A method and system for evaluating the reliability of an AC / DC power system based on a deep neural network

Abstract

The application discloses a kind of based on deep neural network AC-DC power system operation reliability evaluation method and system, it is related to power system operation optimization and artificial intelligence technical field, including based on AC-DC power system composition characteristics, according to hierarchical progressive logic, constructs the multi-state reliability model of AC-DC power system;System state sampling is carried out using state enumeration method, the optimal power flow under different system states is calculated and the corresponding optimal load shedding amount is obtained, training data set is constructed based on input feature matrix;AC-DC power system optimal load shedding regression model based on deep residual network is constructed, and training data set is trained;The optimal load shedding of the state to be evaluated of target AC-DC power system is predicted using the trained deep residual network, and the system operation reliability index is calculated in combination with the prediction result.The method ensures the evaluation accuracy while improving the evaluation efficiency, and provides reliable decision support for AC-DC power system operation dispatching.

Description

Technical Field

[0001] This invention relates to the fields of power system operation optimization and artificial intelligence technology, specifically to a method and system for evaluating the operational reliability of AC / DC power systems based on deep neural networks. Background Technology

[0002] With the large-scale grid connection of new energy sources, AC / DC hybrid power systems are playing an important role in promoting the cross-regional distribution of new energy. However, the deep coupling between DC transmission systems and AC power systems has led to an increase in the size of system components and topological complexity, resulting in a surge in the number of anticipated fault scenarios.

[0003] Existing reliability assessment methods for AC / DC power systems, whether analytical or simulation-based (such as Monte Carlo simulation), heavily rely on repetitive calculations of massive amounts of fault scenarios. This results in extremely low assessment efficiency, making it difficult to meet the timeliness requirements of power system operation phases. In recent years, data-driven methods have been used to improve assessment efficiency; however, traditional machine learning algorithms, such as support vector machines, have simple model structures and limited feature learning capabilities, making it difficult to guarantee assessment accuracy when processing high-dimensional, highly nonlinear power system data.

[0004] Conventional deep learning methods, such as convolutional neural networks, are prone to gradient vanishing and network degradation problems when the number of network layers increases. This leads to the paradoxical phenomenon that "the deeper the network, the lower the accuracy," which limits its ability to learn complex failure modes.

[0005] Therefore, there is an urgent need for a new evaluation method that can significantly improve evaluation efficiency while ensuring high accuracy, so as to meet the online decision-making needs during the operational phase. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by this invention is that existing AC / DC power system reliability assessment methods have low assessment efficiency, simple algorithm model structure, limited learning ability, and the problem of how to improve assessment accuracy.

[0008] To address the aforementioned technical problems, this invention provides the following technical solution: a method for evaluating the operational reliability of AC / DC power systems based on deep neural networks, comprising: constructing a multi-state reliability model of the AC / DC power system according to the composition characteristics of the AC / DC power system and following a hierarchical progressive logic; sampling the system state using a state enumeration method, calculating the optimal power flow under different system states and obtaining the corresponding optimal load shedding amount, and constructing a training dataset based on the input feature matrix; constructing an optimal load shedding regression model of the AC / DC power system based on a deep residual network, and training the training dataset; using the trained deep residual network to predict the optimal load shedding for the target AC / DC power system under evaluation, and calculating the system operational reliability index based on the prediction results; wherein the deep residual network superimposes the module input and the convolutional layer output through jump connections of residual modules.

[0009] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the multi-state reliability model of the AC / DC power system includes: firstly, establishing a DC power equipment reliability model; secondly, establishing a flexible multi-terminal DC transmission system reliability model based on the DC transmission equipment reliability model; and finally, integrating the multi-state reliability models of the sending-end wind farm and the receiving-end AC power grid to establish a multi-state reliability model of the AC / DC system based on flexible transmission.

[0010] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the establishment of the DC power equipment reliability model includes: merging the state space based on the available transmission capacity of the DC transmission equipment, and modeling the DC transmission equipment as a three-state model, including a normal operation state with 100% available transmission capacity, a partial failure state with 50% available transmission capacity, and a complete failure state with 0% available transmission capacity.

