Power system transient stability prevention and emergency coordination method

Abstract

The invention relates to the technical field of power system automation, in particular to a transient stability prevention and emergency coordination method for a power system. Data collection and real-time monitoring: collecting real-time data of the power system; transient stability evaluation: selecting features related to transient stability from the collected data, training a deep belief network DBN, and evaluating the transient stability of the power system by using the trained deep belief network DBN; risk identification and prevention control strategy generation: constructing a risk assessment function to determine the risk level of the power grid based on the transient stability assessment result, and generating a prevention control strategy according to the risk level; performing strategy evaluation and optimization, and constructing a dominant function evaluation and optimization control strategy; and the constructed dominant function is used as a target function, and an optimal control strategy is searched by using an NSGA-II algorithm. According to the invention, through advanced data analysis and optimization technologies, the transient stability of the power system is efficiently evaluated, and an economical and effective control strategy is generated.

Description

Technical Field

[0001] The invention relates to the technical field of power system automation, and in particular to a method for preventing transient stability and emergency coordination of a power system. Background Art

[0002] Transient stability assessment of power systems is a key link in ensuring the safe operation of power grids. Existing technologies usually use model-driven time-domain simulation methods to simulate the dynamic behavior of power systems after faults occur. By simulating the physical processes of the system in detail, in-depth analysis of system stability can be provided. However, with the continuous expansion of the scale and complexity of power systems, these traditional methods have gradually shown limitations in terms of real-time performance and computational efficiency.

[0003] High control cost, separated preventive control and emergency control strategies lead to higher control costs in actual operation. Due to the lack of effective integration and coordination mechanisms, these strategies cannot achieve a cost-effective balance between preventive measures and emergency responses, thereby increasing the economic burden on grid operators. Insufficient system stability, existing control strategies often do not comprehensively consider stability and economy, which may lead to excessive sacrifice of economy in the pursuit of system stability, or reduce system stability while considering economy. This unbalanced strategy may not be able to effectively maintain the stability of the power system under certain circumstances. Insufficient control strategy generation and optimization, the existing technology lacks a systematic method to generate and optimize control strategies, especially when multiple economic and stability indicators need to be considered simultaneously. This results in the control strategy may not be able to accurately adapt to the actual operating status of the power system, reducing the effectiveness and adaptability of the strategy.

[0004] In view of the limitations of existing technologies, a new method is urgently needed to improve the efficiency and accuracy of power system transient stability assessment, while reducing control costs and improving system stability and economy. Summary of the invention

[0005] In view of the above deficiencies in the prior art, the purpose of the present invention is: in order to overcome the limitations of the prior art, the present invention proposes a method for preventing and coordinating transient stability of power systems, aiming to achieve efficient evaluation of transient stability of power systems through advanced data analysis and optimization techniques, and generate economical and effective control strategies.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solutions:

[0007] A method for preventing and coordinating transient stability of a power system, comprising the following steps:

[0008] S1. Data collection and real-time monitoring, collect real-time data of the power system, including node voltage, line current, generator output, etc.;

[0009] S2. Transient stability assessment: Select features related to transient stability from the data collected in S1, train a deep belief network (DBN), and use the trained deep belief network (DBN) to assess the transient stability of the power system;

[0010] S3. Risk identification and prevention and control strategy generation: Based on the results of transient stability assessment in S2, a risk assessment function is constructed to determine the risk level of the power grid, and a prevention and control strategy is generated according to the risk level;

[0011] S4. Strategy evaluation and optimization, construct advantage functions to evaluate and optimize the control strategies in S3;

[0012] S5. Multi-objective optimization, using the advantage function constructed in S4 as the objective function, and finding the optimal control strategy by using the NSGA-II algorithm.

[0013] Furthermore, in S2, the method for evaluating the transient stability of a power system using a deep belief network (DBN) includes:

[0014] S21. Feature selection: Select features related to transient stability from the data collected in S1. These features mainly include: line current, system frequency, generator rotor angular velocity, line load fault type, load fluctuation and reactive power, etc.

[0015] S22. Construct a deep belief network (DBN). Use the selected features related to transient stability to construct a deep belief network (DBN). The deep belief network (DBN) is composed of multiple layers of restricted Boltzmann machines (RBMs) stacked together. Each RBM can be regarded as a bipartite graph, which contains a visible layer and a hidden layer.

