AI-based power system stability evaluation and control optimization system and method thereof

Abstract

The invention discloses an AI-based power system stability evaluation and control optimization system and method, and belongs to the field of artificial intelligence and power systems, and the system comprises a data processing module, a power grid dynamic protection module, a fast transient simulation module, a power angle stability and voltage stability analysis module, a power flow optimization calculation module and an information display module. The data processing module extracts power grid topology dynamic characteristics based on a space-time diagram convolutional neural network and maps the power grid topology dynamic characteristics to power angle stability and voltage stability indexes; the power grid dynamic protection module adjusts operation parameters in real time; the fast transient simulation module carries out fast transient simulation analysis through a deep belief network; the power angle stability and voltage stability analysis module generates a power generation load cutting scheme; the power flow optimization calculation module generates an optimization operation mode; according to the method, the limitation of traditional static topology analysis is broken through, the deep understanding of the dynamic characteristics of the power grid is realized, and the accuracy and the real-time performance of stability evaluation are improved.

Description

TECHNICAL FIELD

[0002] The present application relates to the field of artificial intelligence technology and power systems, in particular to an AI-based power system stability evaluation and control optimization system and method thereof, which is used for real-time monitoring, evaluation and optimization of transient stability of power systems, and improves the safe operation level of power grids. BACKGROUND

[0003] With the continuous expansion of the scale of power systems and the large-scale access of new energy, the structure and operation characteristics of power grids are becoming increasingly complex, and the problem of transient stability is becoming increasingly prominent. Traditional power system stability analysis methods mainly rely on physical models and numerical calculations, which have the disadvantages of high computational complexity, poor real-time performance, and difficulty in adapting to complex and variable power grid structures. Especially in the face of sudden failures and extreme working conditions, traditional methods are difficult to provide timely and accurate evaluation and effective control strategies.

[0004] In the prior art, the stability analysis method based on time domain simulation requires a large amount of computing resources and is difficult to meet the needs of real-time control; the method based on energy function method has a limited scope of application and is difficult to handle complex power grid structures; the method based on artificial intelligence has been applied, but is limited to single-dimensional stability evaluation, lacks in-depth understanding of power grid topology dynamic characteristics and comprehensive evaluation of multi-dimensional stability, and is difficult to provide targeted control strategies.

[0005] Therefore, there is an urgent need for a new system and method that can comprehensively utilize power grid multi-source data, deeply understand power grid topology dynamic characteristics, and realize multi-dimensional stability evaluation and optimization control. SUMMARY

[0006] The purpose of the present application is to provide an AI-based power system stability evaluation and control optimization system and method thereof, which realizes the extraction of power grid topology dynamic characteristics, real-time evaluation of multi-dimensional stability and generation of optimization control strategies by innovatively integrating spatio-temporal graph convolutional neural network and deep belief network technology, thereby improving the safe and stable operation level of power grids.

[0007] The present application provides an AI-based power system stability evaluation and control optimization system, which comprises:

[0008] A data processing module is used to receive power grid PMU synchronous phasor time series data and SCADA operation parameters, filter and intercept a data window of 0-5 seconds after power grid disturbance based on time alignment principle, denoise and preprocess the data window, extract power grid topology dynamic characteristics using spatio-temporal graph convolutional neural network and map to power angle stability and voltage stability indicators, and real-time calculate and quantify the power angle stability indicators and voltage stability indicators, and transmit the power angle stability indicators and voltage stability indicators to a power grid dynamic protection module.

[0009] A power grid dynamic protection module, in communication connection with the data processing module, is configured to receive the power angle stability index and the voltage stability index transmitted by the data processing module, to trigger the power grid node time sequence data flow calculated based on the power grid dynamic protection module in real time, to adjust the operation parameters in real time according to the power system operation state, and to transmit the power grid topology structure and parameter information to the fast transient simulation module;

[0010] The fast transient simulation module, in communication connection with the power grid dynamic protection module, is configured to receive the power grid topology structure and parameter information, to perform fast transient simulation analysis on the power grid through a deep belief network, to generate a power grid transient simulation result, and to transmit the power grid transient simulation result to the power angle stability and voltage stability analysis module;

[0011] The power angle stability and voltage stability analysis module, in communication connection with the fast transient simulation module, is configured to receive the power grid transient simulation result, to extract the voltage stability index and the time stability index, to analyze and evaluate the voltage stability index and the time stability index, to generate a cut-off power load scheme, and to transmit the cut-off power load scheme and the stability evaluation result to the power flow optimization calculation module and the power grid dynamic protection module;

[0012] The power flow optimization calculation module, in communication connection with the power angle stability and voltage stability analysis module, is configured to set power limits for power grid nodes, to generate an optimized operation mode based on the stability index of the power grid dynamic protection module and the current operation state of the power grid, and to transmit the optimized operation mode to the information display module;

[0013] The information display module, in communication connection with the power flow optimization calculation module and the power angle stability and voltage stability analysis module, is configured to receive and visually display the power grid topology structure, the stability index value, the voltage distribution after the cut-off power load, and the power flow distribution.

[0014] Preferably, the data processing module comprises:

[0015] A topology feature extraction unit is configured to select PMU synchronous phasor data of the power grid according to a time scale alignment principle, to average the data according to sampling points, to intercept the data interval from before the disturbance to 0-5 seconds after the disturbance, to calculate the dynamic connection between power grid nodes based on a spatio-temporal graph convolutional neural network, to generate power grid topology dynamic features, to map the power grid topology dynamic features into the power angle stability and time stability index, and to update the power grid topology dynamic features in real time;

[0016] A risk quantification unit, in communication connection with the topology feature extraction unit, is configured to receive the power grid topology dynamic features, to comprehensively consider the influence of power generation output disturbance and voltage fluctuation change on the stability of the power grid, to sum the voltage stability index and the time stability index calculated by the spatio-temporal graph convolutional neural network and the deep learning neural network according to weights, and to dynamically quantify the operation risk of the power grid.