[0011] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the step of calculating the optimal power flow and obtaining the corresponding optimal load shedding amount under different system states includes: enumerating system states by state enumeration method, solving the AC / DC system optimal power flow model for each state, taking the minimization of the system load shedding cost under each state as the objective function, and taking into account the calculation constraints of the optimal power flow.

[0012] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the calculation constraints of the optimal power flow include: DC system power balance constraints, AC / DC system common coupling node power balance constraints, AC system power balance constraints, generator power constraints, node load reduction constraints, node voltage constraints, VSC converter station transmission capacity constraints, and AC / DC power constraints.

[0013] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the input feature matrix includes node load level, wind farm available capacity, conventional unit available capacity, node self-admittance used to characterize line operation status, and DC transmission line available capacity.

[0014] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the deep residual network includes one or more residual modules. The residual modules adopt a skip connection structure, and short-circuit paths are set in parallel by bypass on the basis of stacking two convolutional layers to learn the residual mapping function, that is, the residual between the input features and the expected output.

[0015] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the optimal load shedding regression model includes an architecture based on a deep residual network, training which uses mean square error as the loss function, and fitting the load shedding amount by learning the mapping relationship between the system state and the optimal load shedding amount.

[0016] As a preferred embodiment of the AC / DC power system operation reliability assessment method based on deep neural networks described in this invention, the system operation reliability indicators include: expected power supply shortage and load shedding probability; the expected power supply shortage is expressed as:

[0017] ,

[0018] in, This represents the average power load loss. The total number of system states. This represents the total number of nodes in the system. In the state Down Time Node The shear load.

[0019] The probability of load failure is expressed as:

[0020] ,

[0021] in, The probability of power load loss in the system. In the state Down Total load shedding of the system at any given time.

[0022] Another objective of this invention is to provide a reliability assessment system for AC / DC power systems based on deep neural networks. This system can solve the problem of low assessment efficiency in current AC / DC power system reliability assessments by constructing an optimal load shedding regression model for AC / DC power systems based on deep residual networks and training it on a training dataset.

[0023] As a preferred embodiment of the AC / DC power system operation reliability assessment system based on deep neural networks described in this invention, the system includes: a reliability modeling module, a dataset construction module, a model training module, and an evaluation and prediction module. The reliability modeling module is used to establish multi-state reliability models for DC transmission equipment, flexible multi-terminal DC transmission systems, sending-end wind farms, and receiving-end AC power grids according to hierarchical progressive logic. The dataset construction module is used to enumerate system states using a state enumeration method, calculate the optimal power flow of the AC / DC system under each state to obtain the optimal load shedding amount, and construct an input feature matrix based on node load levels, available wind farm capacity, available capacity of conventional units, node self-admittance, and available capacity of DC transmission lines to form a training dataset. The model training module is used to construct an optimal load shedding regression model based on a deep residual network, learn residual mapping using residual modules and skip connection structures, and train the model using the training dataset to fit the mapping relationship between system states and the optimal load shedding amount. The evaluation and prediction module is used to predict the optimal load shedding amount for the target AC / DC power system under evaluation using the trained deep residual network model, and calculate the operation reliability index based on the prediction results.

[0024] The beneficial effects of this invention are:

[0025] The reliability assessment method for AC / DC power systems based on deep neural networks provided by this invention is based on the composition characteristics of AC / DC power systems. Following a hierarchical progression, it constructs a multi-state reliability model for the AC / DC power system, employing layered modeling to more precisely describe system states and improve the accuracy and applicability of the reliability model. It uses state enumeration to sample system states, calculates optimal power flow under different system states, and obtains the corresponding optimal load shedding amounts. A training dataset is constructed based on the input feature matrix, comprehensively covering the system's operating states and generating representative training data to provide high-quality input data for the model. An optimal load shedding regression model for AC / DC power systems based on a deep residual network is constructed and trained on the training dataset, mitigating gradient vanishing and network degradation problems in deep network training and improving the model's ability to fit complex nonlinear relationships. Using the trained deep residual network, the optimal load shedding is predicted for the target AC / DC power system's state to be assessed. The system's operational reliability index is calculated based on the prediction results. The trained model can quickly predict the optimal load shedding amount, avoiding repeated, time-consuming optimal power flow calculations, and significantly improving the efficiency and real-time performance of operational reliability assessment. This invention achieves better results in terms of model refinement, model training stability, and assessment efficiency and real-time performance. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0027] Figure 1 The above is an overall flowchart of a method for evaluating the operational reliability of AC / DC power systems based on deep neural networks, provided in Embodiment 1 of the present invention.

