated convolutional neural network (Gated CNN) with a gated linear unit (GLU) is used in the temporal dimension to capture the dynamic changes of the time series during operation, and residual connections are used to mitigate overfitting.

[0049] The temporal convolutional layer is designed to consist of a one-dimensional causal convolution followed by a gated linear unit as a nonlinear layer. This design enables feature extraction from time-series data of varying lengths. The introduction of the gated linear unit effectively fuses linear and nonlinear transformations. The input to each node's temporal convolutional layer is active power output and load data. ,in Indicates the number of electrical features selected. This represents the length of the input time series. The convolution kernel is designed as follows: Used for The output obtained by performing a one-dimensional convolution operation can be defined as: In the formula, and These are the inputs of the GLU gates. It is an activation function. It is the number of output channels. It is the size of the convolution kernel.

[0050] Construction of the Spatiotemporal Graph Convolutional Layer ST-Conv: To simultaneously extract the complex temporal and spatial dependencies in the unit operation optimization problem, a spatiotemporal graph convolutional layer ST-Conv is constructed by combining a temporal gated convolutional layer and a spatial graph convolutional layer. The structure of the spatiotemporal graph convolutional layer is designed with a spatial layer sandwiched between two temporal layers, thereby achieving fast spatial state propagation between the two temporal convolutions. The ST-Conv structure is as follows: Figure 5 As shown.

[0051] In ST-Conv, the temporal convolutional layer Γ0 utilizes active power output and load data. As its input, the adjacency matrix adopts the power grid graph structure. As a spatial convolutional layer The input. First, Input temporal convolutional layer This method utilizes one-dimensional causal convolution and GLU to capture temporal dependencies, achieving convolution operations in the temporal dimension. Then, the obtained temporal features are combined with... pass The obtained spatial features are combined and activated to achieve graph convolution in the spatial dimension. This is then passed through a temporal convolutional layer. The process further integrates temporal and spatial features to ultimately obtain a fused spatiotemporal feature result. The expression is: ST-Conv not only extracts the spatial characteristics of the power grid, but also integrates the temporal characteristics of the power grid operation process, providing strong support for achieving comprehensive and reliable prediction of the optimal operating mode of the generating units.

[0052] The Spatiotemporal Graph Convolutional Network (STGCN) model comprises two spatiotemporal graph convolutional layers (ST-Conv) connected by a residual structure, one pooling layer, five fully connected layers, one physical constraint layer, and a feedback mechanism that incorporates grid operation characteristics. ST-Conv extracts the spatial characteristics of the grid and the temporal variation characteristics of electrical quantities during operation. The output is then passed through the pooling layer, fully connected layer, and physical constraint layer, and a linear transformation is used to obtain the prediction results of the optimal operating mode of the generating units.

[0053] In one alternative implementation, the spatiotemporal graph convolutional network model constructed in S11 has a specific structure that includes two spatiotemporal graph convolutional layers (ST-Conv), one pooling layer, five fully connected layers, and one physical constraint layer, wherein the two ST-Conv layers are connected by residual connections.

[0054] In another alternative implementation, the construction of the spatiotemporal graph convolutional network model in S11 is achieved through the physical characteristics and constraints during training, using a physical constraint layer and a feedback mechanism that integrates the characteristics of power grid operation.

[0055] The physical constraint layer directly applies hard constraints on the network output regarding unit output and power balance; the feedback mechanism guides the network to learn the system optimization objective in the form of soft constraints by introducing DC transmission capacity and new energy consumption capacity indicators into the loss function.

[0056] This invention, by constructing and training a Spatiotemporal Graph Convolutional Network (STGCN) model, lays the foundation for an intelligent core system that combines high-speed inference capabilities, physical plausibility, and high prediction accuracy for the entire unit operation optimization system. It is not merely a simple model training process, but a fundamental reconstruction of the solution path for the optimization problem: First, during the training process, the model internalizes physical constraints such as power balance and unit ramping into its prediction logic through guided fusion, enabling its rapid forward calculation results for any new cross-section to spontaneously meet the basic operating rules of the power system. This provides a high-quality initial solution with low violation risk for the optimization process, shortening the iterative process of fine-grained verification and adjustment.