[0016] A single RBM defines the energy function between the visible layer and the hidden layer, as well as the probability distribution based on this energy function. Given the visible layer input v and the hidden state h, the probability distribution is defined as follows:

[0017] The probability of the visible layer given the hidden layer:

[0018] The probability of the hidden layer given the visible layer:

[0019] Among them, σ(x) is the Sigmoid function, b i is the bias of the ith node in the visible layer; c j is the bias of the jth node in the hidden layer; Wij and W ji is the weight connecting the visible layer node i and the hidden layer node j; v is the input data of the visible layer, v i is the input vector of the i-th node in the visible layer; h is the hidden state, h i is the activation state of node i in the hidden layer;

[0020] The input v of the visible layer refers to the node vector of the visible layer, specifically the selected features related to transient stability;

[0021] The hidden state h refers to the node vector of the hidden layer in the RBM, which represents the potential features or patterns learned from the input data v of the visible layer;

[0022] DBN is composed of multiple layers of RBM stacked together. Each layer of RBM is built on the hidden layer output of the previous layer of RBM. The mathematical model of the k-th layer RBM is expressed as:

[0023]

[0024] h (k) is the hidden state of the kth layer; v (k-1) is the visible state of the k-1th layer, that is, the hidden layer output of the kth RBM; c (k) is the bias vector of the kth layer, is the weight matrix of the kth layer; σ(x) is the Sigmoid function,

[0025] S23. Train DBN. Train the deep belief network (DBN) obtained in S22. Train each layer of RBM of the deep belief network (DBN) independently using the contrast divergence algorithm.

[0026] Furthermore, in S23, the method of using the contrastive divergence algorithm to train each layer of RBM of the deep belief network (DBN) includes:

[0027] S231. Randomly initialize weight W, visible layer bias b, and hidden layer bias c;

[0028] S232. The input v of the visible layer is fed into the RBM, and the Sigmoid function is used to calculate the weights and biases of each node h in the hidden layer based on the weights and biases initialized in S231. j The probability of being activated is:

[0029] P(h ij =1 / v)=σ(c j +∑ i W ji v i );

[0030] S233. According to the hidden layer state h obtained by forward propagation, back propagation is performed to reconstruct the visible layer v. The reconstructed visible layer is called "negative" data, which is the output of the model in the current state. The formula is expressed as:

[0031] P(v i / h)=σ(b j +∑ i W ij h j );

[0032] S234. Use the generated "negative" data to perform forward propagation again to update the probability distribution of the hidden layer, which is shown as:

[0033]

[0034] is the visible layer data reconstructed by the model based on the hidden state h; is the visible layer data at a given reconstruction The probability of hidden layer node j being activated under

[0035] S235. Calculate the gradient of weights and biases based on the difference between the forward propagation and the reconstructed hidden layer state. The formula is:

[0036] ΔW=η( <v i h j > data - <v i h j > recon );

[0037] Δb i =η( <v i > data - <v i > recon );

[0038] Δc j =η( <h j > data - <h j > recon );

[0039] Where η is the learning rate, which controls the amplitude of parameter update; data Represents the statistical expectation under the input v of the actual visible layer, which is estimated by multiple sampling; <> recpn Represents the visible layer input in the reconstruction The statistical expectation under , which is estimated by multiple sampling;

[0040] S236. Repeat S232 to S235 until the performance of the model on the training data is no longer significantly improved, or a predetermined number of iterations is reached, each iteration is intended to reduce the difference between the data and the model reconstruction, thereby improving the model's ability to represent the data.

[0041] Furthermore, in S3, the risk assessment function is constructed as follows:

[0042] R=ω1·S v1 +ω2·S f1 +…+ω k ·S nk ;

[0043] Among them, S v1 , S f1 , …, S nk are the standardized scores of various stability indicators after DBN evaluation; ω1, ω2, ω k It is the weight corresponding to each stability index, which is determined according to their influence on system stability.