[0028] Figure 2 The state space diagram of DC transmission equipment is provided for a method for evaluating the operational reliability of AC / DC power systems based on deep neural networks, as shown in Embodiment 1 of the present invention.

[0029] Figure 3 The state space diagram of a flexible four-terminal DC transmission system for a method of evaluating the operational reliability of AC / DC power systems based on deep neural networks, provided in Embodiment 1 of the present invention.

[0030] Figure 4 The residual network jump connection diagram for a method for evaluating the operational reliability of AC / DC power systems based on deep neural networks provided in Embodiment 1 of the present invention. Detailed Implementation

[0031] 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.

[0032] Example 1, referring to Figures 1-4 As an embodiment of the present invention, a method for evaluating the operational reliability of AC / DC power systems based on deep neural networks is provided, comprising:

[0033] S1: Based on the composition characteristics of AC / DC power systems, a multi-state reliability model 100 for AC / DC power systems is constructed according to hierarchical progressive logic.

[0034] Specifically, following the hierarchical logic of "DC transmission equipment - DC transmission system - AC / DC power system," a multi-state reliability model 100 for the AC / DC power system is established. First, a multi-state reliability model 100 based on flexible multi-terminal DC transmission is established. The entire system consists of a sending-end wind farm, a flexible multi-terminal DC transmission system, and a receiving-end AC grid. Both the sending-end wind farm and the receiving-end AC grid adopt conventional multi-state reliability models 100 based on Markov processes. The flexible multi-terminal DC transmission system mainly consists of sending-end DC equipment and receiving-end DC equipment; a failure in any part will affect the operating state of the entire system.

[0035] A multi-state reliability model based on Markov processes is used to model a wind farm at the sending end, which consists of multiple wind turbines. Therefore, the power generation capacity of the wind farm is determined by the combined power generation capacity of each wind turbine. The power generation capacity of each wind turbine is affected by the coupled influence of turbine state and wind speed fluctuations, and typically exhibits multiple states. If we assume that all wind turbines in the wind farm are of the same model, the reliability model of the wind farm can be modeled as a multi-state model. Specifically, assuming the states... Downwind speed is Then the wind farm's power generation capacity at this time It can be represented as:

[0036] ,

[0037] in, For the power generation capacity of wind farms, For state The number of wind turbines operating normally within the wind farm. This refers to the serial number of the wind turbine unit. For wind turbines operating normally within a wind farm, at wind speeds The power generation capacity below.

[0038] If we assume that the fault state of the wind turbine at the current moment is independent of its previous fault states, then the transition relationships between the multiple states of the wind farm can be considered as a Markov process. Therefore, when the parameters of each unit in the wind farm are the same and the unit faults are independent of each other, the calculation process of the probability of each state of the wind farm at each moment can be expressed as follows:

[0039] ,

[0040] in, These represent different states within the multi-state structure of a wind farm. This represents the total number of states of the wind farm. for Moment State The state probability, for Moment State The state probability, for From the state of mind to state The transfer rate for From the state of mind to state The transfer rate.

[0041] Furthermore, the multi-state reliability model 100 of the AC / DC power system includes: firstly, establishing a DC power equipment reliability model M1; secondly, establishing a flexible multi-terminal DC transmission system reliability model M2 based on the DC transmission equipment reliability model; and finally, integrating the multi-state reliability models 100 of the sending-end wind farm and the receiving-end AC power grid to establish a multi-state reliability model 100 of the AC / DC system based on flexible transmission.