[0057] Secondly, the spatiotemporal graph convolutional architecture adopted by the model can process the spatial correlation of the power grid topology and the temporal evolution of operating parameters in parallel at one time. It realizes efficient learning and instantaneous calculation of complex nonlinear mapping relationships in massive time series cross-sectional data, transforming the time-consuming search process in traditional iterative optimization algorithms into an efficient forward prediction, breaking through the bottleneck of traditional methods in terms of computational speed.

[0058] Third, since the predicted starting point is physically feasible and close to the global optimum in terms of trend, the search space of the subsequent simplified optimization algorithm is greatly reduced. This not only improves the convergence speed and stability of the overall optimization process, but also effectively avoids the risk of getting trapped in local optima due to poor initial solution quality.

[0059] S2. Using a spatiotemporal graph convolutional network model to predict the current power grid time series profile and output a preliminary unit operation mode scheme includes the following steps: S21. Input the time-series cross-sectional data representing the current system operating state into the trained spatiotemporal graph convolutional network model.

[0060] Collect or receive real-time operating data of the target DC transmission system at the time to be optimized and several preceding time sections.

[0061] A matrix with dimensions [number of nodes, time step, number of features], where the features at least cover the active power output and active power load values ​​for each node (e.g., power plant, load point). This data serves as the active power output and load data input to the STGCN model.

[0062] The graph structure is represented by G=(V, E, A), where V is the set of nodes, E is the set of edges (transmission lines), and A is the adjacency matrix representing the electrical connection relationship between nodes.

[0063] This graph structure data forms the basis for graph convolution operations, corresponding to the power grid graph structure G defined in the original text and the adjacency matrix A, which serves as the input to the spatial convolutional layer.

[0064] S22. The spatiotemporal graph convolutional network model, after training, generates the corresponding unit output prediction results based on the learned mapping relationship.

[0065] S23. Output the prediction results as a preliminary unit operation mode plan.

[0066] The last layer of the STGCN model (physical constraint layer and output layer) performs linear transformation and constraint mapping on the high-dimensional features output by the aforementioned fully connected layer, and finally generates a unit active power output prediction vector.

[0067] The dimension of this vector is equal to the number of all schedulable generator units (including synchronous generator units and new energy generator units) in the system. The value of each element in the vector directly corresponds to the suggested active power output value of a specific generator unit at the target optimization time.

[0068] This vector represents the preliminary unit operation mode scheme. This scheme is generated in one go by the model during forward propagation, with a computation time on the order of seconds. Because physical constraints were embedded during model training, this scheme possesses inherent rationality in terms of power balance and the feasible output range of the unit.

[0069] The active power output prediction vector of this unit will be provided directly to the simplified DC transmission optimization model downstream as a high-quality initialization scheme to initiate its rapid and refined iterative verification and adjustment process.

[0070] In this embodiment, the mapping relationship in S22 involves feeding the node-time series feature matrix (X) and the power grid topology adjacency matrix (A) constructed in step S21 into the loaded and fixed-parameter STGCN model. The model automatically performs calculations according to a predefined architecture. First, the input data flows through the first spatiotemporal graph convolutional layer (ST-Conv Block), which consists of a temporally gated convolutional layer, a spatial graph convolutional layer, and another temporally gated convolutional layer connected in series, completing the initial extraction and fusion of the spatiotemporal correlation features of the input data.

[0071] Subsequently, the feature data is processed through a second spatiotemporal graph convolutional layer for deeper spatiotemporal feature mining.

[0072] After dimensionality reduction by pooling layers, the data is fed into a subnetwork consisting of multiple fully connected layers for high-dimensional nonlinear transformation and feature integration.