[0017] As preferred, the spatio-temporal graph convolutional neural network adopted by the topological feature extraction unit comprises:

[0018] The network parameter initialization component is configured to form a graph network according to the spatial topological relationship of all nodes in the network, define the connection between nodes, and generate a spatial topological structure of the power grid;

[0019] The node feature extraction component is in communication connection with the network parameter initialization component, configured to extract the features of the changes in voltage stability indicators and time stability indicators between nodes in the spatial topological structure of the power grid, and generate node electrical characteristics;

[0020] The dynamic correlation analysis component is in communication connection with the node feature extraction component, configured to analyze the complex connection between nodes, extract network topological dynamic features, and generate dynamic connection features between nodes.

[0021] As preferred, the power grid dynamic protection module comprises:

[0022] The index receiving unit is configured to receive the power angle stability indicators and voltage stability indicators transmitted by the data processing module, perform data verification on the time series data stream, and generate verified index data;

[0023] The operation mode regulation unit is in communication connection with the index receiving unit, configured to receive the verified index data, change the generator output and reactive power compensation under the condition of a set threshold value, control the operation mode of the power grid, wherein the set threshold value is determined based on the power angle stability indicators and time stability indicators, and the active power limit value and the reactive power limit value are calculated according to the operation state of the power grid, and the optimal adjustment scheme of the power flow distribution to the active power and reactive power disturbance is calculated.

[0024] As preferred, the fast transient simulation module comprises:

[0025] The parameter input unit is configured to receive power grid operation mode data, set the active power limit value and the reactive power limit value, and input the power grid operation mode data and the power limit value into the deep belief network;

[0026] The simulation execution unit is in communication connection with the parameter input unit, configured to perform power grid model matching through the deep belief network based on the power grid operation mode data and the power limit value, and obtain power grid transient stability analysis results;

[0027] The stability quantification unit is in communication connection with the simulation execution unit, configured to calculate the voltage stability indicators and time stability indicators based on the power angle stability and voltage stability index definition formula, assign weight coefficients to the voltage stability indicators and time stability indicators according to the state of the power grid, and calculate a comprehensive power grid operation risk judgment index.

[0028] As preferred, the power angle stability and voltage stability analysis module comprises:

[0029] A fault prediction unit is configured to receive power grid topology and system stability indicators, construct a set of predicted faults based on power grid instability indicators calculated by the power grid operation risk calculation module, and assign probabilities to faults in the set of predicted faults;

[0030] A load shedding scheme generation unit is in communication with the fault prediction unit and is configured to calculate an optimal load shedding scheme based on a set power limit, taking into account the impact of active power distribution and reactive power distribution of the power grid on the load shedding scheme, wherein the optimal load shedding scheme includes the magnitude and location of the load to be shed;

[0031] A generator adjustment unit is in communication with the load shedding scheme generation unit and is configured to optimize generator output to increase voltage stability margin under the premise of ensuring power grid frequency stability through an optimization algorithm, and generate a generator output adjustment scheme;

[0032] A reactive power compensation control unit is in communication with the generator adjustment unit and is configured to minimize power grid voltage deviation and increase voltage stability margin through an optimization algorithm, wherein the optimization algorithm is based on frequency stability constraints and active power balance constraints to construct an optimization objective.

[0033] As preferred, the power flow optimization calculation module comprises:

[0034] A boundary constraint unit is configured to receive active power and reactive power data of power grid nodes, set active power limits and reactive power limits, and generate node power constraint conditions;

[0035] An optimization objective unit is in communication with the boundary constraint unit and is configured to construct a power grid operation optimization objective function based on stability indicators of the power grid dynamic protection module, including minimizing network loss, balancing load distribution, and improving voltage stability margin;

[0036] A solution calculation unit is in communication with the optimization objective unit and is configured to calculate the optimal distribution of power grid power flow based on the node power constraint conditions and the optimization objective function using a multi-objective optimization algorithm, and generate an optimized operation mode.

[0037] As preferred, the information display module comprises:

[0038] A topology display unit is configured to draw regional power grids from multiple perspectives based on power grid topology data, and realize visual display of power grid topology structure;

[0039] A fault playback unit is in communication with the topology display unit and is configured to display full-network fault recording based on time series view and geographic view, and support playback and query of historical faults;

[0040] An index display unit, in communication connection with the fault playback unit, is configured to display instability index changes based on a time sequence view, and realize dynamic visualization of stability indexes.

[0041] A control scheme display unit, in communication connection with the index display unit, is configured to display active and reactive power distribution based on a geographic view, give an optimal load shedding scheme, and display generator reactive power output changes.

[0042] A voltage early warning unit, in communication connection with the control scheme display unit, is configured to highlight and display node voltage early warning based on voltage stability calculation results.

[0043] Preferably, the system further comprises:

[0044] An offline training module is configured to use historical PMU synchronous phasor data and SCADA operation parameters to build a power grid historical operation scenario library, train a spatio-temporal graph convolutional neural network and a deep belief network model, and generate trained AI model parameters.

[0045] A model updating module, in communication connection with the offline training module, is configured to receive real-time operation data and control effect feedback, regularly update AI model parameters, and improve system adaptability and accuracy, wherein the model updating adopts an incremental learning method to retain effective experience knowledge.

[0046] The AI-based power system stability evaluation and control optimization method comprises the following steps:

[0047] PMU synchronous phasor time sequence data and SCADA operation parameters of the entire power grid are obtained, data preprocessing is performed, power grid topology dynamic characteristics are extracted based on a spatio-temporal graph convolutional neural network model, and power grid operation risks are evaluated in real time.

[0048] Based on the power grid topology dynamic characteristics, power grid time sequence node voltages, time, and fault parameters are calculated based on a power grid dynamic protection system, system operation modes are quickly adjusted by setting trigger thresholds, and power grid risks are secondarily prevented and controlled.

[0049] Based on power grid topology structure and parameters, power grid operation states are quickly simulated by a deep belief network model, voltage time stability indexes are calculated, and power angle and voltage instability trajectories are predicted.