[0042] Before constructing a flexible multi-terminal DC transmission system, it is necessary to first establish a reliability model for DC transmission equipment. Establishing a reliability model for DC transmission equipment includes merging states in the state space based on the available transmission capacity of the DC transmission equipment, and modeling the DC transmission equipment as a three-state model, which includes a normal working state 101 with 100% available transmission capacity, a partial failure state 102 with 50% available transmission capacity, and a complete failure state 103 with 0% available transmission capacity.

[0043] A typical DC transmission system based on a VSC converter station consists of one subsystem A, two subsystems B, and two subsystems C connected in series and parallel. The reliability parameters of each subsystem, namely the repair rate and the failure rate, can be calculated based on the repair rate and failure rate of the power electronic components that make up each subsystem, combined with the series and parallel connection formula. This represents the failure rate and repair rate of subsystem A, and so on. Based on the topology of the DC transmission equipment and the reliability models of each subsystem, the state space of the DC transmission equipment can be further constructed, specifically as follows: Figure 2 As shown, by Figure 2 Analysis shows that the transition rate between states can be expressed as the failure rate or recovery rate of the corresponding subsystem. Furthermore, the state space of the DC transmission equipment contains many equivalent operating states, such as state 2 and state 4. These states have different failure types, but the transmission capacity of the DC transmission equipment in the corresponding states is 50%. Therefore, the state space can be merged based on the available transmission capacity, and the reliability model 100 of the DC transmission equipment can be modeled as a three-state model: a normal operating state 101 (SN) with 100% available transmission capacity, a partial failure state 102 (SP) with 50% available transmission capacity, and a complete failure state 103 (SF) with 0% available transmission capacity. After state merging, the state transition rate between the equivalent states can be expressed as:

[0044] ,

[0045] ,

[0046] in, for From the state of mind to state The transfer rate These are the two equivalent states after state merging. For the states within these two state sets, Let these be the sets of states corresponding to these two equivalent states. For state The state probability, For state The state probability, In the state The number of normally functioning components within the DC transmission equipment. The availability of normally functioning components, This represents the total number of components within the DC transmission equipment. The unavailability of the faulty component.

[0047] After calculating the state transition rates between the equivalent states of the DC transmission equipment, the system of differential equations can be solved based on a Markov process model to obtain the system state probability curves. This allows us to obtain the state probabilities corresponding to each equivalent state of the DC transmission equipment at each moment.

[0048] It should be noted that a reliability model M2 for a flexible multi-terminal DC transmission system can be established based on the DC transmission equipment reliability model 100. Taking a flexible four-terminal DC transmission system as an example, it consists of two sending-end DC transmission devices and two receiving-end DC transmission devices. Therefore, the state space of the flexible four-terminal DC transmission system can be specifically represented as follows: Figure 3 . This represents the maximum available transmission capacity of the DC transmission system, assuming that the maximum transmission capacity of each DC transmission device is the same.

[0049] Depend on Figure 3 Analysis shows that, similar to DC transmission equipment, flexible four-terminal DC transmission systems can also merge states in their state space based on the system's available transmission capacity. After merging, the reliability model of the flexible four-terminal DC transmission system can be represented as an eight-state model: 0, ... , , , , , , .in, and This is used to differentiate the reduction in available transmission capacity caused by faults in different receiving-end DC transmission equipment, and so on. Next, the state transition rates between the eight equivalent states can be solved using the aforementioned formula. After calculating the state transition rates between each equivalent state of the flexible four-terminal DC transmission system, a Markov process model can also be used to obtain the system state probability curves, thereby obtaining the state probabilities corresponding to each equivalent state of the flexible multi-terminal DC transmission system at each moment.

[0050] Therefore, the multi-state reliability model 100 integrating the sending-end wind farm and the receiving-end AC grid, and the multi-state reliability model 100 of the AC / DC system based on flexible power transmission have been established.