[0073] In an optional implementation, the mapping relationship of S22 is achieved by changing the propagation rules of graph convolution: the aggregation method of node features in the spatial graph convolution layer adopts the graph attention network (GAT) mechanism based on attention weights, which dynamically adjusts the weight of neighbor node information aggregation by calculating the attention coefficients between adjacent nodes, thereby replacing the graph convolution propagation rules based on fixed normalized adjacency matrices in the original scheme.

[0074] In another alternative implementation, the mapping relationship in S22 is achieved by replacing the gating mechanism of the temporal convolutional layer: the gated linear units (GLUs) in the temporal gated convolutional layer are replaced by a recurrent neural network module based on a long short-term memory network (LSTM) or a gated recurrent unit (GRU). This recurrent neural network module processes sequential data step by step, captures temporal dependencies through its internal gating mechanism, and ultimately outputs temporal features of the same dimension.

[0075] This invention transforms the time-consuming iterative search process in traditional optimization methods into a single, efficient neural network forward computation, thereby reducing the time to obtain a feasible initial solution from tens of minutes to seconds or even milliseconds, fundamentally solving the bottleneck of real-time dispatch response speed. Furthermore, since this predictive solution originates from an intelligent model that integrates the physical constraints of the power system, its output spontaneously satisfies basic operating rules such as power balance and unit output limits, possessing excellent physical feasibility and quality as a starting point for optimization.

[0076] S3. Using the prediction results as the initial solution set, the prediction results are quickly optimized, verified, and corrected using the DC transmission optimization model, including the following steps: S31. Based on the preliminary unit operation mode scheme, construct an initial solution set for iterative optimization.

[0077] Based on the statistical characteristics of the prediction error of the STGCN model, the preliminary active power output prediction scheme output by the model is set as the baseline. Around this baseline, and considering the allowable output range of each unit, a group (e.g., dozens) of unit output combinations are generated according to preset perturbation rules (e.g., small-range random fluctuations based on error distribution), forming the initial particle swarm for subsequent optimization algorithms. This process expands a single prediction point into a high-quality initial population that can be directly used by swarm intelligence algorithms.

[0078] S32. An optimization model for DC power transmission is adopted, and iterative calculations are performed based on the initial solution set.

[0079] A pre-configured simplified DC-OCO optimization model is invoked. This model accelerates computation by significantly reducing the number of iterations, limiting the particle population size, and eliminating the mutation operator.

[0080] The initial particle swarm generated in step S31 is directly loaded into the simplified model as the starting point for its iterative search. The model takes the imbalance between DC transmission capacity, new energy absorption capacity, and voltage support capacity as optimization objectives, and performs rapid optimization calculations within a limited number of iterations based on this initial population.

[0081] S33. The preliminary scheme is revised and its feasibility is verified through iterative calculations, and the optimized unit operation mode scheme is output.

[0082] During the rapid iteration of the simplified DC-OCO model, two aspects are processed simultaneously: First, by using the objective function as a guide, the power output combination of the units represented by the particle swarm is updated, thereby correcting and improving the accuracy of the initial scheme in the optimization direction.

[0083] Secondly, during each iteration of the evaluation, the best-performing particles undergo rapid operational mode verification, including calculating the multi-station short-circuit ratio (MRSCR) of the corresponding key operating sections and the line load rate. After the iteration terminates, the unit output combination represented by the particles with the best overall performance and that have passed the preliminary verification is selected from the final population, and the output is the optimized unit operation mode scheme after a rapid optimization and coarse screening.

[0084] In the implementation of this application, the DC transmission optimization model of S3 is designed to meet the high requirements for the feasibility of optimization results after a large number of new energy sources are connected. The above-mentioned STGCN-based unit optimal operation mode prediction model is connected in series with the simplified DC-OCO model. The prediction results of the prediction model are further corrected by the simplified DC-OCO model. While ensuring accuracy and reliability, the time for unit operation mode optimization is greatly shortened, and the STGCN prediction model is updated as optimization data accumulates.