[0044] Furthermore, in S3, the step of determining the risk level of the power grid according to the risk assessment function includes:

[0045] Set the risk level threshold, and set the low risk threshold T l and medium risk threshold T m To classify the risk level, T l and T m It is set based on historical events, system characteristics and acceptable risk level. l and T m It is dynamically adjusted and updated according to system operation experience and external environment changes; the risk level is expressed as:

[0046] Low risk: R < T l ;

[0047] Medium risk: T l <R<T m ;

[0048] High risk: R ≥ T m ;

[0049] Generate a preventive control strategy and develop preventive measures based on risk levels:

[0050] Low risk, monitor system status, maintain current operations, and prepare for potential risks;

[0051] Medium risk: optimize power generation and load planning, adjust grid operation mode, and reduce risks;

[0052] High risk, take emergency measures, such as shedding load, adjusting generator output, and activating backup equipment;

[0053] Preventive control strategies are regularly evaluated and updated to adapt to changes in grid operating conditions. Preventive control strategy evaluation is performed through simulation, historical data analysis, or real-time monitoring data.

[0054] Furthermore, in S4, the method for constructing the advantage function includes:

[0055] S411. Determine the impact indicators and determine the expected output or effect of each preventive control strategy based on the strategy generated in S3;

[0056] S412. Quantify the impact of strategies. For each preventive control strategy, quantify its impact on stability and economic indicators;

[0057] S413. Construct an advantage function, which combines stability and economic indicators. The advantage function is expressed as:

[0058] A=α·R t +(1-α)·R e ;

[0059] R e =λ1·e1+λ2·e2+…;

[0060] α is the weight for weighing the importance of system stability and economy. The selection of weight α should be based on the judgment of the importance of system stability and economy, and needs to be adjusted according to different situations. α is a weight between 0 and 1, which is used to balance the stability and economy indicators; R t is the output of the risk assessment function R; R e Comprehensive economic score; i is the weight of the i-th economic indicator, reflecting the importance of this indicator in the overall economic evaluation, e i is the cost-effectiveness score of the ith economic indicator, e i It is a monetary unit or other quantitative form, such as a percentage of cost savings.

[0061] Furthermore, methods for evaluating and optimizing control strategies using the constructed advantage function include:

[0062] S421. Quantify the stability index of the advantage function, extract the index related to stability from the constructed advantage function, and quantify the stability index; the stability index includes voltage stability and frequency stability, and use quantitative methods to measure these indexes;

[0063] S422. Design preventive control strategies based on quantitative stability indicators. The preventive control strategies include:

[0064] Adjustment of power generation plan: adjusting the output of generator sets according to system load and voltage stability indicators;

[0065] Load distribution optimization, redistributing grid load to reduce stress points and improve overall stability;

[0066] Activation of backup resources: activating backup generators or adjusting the operation of existing resources when necessary;

[0067] For each control strategy, calculate the change in the advantage function after applying the strategy:

[0068] ΔA=A(C)-A;

[0069] A(C)=α·R t (C)+(1-α)·R e ;

[0070] A(C) is the system advantage score after applying the control strategy; R t (C) is the output of the risk assessment function R after applying the control strategy, α is the weight that weighs the importance of system stability and economy, and R e Comprehensive economic rating;

[0071] The effect of the control strategy is evaluated according to the value of ΔA. If ΔA>0, it means that the control strategy improves the advantage score of the system, that is, it improves the stability and economy of the system.

[0072] Further, the method of finding the optimal control strategy using the NSGA-II algorithm includes:

[0073] S51. Initialize the population and randomly generate a set of potential control strategies as the initial population P0. For each solution P in the population, use S4 to construct the advantage function and calculate each advantage function A(P). A(P) is expressed as:

[0074] A(P)=α·R t (P)+(1-α)·R e ;

[0075] R t (P) is the stability score of solution P, R e is the economic score, α is the weight that weighs the importance of system stability and economy;

[0076] S52. Non-dominated sorting, for each pair of solutions P in the population i and P j , if P i No better than P on all targets j worse, and better in at least one objective, then P i Non-dominated Pj ; All solutions that do not dominate other solutions are divided into a level, usually the first level; then the solutions that are not dominated by them are removed from these solutions, and the remaining solutions constitute the second level, and so on;

[0077] S53. For each non-dominated level S i The strategy in , calculates the crowding distance D(P) to measure the distribution density of solutions within their non-dominated levels;

[0078]

[0079] A max is the maximum value of the optimization function in the entire population; A min is the minimum value of the optimization function in the entire population; A max,s is the non-dominated level S of the solution P i The maximum value of the optimization function in A min,s is the non-dominated level S of the solution P i Minimum value of the optimization function in ;

[0080] S54. Select excellent strategies based on non-dominated sorting and crowding distance to form a new generation of population;

[0081] Select solutions with high non-dominated levels and large crowding distances as parent solutions to form a new generation of population;