[0051] It should also be noted that by constructing a hierarchical and progressive multi-state Markov reliability model, wind farms, flexible multi-terminal DC transmission systems, and receiving-end AC power grids are integrated into a single model. This solves the problem of reliability assessment caused by the coupling of random faults and fluctuating power sources in AC / DC hybrid systems, and provides a refined description of the system's operating state, thereby improving the safety and economy of power grid planning and operation.

[0052] S2: The system state is sampled using the state enumeration method, the optimal power flow is calculated under different system states and the corresponding optimal load shedding is obtained, and a training dataset is constructed based on the input feature matrix.

[0053] Specifically, a training dataset is constructed based on the state enumeration method. Given a training dataset size N, N system states are enumerated, and the corresponding optimal power flow of the AC / DC system is calculated to obtain the optimal load shedding amount for different system states.

[0054] Calculating the optimal power flow and obtaining the corresponding optimal load shedding amount under different system states includes enumerating system states using the state enumeration method, solving the optimal power flow model for AC / DC systems for each state, taking the minimum system load shedding cost under each state as the objective function, and considering the calculation constraints of the optimal power flow.

[0055] The constraints for calculating optimal power flow include DC system power balance constraints, AC / DC system common coupling node power balance constraints, AC system power balance constraints, generator power constraints, node load reduction constraints, node voltage constraints, VSC converter station transmission capacity constraints, and AC / DC power constraints.

[0056] Furthermore, in different system states Under different conditions, the output level of the sending-end wind farm, the available transmission capacity of the DC transmission system, and the status of generators and lines in the AC system all vary. To assess the operational reliability of the AC / DC power system, it is necessary to calculate the optimal power flow under different conditions, analyze whether the system can guarantee reliable power supply under different conditions, and calculate reliability indicators based on this. Therefore, the optimal power flow model of the AC / DC power system for operational reliability assessment can be expressed as:

[0057] ,

[0058] ,

[0059] ,

[0060] ,

[0061] ,

[0062] ,

[0063] ,

[0064] ,

[0065] ,

[0066] ,

[0067] ,

[0068] in, For state Load shedding costs of AC / DC power systems This is the load shedding cost coefficient. This refers to the total number of nodes in the AC / DC power system, including both DC and AC system nodes. For state The load shedding amount at each node, Let be the transmission power on the converter station in state s. For state Lower DC node voltage amplitude, This represents the number of nodes in a DC system. For DC line conductance, For state Lower DC node voltage amplitude, For resistance, For reactance, To improve the DC voltage utilization rate of the converter station, The modulation ratio of the converter station. For state The voltage amplitude at the common coupling node of the AC / DC system. For state Control angle of the downstream converter station It is the impedance angle. The number of common coupling nodes, For state The power injected into the AC system from the converter station For state Next communication system node Injected active power, For the number of nodes in the communication system, For state Next communication system node voltage amplitude, For communication system nodes and The electrical conductance of the lines between them For state Phase angle difference between nodes in the AC system For communication system nodes and The susceptance of the line between them For generator sets Minimum allowable output For generator sets Actual effort For generator sets Maximum allowable output For generator sets, For total load demand, For nodes The lower limit of voltage amplitude, For nodes The actual voltage amplitude, For nodes The upper limit of voltage amplitude, For state The maximum transmission capacity of the VSC converter station For AC line power transmission capacity, This refers to the power transmission capacity of a DC line.

[0069] It should be noted that, in order to efficiently characterize the system state, the input feature matrix includes node load levels, available wind farm capacity, available conventional unit capacity, node self-admittance used to characterize the line operating state, and available DC transmission line capacity, which can be specifically expressed as:

[0070] ,

[0071] in, For the input feature matrix, Represents a node The load is used to characterize changes in the system's load level. Represents a node The corresponding available capacity of the wind farm is used to characterize the time-varying characteristics of wind power output. Represents a node The available capacity of conventional generating units is used to characterize the operating status information of conventional generating units. Represents a node The self-admittance is used to characterize the operating state of a transmission line. When the transmission line is in a fault failure state, this characteristic will change accordingly. Represents nodes The available capacity of connected DC transmission lines is used to characterize the operating status of the DC transmission system.