[0085] In one alternative implementation, the DC transmission optimization model of S3 is achieved by adjusting its simplification strategy.

[0086] In the simplified DC-OCO model, the number of particles in its particle swarm optimization algorithm is fixed to a small-scale constant (e.g., 10-30), the maximum number of iterations is limited to a low range (e.g., 20-50 times), the mutation operator is completely eliminated, and the center of its search space is set to the initial scheme output by the STGCN model.

[0087] In another alternative implementation, the DC transmission optimization model of S3 is achieved by adjusting the combination or weights of its optimization objectives.

[0088] In the optimization objectives of the simplified DC-OCO model, in addition to DC transmission capacity and renewable energy consumption capacity, other objective items (such as minimizing system network losses, reducing the operating cost of synchronous generators, etc.) and their corresponding weight coefficients are selectively added or adjusted according to the real-time operating status of the system.

[0089] This invention uses the prediction results as a high-quality initial solution to guide a deeply simplified DC transmission optimization model to perform rapid local optimization, and can accurately correct the prediction scheme within a specified iteration.

[0090] During the optimization process, preliminary online calculations of core safety indicators such as short-circuit ratio and load rate are performed simultaneously, enabling preliminary feasibility screening of the proposed solutions. This improves the decision-making efficiency and deliverability of the optimization system.

[0091] S4. Verify the safety and stability of the modified operation mode scheme using the DC transmission optimization model, including the following steps: S41. Calculate and verify the multi-site short-circuit ratio of each new energy power station, and require it to be no less than the critical value.

[0092] Power flow calculations are performed on the system operating profile formed by the revised unit operation scheme to obtain the voltage and phase angle of each node in the entire network. For each renewable energy collection station or power plant, its multi-station short-circuit ratio (MRSCR) is calculated based on its grid connection point electrical parameters and the overall network impedance matrix. This calculated value is then compared and verified one by one with a pre-set critical MRSCR value to ensure system voltage stability, ensuring that the MRSCR value of all renewable energy power plants is not lower than its corresponding critical value. This verification directly guarantees the voltage support strength of the system under a high proportion of renewable energy access.

[0093] S42. Evaluate and verify that the maximum transient frequency deviation of the system does not exceed the safety limit under DC blocking fault scenarios.

[0094] A system dynamic model is constructed with the modified scheme as the initial state. DC blocking is set as a simulated fault, which simulates the instantaneous active power surplus at the maximum capacity of the sending-end system. A power system time-domain simulation is performed to obtain the dynamic response curve of the system frequency after the fault occurs. The maximum transient frequency deviation value (usually the difference between the lowest frequency drop point or the highest frequency spike point relative to the rated frequency) is extracted from this curve. This maximum transient frequency deviation value is compared and verified with the frequency deviation safety limit specified in the power system safe operation regulations to ensure that its absolute value does not exceed the limit. This verification directly ensures the frequency stability of the system under severe faults.

[0095] S43. Verify and confirm that the power flow distribution of each line in the system does not exceed the load capacity constraint.

[0096] Based on the modified scheme, power flow calculations are performed to obtain the active and reactive power flow distributions for all transmission lines (edges) in the system. The power flow calculation results for each line (usually expressed as apparent power or current) are compared with the line's rated current-carrying capacity (or thermal stability limit, static safe transmission capacity). This verifies whether the real-time power flow of each line does not exceed its corresponding load capacity constraint. This verification directly ensures the safe operation of transmission equipment and the overall static safety of the network.

[0097] S5. Output the final optimized unit operation mode scheme that satisfies all safety constraints, and update the spatiotemporal graph convolutional network model based on the newly added optimized samples, including the following steps: S51. Output the unit operation mode scheme that has passed the safety and stability verification as the final optimized scheme that can be used for scheduling and execution.

[0098] The fully approved unit operation mode scheme is converted into a standard instruction format that the target scheduling system can directly recognize and execute.

[0099] This format is typically a common, structured data exchange format in the field of power system dispatching (such as dispatching plan files based on CIM / E specifications or specific API data packets).