[0050] Based on the power angle stability index and the voltage stability index, a load shedding scheme is given in advance before power grid instability, reactive power is optimally distributed, generator output is controlled, and power grid stability control under expected faults is realized.

[0051] The information display module realizes visual display of power grid topology structure, active power and reactive power distribution, dynamically displays power grid transient stability risk, and displays control results.

[0052] The application has the following beneficial effects:

[0053] 1. The power grid topology dynamic characteristics are extracted by the spatio-temporal graph convolutional neural network, the limitations of traditional static topology analysis are broken through, the depth understanding of the power grid dynamic characteristics is realized, and the accuracy of the stability evaluation is improved.

[0054] 2. The deep belief network is used for fast transient simulation, the simulation speed is greatly improved, the traditional calculation which may need minutes is shortened to seconds, and the needs of real-time control are met.

[0055] 3. A multi-dimensional evaluation system of power angle stability and voltage stability is established, the dynamic characteristics of the power grid topology are combined, the dynamic quantification of the power grid risk is realized, and comprehensive basis is provided for control decision.

[0056] 4. A complete closed-loop system from data processing to control execution is established, real-time monitoring of the power grid state, timely early warning of the risk and optimized generation of the control strategy are realized, and a multi-level power grid protection system is formed.

[0057] 5. The early warning time of the system is 50%-200% earlier than that of the traditional method, the false alarm rate is reduced from 15%-20% of the traditional method to less than 5%, the calculation efficiency is improved by 10-50 times, and the safe and stable operation level of the power grid is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 It is a whole structure schematic diagram of the application based on AI power system stability evaluation and control optimization system;

[0059] Figure 2 It is an internal structure schematic diagram of the data processing module of the application;

[0060] Figure 3 It is an internal structure schematic diagram of the power grid dynamic protection module of the application;

[0061] Figure 4 It is an internal structure schematic diagram of the fast transient simulation module of the application;

[0062] Figure 5 It is an internal structure schematic diagram of the power angle stability and voltage stability analysis module of the application;

[0063] Figure 6 It is an internal structure schematic diagram of the power flow optimization calculation module of the application;

[0064] Figure 7It is an internal structure schematic diagram of the information display module of the application;

[0065] Figure 8 It is a flow chart of the AI-based power system stability evaluation and control optimization method of the application. DETAILED DESCRIPTION

[0066] Please refer to Figures 1-8 , the application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the embodiments described herein are only for illustrating the application, not limiting the application.

[0067] As Figure 1 shown, the AI-based power system stability evaluation and control optimization system provided by the application includes a data processing module 1, a power grid dynamic protection module 2, a fast transient simulation module 3, a power angle stability and voltage stability analysis module 4, a power flow optimization calculation module 5 and an information display module 6.

[0068] As Figure 2 shown, the data processing module 1 is used for receiving power grid PMU synchronous phasor time series data and SCADA operation parameters, screening and intercepting a data window of 0-5 seconds after power grid disturbance based on time alignment principle, denoising and preprocessing the data window, extracting power grid topology dynamic characteristics using a spatio-temporal graph convolutional neural network and mapping to power angle stability and voltage stability indicators, real-time calculating and quantifying the power angle stability indicators and voltage stability indicators, and transmitting the power angle stability indicators and voltage stability indicators to the power grid dynamic protection module 2.

[0069] In a preferred embodiment of the application, the data processing module 1 includes a topology feature extraction unit 11 and a risk quantification unit 12.

[0070] The topology feature extraction unit 11 is used for selecting PMU synchronous phasor data of the power grid according to the time alignment principle, taking the average according to the sampling points, and then intercepting the data interval from before disturbance to 0-5 seconds later, calculating the dynamic connection between power grid nodes based on a spatio-temporal graph convolutional neural network, generating power grid topology dynamic characteristics, mapping into power angle stability and time stability indicators, and real-time updating the power grid topology dynamic characteristics.

[0071] Specifically, the spatio-temporal graph convolutional neural network used by the topology feature extraction unit 11 includes a network parameter initialization component 111, a node feature extraction component 112 and a dynamic correlation analysis component 113.

[0072] The network parameter initialization component 111 is used to construct a graph network of all network nodes according to spatial topological relationships, define the connections between nodes, and generate the power grid spatial topology. In practical applications, the power grid can be represented as a graph structure G=(V,E), where V represents the set of nodes (such as power plants, substations, etc.) and E represents the set of edges (such as transmission lines). The connections between nodes can be represented by an adjacency matrix A ∈ Rn×n, where n is the number of nodes. =1 indicates that there is a connection between node i and node j; otherwise... =0.

[0073] The node feature extraction component 112 is communicatively connected to the network parameter initialization component 111, and is used to extract features from the changes in voltage stability index and time stability index between nodes in the power grid spatial topology, generating node electrical features. In this invention, node features can be represented as a feature matrix X ∈ R^{n×d}, where n is the number of nodes and d is the feature dimension, including electrical quantities such as voltage amplitude, phase angle, active power, and reactive power.

[0074] The dynamic correlation analysis component 113 is communicatively connected to the node feature extraction component 112, and is used to analyze complex relationships between nodes, extract dynamic features of the network topology, and generate dynamic relationship features between nodes. Dynamic correlation analysis is based on spatiotemporal graph convolution operations, with the specific formula as follows:

[0075] ,

[0076] in: For the first The node feature matrix of the layer; For the first The node feature matrix of the layer, when hour, That is, the input feature matrix; To add self-connected adjacency matrices, This is the original adjacency matrix. for An identity matrix of order 1; for The degree matrix, whose diagonal elements , i.e., node The number of connections (including self-connections); For the first The weight matrix of a layer consists of parameters that need to be learned through training. Activation functions, such as ReLU or tanh functions, are used to introduce nonlinear characteristics.

[0077] Through multi-layer convolution operation, the spatio-temporal graph convolutional neural network can extract high-order connection relationship between nodes and capture dynamic change characteristics of power grid topology.