[0072] The output feature is the optimal load shedding amount corresponding to each system state. A training dataset is constructed based on the input and output features.

[0073] It should also be noted that by systematically sampling the operating states through the state enumeration method and combining it with AC / DC optimal power flow calculation to accurately obtain the optimal load shedding amount under each state, a feature dataset is constructed. This solves the huge computational burden caused by the complex power flow that involves traversing system states and solving them in the operation reliability assessment, and provides a key high-quality data foundation for training an efficient data-driven assessment model, thereby achieving fast and accurate online reliability assessment.

[0074] S3: Construct an optimal load shedding regression model for AC / DC power systems based on a deep residual network and train it on the training dataset.

[0075] Specifically, based on the constructed training dataset, an optimal load shedding regression model 200 for AC / DC systems based on a deep residual network is trained. The residual network is designed with a residual module structure on the basis of a traditional convolutional neural network, and then multiple residual modules are stacked to increase the network depth.

[0076] Deep residual networks include one or more residual modules. The residual modules adopt a skip connection structure, which sets short-circuit paths in parallel by bypass on the basis of stacking two convolutional layers, and learns the residual mapping function, that is, the residual between the input features and the expected output.

[0077] Furthermore, the residual module, based on the stacking of two convolutional layers, establishes short-circuit paths in parallel bypasses, enabling skip connections. This structure changes the learning objective of the residual network during training; it no longer focuses on directly learning the mapping from input to output, but instead learns the residual mapping function. input features and expected output The residuals between them can be specifically expressed as:

[0078] ,

[0079] in, For the desired output, The output is for the shortest path. This represents the last layer number of the residual module. This represents the initial number of layers for the residual module. For the first Input features of layer networks For the first The weight parameters of the layer network.

[0080] The output of the short-circuit path is typically superimposed on the output of the residual path via skip connections. Therefore, skip connections provide a direct path for gradients, allowing them to propagate directly from deeper layers to shallower layers, effectively mitigating the problems of gradient explosion and vanishing gradients. This allows the network to be trained at greater depths while maintaining training stability. The gradient propagation process in a residual network can be represented as:

[0081] ,

[0082] in, For loss function, These are the corresponding weight parameters.

[0083] The residual network skip connections assume that the network input consists of input features, and the expected output is a target mapping. Figure 4Analysis shows that traditional neural networks aim to directly learn this complex mapping. Residual networks, however, change the learning objective; they aim to learn a residual mapping function. That is, input and expected output The residual between them. The specific calculation process can be expressed as follows:

[0084] ,

[0085] in, The mapping learned by the stacked convolutional layers (typically two to three layers) inside the residual block. To connect the original input via skip connections Add directly to On the output.

[0086] Therefore, skip connections provide a direct path for gradients, allowing them to propagate directly from deeper layers to shallower layers, effectively mitigating the vanishing gradient problem during backpropagation. This enables the network to be trained at greater depths while maintaining training stability.

[0087] It should be noted that the optimal load shedding regression model 200 includes an architecture based on a deep residual network, which uses mean squared error as the loss function during training and fits the load shedding amount by learning the mapping relationship between the system state and the optimal load shedding amount.

[0088] The optimal load shedding model is a regression task; therefore, the mean squared error is used as the loss function, specifically expressed as follows:

[0089] ,

[0090] in, For the sample size, This represents the actual load shedding amount of the AC / DC power system. This represents the shear load applied to the residual network fitting.

[0091] It should also be noted that by introducing a deep residual network with a jump connection structure, the complex mapping relationship between system state characteristics and optimal load shedding is learned, solving the problem that traditional reliability assessment methods based on optimal power flow are computationally time-consuming and difficult to meet online application requirements. This provides an efficient and reliable tool for online rapid assessment and decision-making of AC / DC system operation risks.