[0100] Through a communication link equipped with secure authentication and encryption functions, the converted final optimization scheme instruction file or data stream is sent to the planning execution subsystem or advanced application software of the power grid dispatch and control center.

[0101] This method completes the transformation from optimization calculations to production instructions, enabling the optimization results to be directly embedded and drive actual power grid dispatching operations.

[0102] S52. Save the power grid time-series section data, the final optimization scheme and the verification results corresponding to the current optimization as new samples.

[0103] The system automatically aligns all key data generated throughout the entire optimization process with a unified time stamp and establishes an index association to form a complete data record, which is then stored in a dedicated optimization case database.

[0104] The archived data includes, but is not limited to: (1) the original input that triggered this task, namely the time sequence section feature matrix X and the grid topology adjacency matrix A; (2) intermediate process data, such as the initial output prediction vector of the STGCN model, the population snapshot and objective function value sequence in the simplified DC-OCO model iteration process; (3) verification process data, including the specific calculation results of the short-circuit ratio of each new energy power station, the response curve of transient frequency simulation, and the power flow and load rate details of all lines; (4) the final output optimization scheme itself. This complete record is marked as a newly added "optimization sample".

[0105] S53. When the number of new samples accumulates to a preset size, the spatiotemporal graph convolutional network model is retrained and continuously updated using the updated sample set containing the new samples.

[0106] The system has a built-in automatic triggering mechanism that uses both sample size and calendar time as dual judgment conditions.

[0107] When the cumulative number of newly added optimized samples stored in the database reaches a preset threshold (e.g., 500), or when the time since the last model update has reached a preset period (e.g., 30 calendar days), the mechanism will automatically start the model update task.

[0108] All historical and new samples are extracted from the database and merged to generate a new training sample set covering a wider range of operating scenarios. Subsequently, the spatiotemporal graph convolutional network model is completely retrained using this new sample set. The training process strictly adheres to the network structure, physical constraint layers, and loss function incorporating physical objectives set during the initial model construction. After training, the generated new model parameters seamlessly replace the old model parameters in the online prediction service, thereby achieving closed-loop iteration and continuous enhancement of the model's predictive capabilities.

[0109] In the implementation of this application, the spatiotemporal graph convolutional network model in S5 is updated by merging the newly added optimized samples with the historical sample library when the accumulated number of samples reaches a preset threshold (e.g., 1000 sets) or a time period (e.g., one month), forming an updated training sample set. The spatiotemporal graph convolutional network model is then completely retrained using this sample set. That is, after initializing the model parameters, the model is trained from scratch using a mixed sample set containing new data, following the process described in S1, to obtain entirely new model parameters and achieve a comprehensive iteration and update of the model's predictive capabilities.

[0110] In one optional implementation, the spatiotemporal graph convolutional network model S5 is updated using an incremental learning or fine-tuning strategy. Specifically, the existing network structure and parameters of the spatiotemporal graph convolutional network model are kept unchanged, and the model is trained for additional rounds using only the newly added sample set. During this process, a lower learning rate can be used to enable the model to adaptively learn the latest operating characteristics and optimization patterns of the power grid reflected by the new samples, based on existing knowledge, thus achieving a gradual enhancement of the model's capabilities.

[0111] In another optional implementation, the S5 spatiotemporal graph convolutional network model updates by performing importance sampling or representativeness screening on the accumulated new samples before model updates. For example, based on the degree of improvement of the optimization objective corresponding to the sample or the scarcity of the operating scenario it represents, different weights are assigned to the samples, or the most representative subset of samples is selected. Subsequently, the model is retrained or fine-tuned using this weighted sample set or selected subset of samples. This approach aims to improve the efficiency and relevance of model updates, enabling the model to prioritize learning new knowledge that contributes the most to performance improvement or best covers potential future scenarios.

[0112] This invention automatically converts the optimization scheme generated by the algorithm into standard instructions that the scheduling system can recognize, enabling it to be directly executed, thus solving the engineering problem of interfacing intelligent algorithms with existing production control systems.