[0078] The risk quantification unit 12 is in communication connection with the topology feature extraction unit 11, used for receiving the power grid topology dynamic characteristics, comprehensively considering the influence of power output disturbance and voltage fluctuation change on power grid stability, comprehensively considering the voltage stability index and time stability index calculated by the spatio-temporal graph convolutional neural network and the deep learning neural network, and performing weighted summation, to dynamically quantify the power grid operation risk.

[0079] The power grid operation risk can be calculated by the following formula:

[0080] ,

[0081] Wherein: represents the power grid operation risk index, the value range is [0, 1], and the greater the value is, the higher the risk is; represents the voltage stability index, the value range is [0, 1]; represents the time stability index, the value range is [0, 1]; and are weight coefficients, satisfying , used for balancing the importance of voltage stability and time stability in risk assessment, in actual application, the weight coefficients can be set according to the characteristics of the power grid, and preferably , .

[0082] In the application, the voltage stability index can be calculated by weighted average of node voltage deviation:

[0083] ,

[0084] Wherein: is the voltage stability index; is the voltage amplitude of node i, and the unit is kilovolt (kV); is the reference voltage, usually taking the rated value, such as 220kV or 500kV; is the weight of node i, indicating the importance of the node in the system, satisfying ; is the total number of nodes in the system; represents taking absolute value operation.

[0085] The time stability index can be calculated by system damping ratio:

[0086] ,

[0087] Wherein: For time-stable indicators; Let be the damping ratio of the j-th mode of the system, and represent the rate at which the system oscillations decay. To determine the number of modes of interest, the dominant low-frequency oscillation modes in the system are typically selected. This is used as an averaging factor to ensure that the index value is within a reasonable range. Damping ratio The smaller the value, the slower the system oscillation decays, and the worse the stability.

[0088] like Figure 3 As shown, the power grid dynamic protection module 2 is communicatively connected to the data processing module 1. It is used to receive the power angle stability index and voltage stability index transmitted by the data processing module 1, trigger the power grid node time-series data stream calculated by the power grid dynamic protection module in real time, adjust the operating parameters in real time according to the power system operating status, and transmit the power grid topology and parameter information to the fast transient simulation module 3.

[0089] In a preferred embodiment of the present invention, the power grid dynamic protection module 2 includes an indicator receiving unit 21 and an operation mode control unit 22.

[0090] The indicator receiving unit 21 is used to receive the power angle stability indicator and voltage stability indicator transmitted by the data processing module 1, perform data verification on the time-series data stream, and generate verified indicator data. The data verification adopts the median filtering method to remove outliers and ensure data quality.

[0091] The operation mode control unit 22 is communicatively connected to the index receiving unit 21 and is used to receive the verified index data. Under the condition of a set threshold, the power grid operation mode is controlled by changing the generator output and reactive power compensation. The set threshold is determined based on the power angle stability index and the time stability index. The active power limit and reactive power limit are calculated according to the power grid operation status, and the optimal adjustment scheme of the power flow distribution on the active power and reactive power disturbances is calculated.

[0092] In this invention, the threshold values ​​can be determined based on power grid characteristics and safety requirements. Preferably, the power angle stability index threshold is set to 0.2, and the time stability index threshold is set to 0.15. When the index values ​​exceed the thresholds, control measures are triggered.

[0093] The formulas for calculating the active power limit and reactive power limit are as follows:

[0094] ,

[0095] ,

[0096] in: This indicates the active power limit, in megawatts (MW). P ref represents the reference active power, i.e. the set power in normal operating conditions; is the active regulation coefficient, representing the power reduction rate when the risk exceeds the threshold, preferably taking the value of 0.3; is the risk indicator, taking the value in the range of [0, 1]; is the risk threshold, representing the risk level triggering the power limitation, preferably taking the value of 0.25; represents the larger value between 0 and , ensuring that the power limitation is only performed when the risk exceeds the threshold; represents the reactive power limit, in units of mega var (Mvar); represents the reference reactive power; is the reactive regulation coefficient, preferably taking the value of 0.5.

[0097] As shown in Figure 4 , the fast transient simulation module 3 is in communication connection with the power grid dynamic protection module 2, for receiving the power grid topology and parameter information, performing fast transient simulation analysis on the power grid through a deep belief network, generating a power grid transient simulation result, and transmitting the power grid transient simulation result to the power angle stability and voltage stability analysis module 4.

[0098] In a preferred embodiment of the present application, the fast transient simulation module 3 includes a parameter input unit 31, a simulation execution unit 32, and a stability quantification unit 33.

[0099] The parameter input unit 31 is configured to receive power grid operating mode data, set active power limits and reactive power limits, and input the power grid operating mode data and power limits into a deep belief network. The power grid operating mode data includes node parameters (such as generator output, load power, node voltage, etc.) and branch parameters (such as line impedance, transformer parameters, etc.).

[0100] The simulation execution unit 32 is in communication connection with the parameter input unit 31, and is configured to perform power grid model matching through a deep belief network based on the power grid operating mode data and power limits, to obtain a power grid transient stability analysis result. The deep belief network (DBN) is a deep learning model stacked by multiple layers of restricted Boltzmann machines (RBMs). In the present application, the DBN is used to quickly simulate the transient process of the power grid, avoiding the complex calculation of traditional differential equation solving.

[0101] The pre-training process of the DBN uses the contrastive divergence algorithm, and the energy function is defined as:

[0102] ,

[0103] wherein: represents the energy function, and the lower the energy, the more stable the system; is the visible layer node state vector, represents the state of the i-th visible layer node, corresponding to the observable quantities in the power grid, such as node voltage, generator power, etc. is the hidden layer node state vector, represents the state of the j-th hidden layer node, used to capture the potential features in the data; is the connection weight, representing the connection strength between the visible layer node i and the hidden layer node j; and are the bias terms of the visible layer and the hidden layer, respectively; and are the number of nodes of the visible layer and the hidden layer, respectively.