[0092] S4: Using the trained deep residual network, perform optimal load shedding prediction for the target AC / DC power system under evaluation, and calculate the system operation reliability index based on the prediction results.

[0093] Specifically, based on the load shedding of the AC / DC power system under different states obtained by fitting the deep residual network, the system power supply reliability index can be further calculated, thereby completing the reliability assessment.

[0094] System reliability metrics include expected power shortage and probability of load failure; expected power shortage is expressed as:

[0095] ,

[0096] in, This represents the average power load loss. The total number of system states. This represents the total number of nodes in the system. In the state Down Time Node The shear load.

[0097] The probability of load failure is expressed as:

[0098] ,

[0099] in, The probability of power load loss in the system. In the state Down Total load shedding of the system at any given time.

[0100] It should be noted that, The smaller the value, the less power loss the system is expected to experience due to a fault, and the higher the system's operational reliability. This represents the probability of power load loss in the system; this indicator reflects the frequency with which power load losses occur in the system. Similarly, The smaller the value, the lower the probability of power load loss in the system, and the higher the reliability level of the system operation.

[0101] It should also be noted that by using a well-trained deep residual network model, the optimal load shedding quantity can be predicted quickly for the system's state to be evaluated. The key operational reliability indicators of expected power shortage and load loss probability can be calculated, solving the problem of low computational efficiency caused by repeated solutions in traditional reliability assessment. This provides efficient and accurate quantitative support for real-time safety analysis and decision-making in power systems.

[0102] Example 2, an embodiment of the present invention, provides a reliability assessment system for AC / DC power systems based on deep neural networks, including a reliability modeling module, a dataset construction module, a model training module, and an assessment and prediction module.

[0103] Among them, the reliability modeling module is used to establish a multi-state reliability model 100 for DC transmission equipment, flexible multi-terminal DC transmission system, sending-end wind farm and receiving-end AC power grid according to hierarchical progressive logic.

[0104] The dataset construction module is used to enumerate system states using the state enumeration method, calculate the optimal power flow of the AC / DC system under each state to obtain the optimal load shedding amount, and construct the input feature matrix based on the node load level, wind farm available capacity, conventional unit available capacity, node self-admittance, and DC transmission line available capacity to form the training dataset.

[0105] The model training module is used to build an optimal load shedding regression model 200 based on a deep residual network. It uses the residual module and the skip connection structure to learn the residual mapping and uses the training dataset to train the model to fit the mapping relationship between the system state and the optimal load shedding amount.

[0106] The evaluation and prediction module is used to predict the optimal load shedding amount for the target AC / DC power system under evaluation using a trained deep residual network model, and calculates the operational reliability index based on the prediction results.

[0107] This embodiment also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the AC / DC power system operation reliability assessment method based on deep neural networks proposed in the above embodiment.

[0108] This embodiment also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the AC / DC power system operation reliability assessment method based on deep neural networks as proposed in the above embodiments.

[0109] If the aforementioned functions are implemented as 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 this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing 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 (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0111] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0112] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0113] 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 method for evaluating the operational reliability of AC / DC power systems based on deep neural networks, characterized in that, include: Based on the composition characteristics of AC / DC power systems, a multi-state reliability model (100) of AC / DC power systems is constructed according to hierarchical progressive logic. The multi-state reliability model (100) of the AC / DC power system includes: firstly, establishing a DC equipment reliability model (M1); secondly, establishing a reliability model (M2) of the flexible multi-terminal DC transmission system based on the DC transmission equipment reliability model; and finally, integrating the multi-state reliability models of the sending-end wind farm and the receiving-end AC grid to establish a multi-state reliability model of the AC / DC system based on flexible transmission. The establishment of the DC power equipment reliability model (M1) includes merging the state space based on the available transmission capacity of the DC power transmission equipment, and modeling the DC power transmission equipment as a three-state model, including a normal working state (101) with 100% available transmission capacity, a partial failure state (102) with 50% available transmission capacity, and a complete failure state (103) with 0 available transmission capacity. The system state is sampled using the state enumeration method, the optimal power flow is calculated under different system states and the corresponding optimal load shedding is obtained, and a training dataset is constructed based on the input feature matrix. An optimal load shedding regression model for AC / DC power systems based on a deep residual network is constructed (200), and the model is trained on the training dataset. Using the trained deep residual network, the optimal load shedding prediction is performed on the state to be evaluated of the target AC / DC power system, and the system operation reliability index is calculated based on the prediction results. The deep residual network superimposes the module input and the convolutional layer output through the skip connection of the residual modules.