[0082] Select solutions with high non-dominated levels, and within the same level, select solutions with large crowding distance to maintain population diversity;

[0083] S55. Perform crossover and mutation operations on the selected strategy to generate a new offspring strategy;

[0084] Randomly select a parent solution, convert each parameter of the parent solution into a binary string, select one or more crossover points, the crossover points are random, exchange the gene fragments of the parent solution at the crossover points, generate a new offspring solution, and the crossover operation is one of single-point crossover, multi-point crossover or uniform crossover;

[0085] Perform mutation operations on the offspring solutions generated by crossover to introduce new genetic mutations, and collect the offspring solutions generated by crossover and mutation to form a new offspring population;

[0086] S56. Repeat steps S51 to S54, each iteration including generating a new generation of population, calculating crowding distance, selecting excellent strategies, performing crossover and mutation operations, until the quality of the solution reaches the predetermined standard.

[0087] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are:

[0088] By training DBN, it is possible to learn and predict the transient stability of the power system from a large amount of real-time data. Compared with traditional methods, DBN can capture more complex nonlinear relationships and use deep belief networks to evaluate the transient stability of the power system, thereby improving the accuracy and efficiency of the evaluation.

[0089] By setting risk level thresholds, the grid operation risks can be managed more finely, preventive measures at different levels from low risk to high risk can be implemented, a risk assessment function can be constructed, the risk level of the grid can be dynamically determined, and corresponding preventive control strategies can be generated according to the risk level.

[0090] The construction of the advantage function enables the control strategy to consider not only the stability of the system but also the economy. By weighing different indicators, a more comprehensive strategy optimization is achieved. Through the advantage function, the stability and economy of the power system are comprehensively considered, providing quantitative indicators for the evaluation and optimization of the control strategy.

[0091] Through non-dominated sorting and crowding distance calculation, excellent strategies are selected to form a new generation of population, and new offspring strategies are generated through crossover and mutation operations, ultimately achieving the quality of the solution of the predetermined standard. The NSGA-II algorithm is used for multi-objective optimization to find the optimal control strategy, balancing the stability and economy of the system.

[0092] Through the integrated approach, possible fault situations can be taken into account in the prevention stage and corresponding coordination strategies can be formulated, which improves the grid's ability to respond to emergencies, integrates transient stability prevention and fault coordination assistance, and achieves more comprehensive grid safety management.

[0093] By quantifying stability indicators (such as voltage deviation, frequency deviation) and economic indicators (such as cost savings percentage), the control strategy can be accurately evaluated and optimized, the quantification of stability and economic indicators is achieved, and a clear quantitative basis is provided for the evaluation of control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 A flow chart of the power system transient stability prevention and emergency coordination method;

[0095] Figure 2 Flowchart of the method for evaluating power system transient stability using deep belief network (DBN);

[0096] Figure 3 A flow chart of the method for constructing the advantage function;

[0097] Figure 4 Flowchart of the method for evaluating and optimizing control strategies using the constructed advantage function;

[0098] Figure 5Flowchart of the method for finding the optimal control strategy using the NSGA-II algorithm. DETAILED DESCRIPTION

[0099] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in combination with the embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and are not used to limit the invention.

[0100] like Figure 1 As shown, the power system transient stability prevention and emergency coordination method includes the following steps:

[0101] S1. Data collection and real-time monitoring: PMU (phasor measurement unit) and SCADA (supervisory control and data acquisition) systems are used for real-time monitoring to collect real-time data of the power system, including node voltage, line current, generator output, etc.

[0102] S2. Transient stability assessment: Select features related to transient stability from the data collected in S1, train a deep belief network (DBN), and use the trained deep belief network (DBN) to assess the transient stability of the power system;

[0103] S3. Risk identification and prevention and control strategy generation: Based on the results of transient stability assessment in S2, a risk assessment function is constructed to determine the risk level of the power grid, and a prevention and control strategy is generated according to the risk level;

[0104] S4. Strategy evaluation and optimization, construct advantage functions to evaluate and optimize the control strategies in S3;

[0105] S5. Multi-objective optimization, using the advantage function constructed in S4 as the objective function, and finding the optimal control strategy by using the NSGA-II algorithm.