[0104] The stability quantification unit 33 is in communication connection with the simulation execution unit 32, for calculating the voltage stability index and the time stability index based on the power angle stability and voltage stability index definition formula, and weighting the voltage stability index and the time stability index according to the state of the power grid, to calculate the comprehensive power grid operation risk judgment index.

[0105] The comprehensive power grid operation risk judgment index can be calculated by the following formula:

[0106] ,

[0107] wherein: represents the comprehensive risk index, with a value range of [0, 1], and the larger the value, the higher the risk; represents the voltage stability index, with a value range of [0, 1]; represents the time stability index, with a value range of [0, 1]; represents the frequency stability index, with a value range of [0, 1], and the calculation method is similar to that of the voltage stability index, based on the frequency deviation; 、 and are weight coefficients, satisfying , for balancing the importance of different types of stability in comprehensive evaluation, in actual application, the power grid characteristics and focus can be set, and preferably , , .

[0108] As shown in Figure 5 , the power angle stability and voltage stability analysis module 4 is in communication connection with the fast transient simulation module 3, for receiving the power grid transient simulation result, extracting the voltage stability index and the time stability index, analyzing and evaluating the voltage stability index and the time stability index, generating the cut-off power load scheme, and transmitting the cut-off power load scheme and the stability evaluation result to the power flow optimization calculation module 5 and the power grid dynamic protection module 2.

[0109] In a preferred embodiment of the present application, the power angle stability and voltage stability analysis module 4 comprises a fault prediction unit 41, a load shedding scheme generation unit 42, a generator adjustment unit 43, and a reactive power compensation control unit 44.

[0110] The fault prediction unit 41 is configured to receive the power grid topology and system stability indicators, and based on the power grid instability indicators calculated by the power grid operation risk calculation module, construct a set of expected faults, and assign probabilities to the faults in the set of expected faults. The set of expected faults can be represented as:

[0111] ,

[0112] wherein FS represents the set of expected faults, including all typical faults that can occur in the system; represents the i-th expected fault, including information such as fault location, type (such as short circuit, broken line, etc.), and duration; k represents the number of faults, which is determined according to the size of the system and the analysis requirements.

[0113] For each fault , a probability weight can be assigned according to historical statistical data and the current system state, satisfying . The probability weight reflects the likelihood of the fault occurring, and is used to optimize resource allocation and control strategies.

[0114] The load shedding scheme generation unit 42 is communicatively connected to the fault prediction unit 41, and is configured to calculate an optimal load shedding scheme based on the set power limit, considering the impact of active power distribution and reactive power distribution on the load shedding scheme, wherein the optimal load shedding scheme includes the magnitude and location of the load shedding.

[0115] The optimal load shedding scheme can be obtained by solving the following optimization problem:

[0116] ,

[0117] wherein: represents the load shedding distribution vector, , represents the amount of load shedding at node i, in units of megawatts (MW); f represents the optimization objective function, which can include the total amount of load shedding, the system stability margin, etc.; C represents the load shedding constraint conditions, such as the maximum allowed load shedding amount, critical load protection, etc.; represents the optimization increment of each node, used to adjust the load shedding distribution.

[0118] In practical applications, the sensitivity analysis method can be used to calculate the influence degree of each load node on the stability of the system, and the load with less influence on stability is preferentially cut off. The amount of cut-off load can be determined according to the degree of system instability, and is generally controlled within the range of 5%-15% of the total load.

[0119] The generator adjustment unit 43 is in communication connection with the load shedding scheme generation unit 42, and is configured to optimize the generator output by an optimization algorithm to increase the voltage stability margin under the premise of ensuring the stability of the power grid frequency, and generate a generator output adjustment scheme.

[0120] The generator output adjustment needs to meet the following constraint conditions:

[0121] ,

[0122] ,

[0123] Among them: P Gi represents the active power output of the generator, and the unit is megawatt (MW); and respectively represent the minimum and maximum output limits of the generator, which are determined by the unit characteristics; P Gi represents the active power output of the i-th generator; N G represents the number of generators; P Dj represents the active power of the j-th load node; N D represents the number of load nodes; P Loss represents the system loss, and the unit is megawatt (MW). The first constraint ensures that the generator operates within a safe range, and the second constraint ensures system power balance.

[0124] The reactive power compensation control unit 44 is in communication connection with the generator adjustment unit 43, and is configured to minimize the voltage deviation of the power grid by an optimization algorithm to increase the voltage stability margin, wherein the optimization algorithm is based on the frequency stability constraint and the active power balance constraint to build an optimization target.

[0125] The optimization target of the reactive power compensation control can be represented as:

[0126] ,

[0127] Among them: V i represents the voltage amplitude of node i, and the unit is kilovolt (kV); V ref represents the reference voltage, which is usually taken as the rated value, such as 220kV or 500kV; W i represents the weight of node i, reflecting the importance of the node in the system; N represents the number of nodes. The optimization target aims to minimize the weighted squared difference between the system node voltage and the reference value, and improve the voltage quality.

[0128] The constraints include:

[0129] ,

[0130] ,

[0131] wherein: PgenQ represents the generator reactive power output, in units of mega-volt-ampere (Mvar); PgenQmin and PgenQmax represent the minimum and maximum generator reactive power output limits, respectively, determined by the unit characteristics; Pcomp represents the compensation capacity of the reactive power compensation device, in units of mega-volt-ampere (Mvar); Pcompmin and Pcompmax represent the minimum and maximum capacity limits of the reactive power compensation device, respectively. These constraints ensure that the device operates within a safe range.

[0132] As shown in FIG. 1, the power flow optimization calculation module 5 is in communication connection with the power angle stability and voltage stability analysis module 4, for setting power limits for grid nodes, generating an optimized operation mode based on the stability indexes of the grid dynamic protection module 2 and the current operation state of the grid, and transmitting the optimized operation mode to the information display module 6. Figure 6 In a preferred embodiment of the present application, the power flow optimization calculation module 5 includes a boundary constraint unit 51, an optimization objective unit 52, and a solution calculation unit 53.