[0113] Secondly, by automatically archiving the entire process data of each optimization task, high-quality optimization cases with scenario tags are systematically accumulated. Its built-in automatic trigger update mechanism enables the model to iterate regularly using the newly accumulated knowledge base.

[0114] Example 3 is an embodiment of the present invention. This embodiment provides a rapid optimization system for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization, including a model building module, a preliminary scheme output module, a verification and correction module, and a final scheme output module.

[0115] The model building module is used to build and train a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, and is used to predict the optimal operating mode of the generating units based on the time-series operating data of the power grid.

[0116] The preliminary scheme output module uses a spatiotemporal graph convolutional network model to predict the current power grid time series profile and output a preliminary unit operation mode scheme.

[0117] The verification and correction module uses the prediction results as the initial solution set and employs a DC transmission optimization model to quickly optimize, verify, and correct the prediction results; it also uses the DC transmission optimization model to verify the safety and stability of the corrected operation mode scheme.

[0118] The final solution output module outputs the final optimized unit operation mode scheme that satisfies all safety constraints, and updates the spatiotemporal graph convolutional network model based on the newly added optimized samples.

[0119] This embodiment also provides an electronic device applicable to a rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization as proposed in the above embodiment.

[0120] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization as proposed in the above embodiment.

[0121] The storage medium proposed in this embodiment and the method for rapidly optimizing unit operation mode by integrating spatiotemporal graph convolutional networks and DC power transmission optimization proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0122] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A rapid optimization method for unit operation modes that integrates spatiotemporal graph convolutional networks and DC transmission optimization, characterized in that: include, Construct and train a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, and use it to predict the optimal operating mode of the generating units based on the time-series operating data of the power grid; The spatiotemporal graph convolutional network model is used to predict the current power grid time series section and output a preliminary unit operation mode scheme; Using the prediction results as the initial solution set, the DC transmission optimization model is used to quickly optimize, verify and correct the prediction results. The safety and stability of the modified operation mode scheme were verified using a DC power transmission optimization model. The output is the final optimized unit operation mode scheme that satisfies all safety constraints, and the spatiotemporal graph convolutional network model is updated based on the newly added optimized samples.

2. The rapid optimization method for unit operation mode by integrating spatiotemporal graph convolutional networks and DC transmission optimization as described in claim 1, characterized in that: The construction and training of a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, used to predict the optimal operating mode of generating units based on power grid time-series operating data, includes... A spatiotemporal graph convolutional network model is constructed to learn the spatial correlation of power grid structure and the temporal evolution of operational data through spatiotemporal graph convolutional layers; Supervised training of the spatiotemporal graph convolutional network model was conducted using power grid time-series operation data and a sample set of corresponding optimization target values. In the training process of the spatiotemporal graph convolutional network model, physical characteristics and constraints reflecting the operating laws of the power system are incorporated.

3. The rapid optimization method for unit operation mode by integrating spatiotemporal graph convolutional networks and DC transmission optimization as described in claim 2, characterized in that: The method of using a spatiotemporal graph convolutional network model to predict the current power grid time series profile and outputting preliminary unit operation mode schemes includes, The time-series cross-sectional data representing the current system operating state are input into the trained spatiotemporal graph convolutional network model; The spatiotemporal graph convolutional network model, after being trained, generates corresponding unit output prediction results based on the learned mapping relationships; The prediction results will be output as a preliminary unit operation mode plan.

4. The rapid optimization method for unit operation mode by integrating spatiotemporal graph convolutional networks and DC transmission optimization as described in claim 3, characterized in that: The step of using the prediction results as the initial solution set and employing a DC transmission optimization model to quickly optimize, verify, and correct the prediction results includes... Based on the preliminary unit operation mode scheme, an initial solution set for iterative optimization is constructed; An optimization model for DC power transmission is adopted, and iterative calculations are performed based on the initial solution set. The preliminary scheme is revised and its feasibility is verified through iterative calculations, and an optimized unit operation mode scheme is output.