2. The method for assessing the operational reliability of AC / DC power systems based on deep neural networks as described in claim 1, characterized in that: The calculation of optimal power flow under different system states and the acquisition of corresponding optimal load shedding amounts include... The system states are enumerated using the state enumeration method. For each state, the optimal power flow model of the AC / DC system is solved. The objective function is to minimize the system load shedding cost under each state, while taking into account the calculation constraints of the optimal power flow.

3. The method for assessing the operational reliability of AC / DC power systems based on deep neural networks as described in claim 2, characterized in that: The calculation constraints for the optimal power flow include: DC system power balance constraints, AC / DC system common coupling node power balance constraints, AC system power balance constraints, generator power constraints, node load reduction constraints, node voltage constraints, VSC converter station transmission capacity constraints, and AC / DC power constraints.

4. The method for assessing the operational reliability of AC / DC power systems based on deep neural networks as described in any one of claims 1 to 3, characterized in that: The input feature matrix includes, Node load level, wind farm available capacity, conventional unit available capacity, node self-admittance used to characterize line operating status, and DC transmission line available capacity.

5. The method for assessing the operational reliability of AC / DC power systems based on deep neural networks as described in claim 4, characterized in that: The deep residual network includes, One or more residual modules, wherein the residual modules adopt a skip connection structure, and short-circuit paths are set in parallel by bypass on the basis of stacking two convolutional layers to learn the residual mapping function, that is, the residual between the input features and the expected output.

6. The method for assessing the operational reliability of AC / DC power systems based on deep neural networks as described in any one of claims 1 to 3 and 5, characterized in that: The optimal load shedding regression model (200) includes, Based on the architecture of deep residual networks, the training uses mean squared error as the loss function, and the load shedding amount is fitted by learning the mapping relationship between the system state and the optimal load shedding amount.

7. The method for assessing the operational reliability of AC / DC power systems based on deep neural networks as described in claim 6, characterized in that: The system operational reliability indicators include, Expected power shortage and probability of load failure; The expected power shortage is expressed as: ， in, This represents the average power load loss. The total number of system states. This represents the total number of nodes in the system. In the state Down Time Node The amount of load shedding; The probability of load failure is expressed as: ， in, The probability of power load loss in the system. In the state Down Total load shedding of the system at any given time.

8. A reliability assessment system for AC / DC power systems based on deep neural networks, employing the reliability assessment method for AC / DC power systems based on deep neural networks as described in any one of claims 1 to 7, characterized in that: It includes a reliability modeling module, a dataset construction module, a model training module, and an evaluation and prediction module; The reliability modeling module is used to establish a multi-state reliability model (100) for DC transmission equipment, flexible multi-terminal DC transmission system, sending-end wind farm and receiving-end AC power grid according to hierarchical progressive logic. The dataset construction module is used to enumerate system states using the state enumeration method, calculate the optimal power flow of the AC / DC system under each state to obtain the optimal load shedding amount, and construct an input feature matrix based on node load level, wind farm available capacity, conventional unit available capacity, node self-admittance and DC transmission line available capacity to form a training dataset. The model training module is used to construct an optimal load shedding regression model (200) based on a deep residual network. It uses the residual module and the skip connection structure to learn the residual mapping and uses the training dataset to train the model to fit the mapping relationship between the system state and the optimal load shedding amount. The evaluation and prediction module is used to predict the optimal load shedding amount for the target AC / DC power system under evaluation using a trained deep residual network model, and calculate the operational reliability index based on the prediction results.

Images