[0133] The boundary constraint unit 51 is configured to receive active power and reactive power data of grid nodes, set active power limits and reactive power limits, and generate node power constraints. The constraints are represented as:

[0134]

[0135] ,

[0136] , wherein:

[0137] PnodeP and PnodeQ represent the node active power and reactive power, respectively, in units of mega-watt (MW) and mega-volt-ampere (Mvar); PnodePmin and PnodePmax represent the lower and upper limits of the active power, respectively, determined by the device capacity and system safety requirements; PnodeQmin and PnodeQmax represent the lower and upper limits of the reactive power, respectively. These constraints ensure that the system operates within a safe range.

[0138] ​​​​​​The optimization target unit 52 is in communication connection with the boundary constraint unit 51, configured to construct a power grid operation optimization target function based on the stability index of the power grid dynamic protection module 2, including minimizing the grid loss, balancing the load distribution, and improving the voltage stability margin. The target function can be expressed as:

[0139] ,

[0140] wherein: represents the comprehensive optimization target, which needs to be minimized; represents the grid loss target function, reflecting the system loss; represents the load balancing target function, reflecting the load distribution uniformity; represents the stability margin target function, reflecting the system stability; , and are weight coefficients, satisfying , for balancing the importance of different targets in optimization, in actual application, preferably , , .

[0141] The grid loss target function can be expressed as:

[0142] ,

[0143] wherein: represents the grid loss target function, with the unit of megawatt (MW); represents the resistance of the line , with the unit of ohm ; represents the current of the line , with the unit of ampere (A); represents the line number. The function calculates the total active loss of all lines in the system.

[0144] The load balancing target function can be expressed as:

[0145] ,

[0146] wherein: represents the load balancing target function; represents the load power of the node , with the unit of megawatt (MW); represents the average load power, ; represents the load node number. The function measures the uniformity of the system load distribution, and the smaller the value, the more uniform the distribution.

[0147] The stability margin target function can be expressed as:

[0148] ,

[0149] wherein represents a stability margin target function; represents a comprehensive risk index, the value range is [0, 1]. Since the smaller the value is, the more stable the system is, and therefore the reciprocal of the value is taken as an optimization target, so that the target function is consistent with the minimum direction of other target functions.

[0150] The solving calculation unit 53 is in communication connection with the optimization target unit 52, and is configured to calculate the optimal distribution of power flow of the power grid based on the node power constraint condition and the optimization target function by using a multi-objective optimization algorithm, and generate an optimized operation mode. The multi-objective optimization algorithm can be an improved particle swarm optimization algorithm, a genetic algorithm or an interior point method, etc.

[0151] As shown in Figure 7 , the information display module 6 is in communication connection with the power flow optimization calculation module 5 and the power angle stability and voltage stability analysis module 4, and is configured to receive and visually display the power grid topology structure, the stability index value, the voltage distribution after the power load is cut off and the power flow distribution.

[0152] In a preferred embodiment of the present application, the information display module 6 comprises a topology display unit 61, a fault playback unit 62, an index display unit 63, a control scheme display unit 64 and a voltage early warning unit 65.

[0153] The topology display unit 61 is configured to draw the regional power grid from multiple perspectives based on the power grid topology data, and realize the visual display of the power grid topology structure. The power grid topology can be displayed in the form of a node-edge graph, wherein the node represents an electrical device (such as a generator, a transformer, a bus, etc.), and the edge represents a connection relationship (such as a power transmission line).

[0154] The fault playback unit 62 is in communication connection with the topology display unit 61, and is configured to display the full-network fault recording based on a time sequence view and a geographical view, and support the playback and query of historical faults. The time sequence view displays the change curve of electrical quantities over time, and the geographical view displays the fault position and the influence range.

[0155] The index display unit 63 is in communication connection with the fault playback unit 62, and is configured to display the instability index change based on a time sequence view, and realize the dynamic visualization of the stability index. The index display adopts a curve chart form, and simultaneously displays the index value and the threshold line, so as to facilitate the intuitive judgment of the system stability state.

[0156] The control scheme display unit 64 is in communication connection with the index display unit 63, and is used to display the active and reactive power distribution based on the geographical view, give the optimal load shedding scheme, and display the generator reactive power output change. The control scheme display adopts color coding, and different colors represent different control measures and adjustment amplitudes.

[0157] The voltage early warning unit 65 is in communication connection with the control scheme display unit 64, and is used to perform high-light early warning display on the node voltage based on the voltage stability calculation result. The voltage early warning adopts color classification display, for example, green represents normal (voltage deviation < 5%), yellow represents warning (voltage deviation 5%-10%), and red represents emergency (voltage deviation > 10%).

[0158] In addition, the system of the present application can further include an offline training module and a model updating module (not shown in the figure). The offline training module is used to utilize historical PMU synchronous phasor data and SCADA operation parameters to construct a power grid historical operation scenario library, train a spatio-temporal graph convolutional neural network and a deep belief network model, and generate AI model parameters after training. The model updating module is in communication connection with the offline training module, and is used to receive real-time operation data and control effect feedback, regularly update the AI model parameters, and improve the adaptability and accuracy of the system, wherein the model updating adopts an incremental learning manner to retain effective experience knowledge.

[0159] As shown in Figure 8 The present application also provides an AI-based power system stability evaluation and control optimization method, which includes the following steps:

[0160] Step S1: Obtain the whole network PMU synchronous phasor time series data and SCADA operation parameters, perform data preprocessing, extract the power grid topology dynamic characteristics based on the spatio-temporal graph convolutional neural network model, and evaluate the power grid operation risk in real time.

[0161] Specifically, the data preprocessing includes data cleaning, data alignment and data normalization. The data cleaning is used to remove noise and outliers, and methods such as median filtering or wavelet transform can be used; the data alignment is based on the time alignment principle to ensure the time consistency of different source data; and the data normalization converts the data of different physical quantities to a unified numerical range, which is convenient for subsequent processing.