5. The rapid optimization method for unit operation mode by integrating spatiotemporal graph convolutional networks and DC transmission optimization as described in claim 4, characterized in that: The verification of the safety and stability of the modified operation mode scheme through the DC transmission optimization model includes... Calculate and verify the multi-site short-circuit ratio of each new energy power station, and require it to be no less than the critical value; Evaluate and verify that the maximum transient frequency deviation of the system does not exceed the safety limit under DC blockage fault scenarios; Verify and confirm that the power flow distribution of each line in the system does not exceed the load capacity constraint.

6. The rapid optimization method for unit operation mode by integrating spatiotemporal graph convolutional networks and DC transmission optimization as described in claim 5, characterized in that: The output is the final optimized unit operation mode scheme that satisfies all safety constraints, and the spatiotemporal graph convolutional network model is updated based on the newly added optimized samples. The unit operation mode schemes that have passed safety and stability verification will be output as the final optimized schemes that can be used for scheduling and execution. Save the power grid time-series section data, the final optimization scheme and the verification results corresponding to the current optimization as new samples; When the number of new samples accumulates to a preset size, the spatiotemporal graph convolutional network model is retrained and continuously updated using the updated sample set containing the new samples.

7. The rapid optimization method for unit operation mode by integrating spatiotemporal graph convolutional networks and DC transmission optimization as described in claim 6, characterized in that: The power flow distribution of each line in the verification and validation system does not exceed the load capacity constraints, including unit output constraints, ramping constraints, and power balance constraints. The unit output constraint is expressed as: in, , These represent the lower and upper limits of the generator unit's output at time t, respectively, with the lower limit of the new energy unit's output being 0. Constrained by maximum capacity active power fluctuation faults and DC blocking faults, the lower limit of output for each synchronous generator unit participating in frequency regulation is set based on the current DC transmission situation, expressed as: in, , These represent the minimum output of different hydropower and thermal power synchronous generating units participating in frequency regulation at time t. and These refer to the number of hydropower and thermal power units participating in frequency regulation. The active power surplus is filled by the load participating in frequency regulation after the fault occurs at time t. The ramp constraint for the synchronous machine is set as follows: in, and These represent the upper and lower limits of the unit's operating ramp rate, respectively. , and These represent the unit outputs of the synchronous generator at time t, time t-1, and time 0, respectively. The system power balance constraint is expressed as follows: in, 、 、 These represent the generator active power output, load level, and line active power loss at time t, respectively. Let t be the DC transmission capacity of the system at time t. 、 、 These represent the number of generators, the number of loads, and the number of lines, respectively.

8. A rapid optimization system for unit operation modes integrating spatiotemporal graph convolutional networks and DC transmission optimization, comprising applying the rapid optimization method for unit operation modes integrating spatiotemporal graph convolutional networks and DC transmission optimization as described in any one of claims 1 to 7, characterized in that, include: The system includes a model building module, a preliminary solution output module, a verification and correction module, and a final solution output module. The model building module is used to build and train a spatiotemporal graph convolutional network model that integrates the physical characteristics and operating rules of the power system, and is used to predict the optimal operating mode of the units based on the time series operation data of the power grid. The preliminary output scheme module uses a spatiotemporal graph convolutional network model to predict the current power grid time series section and output a preliminary unit operation mode scheme. The verification and correction module uses the prediction results as the initial solution set and employs a DC transmission optimization model to quickly optimize, verify, and correct the prediction results; the DC transmission optimization model is used to verify the safety and stability of the corrected operation mode scheme. The output final scheme module outputs the final unit operation mode optimization scheme that satisfies all safety constraints, and updates the spatiotemporal graph convolutional network model based on the newly added optimization samples.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the rapid optimization method for unit operation mode that integrates spatiotemporal graph convolutional networks and DC transmission optimization as described in any one of claims 1 to 7.

Images