[0162] The power grid topology dynamic characteristic extraction and operation risk evaluation adopt the same technical scheme as the data processing module in Embodiment One, which will not be described here again.

[0163] Step S2: Based on the power grid topology dynamic characteristics, calculate the power grid time sequence node voltage, time and fault parameters based on the power grid dynamic protection system, quickly adjust the system operation mode by setting a trigger threshold, and perform secondary prevention and control on the power grid risk.

[0164] In this step, the system dynamically adjusts the operating parameters according to the dynamic characteristics and stability indicators of the power grid topology, and realizes real-time prevention and control of the power grid risk. When the stability indicator exceeds the preset threshold, the system triggers the corresponding control measures, such as adjusting the generator output, switching the reactive power compensation device, etc., to maintain the stable operation of the system.

[0165] Step S3: Based on the power grid topology and parameters, the power grid operating state is quickly simulated through a deep belief network model, the voltage time stability indicator is calculated, and the power angle and voltage instability trajectory are predicted.

[0166] The fast transient simulation adopts the same technical solution as the fast transient simulation module in Embodiment One, and replaces the traditional differential equation solution with a deep belief network, which greatly improves the simulation speed and provides time margin for subsequent control decisions.

[0167] Step S4: Based on the power angle stability indicator and the voltage stability indicator, a load shedding scheme is given in advance before the power grid instability, and the generator output is adjusted through reactive power optimization and control, to realize the power grid stability control under the expected fault.

[0168] In this step, the system generates a preventive control strategy based on the stability evaluation results, including load shedding scheme, generator output adjustment and reactive power compensation control. These control measures can be deployed in advance before the actual fault of the power grid, significantly improving the anti-disturbance ability of the system.

[0169] Step S5: The information display module is used to realize the visual display of the power grid topology, active and reactive power distribution, dynamically display the transient stability risk of the power grid and display the control results.

[0170] The information visualization adopts the same technical solution as the information display module in Embodiment One, and through multi-angle and multi-dimensional visual display, helps the operator to intuitively understand the system state and control effect, and supports decision optimization.

[0171] The method of the present application realizes the full-process automation from data acquisition, feature extraction, stability evaluation to control strategy generation, significantly improving the accuracy, real-time performance and effectiveness of power grid stability evaluation and control.

[0172] The above description is only the preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An AI-based power system stability assessment and control optimization system, characterized in that, The method comprises the following steps: The data processing module is used for receiving power grid PMU synchronous phasor time series data and SCADA operation parameters, screening and intercepting a data window of 0-5 seconds after a power grid disturbance based on a time scale alignment principle, denoising and preprocessing the data window, extracting power grid topology dynamic characteristics by using a space-time graph convolutional neural network and mapping the power grid topology dynamic characteristics to power angle stability and voltage stability indexes, and calculating and quantifying the power angle stability indexes and voltage stability indexes in real time, and transmitting the power angle stability indexes and voltage stability indexes to the power grid dynamic protection module; The power grid dynamic protection module is in communication connection with the data processing module, and is used for receiving the power angle stability indexes and voltage stability indexes transmitted by the data processing module, triggering a power grid node time series data stream calculated based on the power grid dynamic protection module in real time, adjusting operation parameters according to a power system operation state in real time, and transmitting power grid topology structure and parameter information to the fast transient simulation module; The fast transient simulation module is in communication connection with the power grid dynamic protection module, and is used for receiving the power grid topology structure and parameter information, performing fast transient simulation analysis on the power grid by using a deep belief network, and generating power grid transient simulation results, and transmitting the power grid transient simulation results to the power angle stability and voltage stability analysis module; The power angle stability and voltage stability analysis module is in communication connection with the fast transient simulation module, and is used for receiving the power grid transient simulation results, extracting voltage stability indexes and time stability indexes, analyzing and evaluating the voltage stability indexes and time stability indexes, generating a cut-off power load scheme, and transmitting the cut-off power load scheme and stability evaluation results to the power flow optimization calculation module and the power grid dynamic protection module; The power flow optimization calculation module is in communication connection with the power angle stability and voltage stability analysis module, and is used for setting power limits for power grid nodes, generating an optimized operation mode based on stability indexes of the power grid dynamic protection module and a current operation state of the power grid, and transmitting the optimized operation mode to the information display module; The information display module is in communication connection with the power flow optimization calculation module and the power angle stability and voltage stability analysis module, and is used for receiving and visually displaying power grid topology structure, stability index values, voltage distribution after cutting off power load, and power flow distribution. 2.The AI-based power system stability evaluation and control optimization system according to claim 1, characterized in that, The data processing module comprises: The topology feature extraction unit is used for selecting PMU synchronous phasor data of the power grid according to the time scale alignment principle, intercepting a data interval from before the disturbance to 0-5 seconds after the disturbance after averaging according to sampling points, calculating dynamic connections between power grid nodes based on a space-time graph convolutional neural network, generating power grid topology dynamic characteristics, mapping the power grid topology dynamic characteristics into power angle stability and time stability indexes, and updating the power grid topology dynamic characteristics in real time; The risk quantification unit is in communication connection with the topology feature extraction unit, and is used for receiving the power grid topology dynamic characteristics, comprehensively considering influences of power generation output disturbance and voltage fluctuation changes on power grid stability, summing up voltage stability indexes and time stability indexes calculated by the space-time graph convolutional neural network and the deep learning neural network according to weights, and dynamically quantifying power grid operation risks. 3.The AI-based power system stability evaluation and control optimization system according to claim 2, characterized in that, The space-time graph convolutional neural network used by the topology feature extraction unit comprises: The network parameter initialization component is configured to form a graph network by arranging all nodes of the network according to a spatial topological relationship, define the connection between the nodes, and generate a spatial topological structure of the power grid; The node feature extraction component is in communication connection with the network parameter initialization component and is configured to extract features of changes in voltage stability indicators and changes in time stability indicators between the nodes in the spatial topological structure of the power grid, and generate electrical features of the nodes; The dynamic correlation analysis component is in communication connection with the node feature extraction component and is configured to analyze complex connections between the nodes, extract dynamic features of the network topology, and generate dynamic connection features between the nodes. 4.The AI-based power system stability evaluation and control optimization system according to claim 1, wherein, The power grid dynamic protection module includes: The index receiving unit is configured to receive the power angle stability indicators and the voltage stability indicators transmitted by the data processing module, perform data verification on the time series data stream, and generate verified index data; The operation mode regulation unit is in communication connection with the index receiving unit and is configured to receive the verified index data, change the generator output and the reactive power compensation under a set threshold condition, and control the operation mode of the power grid, wherein the set threshold is determined based on the power angle stability indicators and the time stability indicators, and the active power limit value and the reactive power limit value are calculated according to the operation state of the power grid, and an optimal adjustment scheme of the power flow distribution to the active power and the reactive power disturbance is calculated. 5.The AI-based power system stability evaluation and control optimization system according to claim 1, wherein, The fast transient simulation module includes: The parameter input unit is configured to receive the power grid operation mode data, set the active power limit value and the reactive power limit value, and input the power grid operation mode data and the power limit value into the deep belief network; The simulation execution unit is in communication connection with the parameter input unit and is configured to perform power grid model matching through the deep belief network based on the power grid operation mode data and the power limit value, and obtain a power grid transient stability analysis result; The stability quantification unit is in communication connection with the simulation execution unit and is configured to calculate the voltage stability indicators and the time stability indicators based on a power angle stability and voltage stability indicator definition formula, assign weight coefficients to the voltage stability indicators and the time stability indicators according to the state of the power grid, and calculate a comprehensive power grid operation risk judgment indicator. 6.The AI-based power system stability evaluation and control optimization system according to claim 1, wherein, The power angle stability and voltage stability analysis module includes: The fault prediction unit is configured to receive the power grid topological structure and the system stability indicators, calculate the power grid instability indicators based on the power grid operation risk calculation module, construct an expected fault set, and assign probabilities to faults in the expected fault set; The load shedding scheme generation unit is in communication connection with the fault prediction unit and is configured to consider the influence of the active power distribution and the reactive power distribution of the power grid on the optimal load shedding scheme, calculate the optimal load shedding scheme based on the set power limit value, wherein the optimal load shedding scheme includes the magnitude and position of the load to be cut off; The generator adjustment unit is in communication connection with the load shedding scheme generation unit and is configured to optimize the generator output under the premise of ensuring the stability of the power grid frequency by using an optimization algorithm, increase the voltage stability margin, and generate a generator output adjustment scheme. The reactive power compensation control unit is in communication connection with the generator adjustment unit, is used for minimizing the grid voltage deviation through an optimization algorithm, increasing the voltage stability margin, wherein the optimization algorithm is based on the frequency stability constraint and the active power balance constraint to build an optimization target. 7.The AI-based power system stability evaluation and control optimization system according to claim 1, wherein, The power flow optimization calculation module comprises: The boundary constraint unit is used for receiving active power and reactive power data of grid nodes, setting active power limit and reactive power limit, and generating node power constraint conditions; The optimization target unit is in communication connection with the boundary constraint unit, is used for building a grid operation optimization objective function based on the stability index of the grid dynamic protection module, including minimizing the network loss, balancing the load distribution and improving the voltage stability margin; The solving calculation unit is in communication connection with the optimization target unit, is used for calculating the optimal distribution of grid power flow based on the node power constraint conditions and the optimization target function, and generating an optimized operation mode by using a multi-objective optimization algorithm. 8.The AI-based power system stability evaluation and control optimization system according to claim 1, wherein, The information display module comprises: The topology display unit is used for multi-view drawing of regional grid based on grid topology data, realizing visual display of grid topology structure; The fault playback unit is in communication connection with the topology display unit, is used for displaying full-network fault recording based on time sequence view and geographic view, supporting playback and query of historical faults; The index display unit is in communication connection with the fault playback unit, is used for displaying instability index changes based on time sequence view, realizing dynamic visualization of stability index; The control scheme display unit is in communication connection with the index display unit, is used for displaying active and reactive power distribution based on geographic view, giving an optimal cut-off power load scheme, and displaying generator reactive power output changes; The voltage early warning unit is in communication connection with the control scheme display unit, is used for highlighting and warning display of node voltage based on voltage stability calculation results. 9.The AI-based power system stability evaluation and control optimization system according to claim 1, wherein, The system further comprises: The offline training module is used for building a grid historical operation scenario library by using historical PMU synchronous phasor data and SCADA operation parameters, training a spatio-temporal graph convolutional neural network and a deep belief network model, and generating trained AI model parameters; The model updating module is in communication connection with the offline training module, is used for receiving real-time operation data and control effect feedback, regularly updating AI model parameters, and improving system adaptability and accuracy, wherein the model updating adopts an incremental learning mode to retain effective experience knowledge.

10. An AI-based power system stability assessment and control optimization method, characterized in that, The method comprises the following steps: Obtain the PMU synchronous phasor time sequence data and the SCADA operation parameters, perform data preprocessing, extract grid topology dynamic characteristics based on a spatio-temporal graph convolutional neural network model, and evaluate grid operation risk in real time; In combination with the grid topology dynamic characteristics, calculate grid time sequence node voltage, time and fault parameters based on a grid dynamic protection system, quickly adjust system operation mode by setting a trigger threshold, and perform secondary prevention and control on grid risk; Based on grid topology structure and parameters, quickly simulate grid operation state by a deep belief network model, calculate voltage time stability index, predict power angle and voltage instability trajectory; Based on the aforementioned power angle stability index and voltage stability index, a load shedding scheme is provided in advance before grid instability occurs, and grid stability control under anticipated fault conditions is achieved through reactive power optimization allocation and generator output control. The information display module enables the visualization of the power grid topology and active and reactive power distribution, dynamically displays the power grid transient stability risks, and shows the control results.

Images