[0106] like Figure 2 As shown, in S2, the method of evaluating the transient stability of the power system using a deep belief network (DBN) includes:

[0107] S21. Feature selection: Select features related to transient stability from the data collected in S1. Features related to transient stability mainly include: line current, system frequency, generator rotor angular velocity, line load fault type, load fluctuation and reactive power, etc.;

[0108] S22. Construct a deep belief network (DBN). Use the selected features related to transient stability to construct a deep belief network (DBN). The deep belief network (DBN) is composed of multiple layers of restricted Boltzmann machines (RBMs) stacked together. Each RBM can be regarded as a bipartite graph, including a visible layer and a hidden layer.

[0109] The mathematical model of a single RBM defines the energy function between the visible layer and the hidden layer, as well as the probability distribution based on this energy function. For a given visible layer input v and hidden state h, the probability distribution is defined as follows:

[0110] The probability of the visible layer v given the hidden layer h is:

[0111] The probability of hidden layer h given visible layer v is:

[0112] Among them, σ(x) is the Sigmoid function, b i is the bias of the ith node in the visible layer; c j is the bias of the jth node in the hidden layer; W ij and W ji is the weight connecting the visible layer node i and the hidden layer node j; v is the input data of the visible layer, v i is the input vector of the i-th node in the visible layer; h is the hidden state, h i is the activation state of node i in the hidden layer;

[0113] The input v of the visible layer refers to the node vector of the visible layer, which is the selected feature related to transient stability;

[0114] The hidden state h refers to the node vector of the hidden layer in the RBM, which represents the potential features or patterns learned from the input data v of the visible layer;

[0115] DBN is composed of multiple layers of RBM stacked together. Each layer of RBM is built on the hidden layer output of the previous layer of RBM. The mathematical model of the k-th layer RBM is expressed as:

[0116]

[0117] h (k) is the hidden state of the kth layer; v (k-1) is the visible state of the k-1th layer, that is, the hidden layer output of the kth RBM; c (k) is the bias vector of the kth layer, is the weight matrix of the kth layer; σ(x) is the Sigmoid function,

[0118] S23. Train DBN. Train the deep belief network (DBN) obtained in S22. Train each layer of RBM of the deep belief network (DBN) independently using the contrastive divergence algorithm. During the training process, the trained DBN is treated as a whole and trained end-to-end to optimize the parameters of the entire network.

[0119] In S23, the method of independently training each layer of RBM of a deep belief network (DBN) using a contrastive divergence algorithm includes:

[0120] S231. Randomly initialize weight W, visible layer bias b, and hidden layer bias c;

[0121] S232. The input v of the visible layer is fed into the RBM, and the Sigmoid function is used to calculate the weights and biases of each node h in the hidden layer based on the weights and biases initialized in S231. j The probability of being activated is:

[0122] P(h ij =1 / v)=σ(c j +Σ i W ji v i );

[0123] S233. According to the hidden layer state h obtained by forward propagation, back propagation is performed to reconstruct the visible layer v. The reconstructed visible layer is called "negative" data, which is the output of the model in the current state. The formula is expressed as:

[0124] P(v i / h)=σ(b j +∑ i W ij h j );

[0125] S234. Use the generated “negative” data to perform forward propagation again to update the probability distribution of the hidden layer. The formula is:

[0126]

[0127] is the visible layer data reconstructed by the model based on the hidden state h; is the visible layer data at a given reconstruction The probability of hidden layer node j being activated under

[0128] S235. Calculate the gradient of weights and biases. The parameter update is based on the difference between the forward propagation and the reconstructed hidden layer state to calculate the gradient. The calculation formula is:

[0129] ΔW=η( <v i h j > data - <v i h j > recon );

[0130] Δb i =η( <vi > data - <v i > recon );

[0131] Δc j =η( <h j > data - <h j > recon );

[0132] Among them, η is the learning rate, which controls the amplitude of parameter update; <> data Represents the statistical expectation under the input v of the actual visible layer, which is estimated by multiple sampling; <> recon Represents the visible layer input in the reconstruction The statistical expectation under , which is estimated by multiple sampling;

[0133] S236. Repeat S232 to S235 until the performance of the model on the training data is no longer significantly improved, or a predetermined number of iterations is reached, each iteration is intended to reduce the difference between the data and the model reconstruction, thereby improving the model's ability to represent the data.

[0134] S3. Risk identification and prevention and control strategy generation: Based on the results of transient stability assessment in S2, a risk assessment function is constructed to determine the risk level of the power grid, and a prevention and control strategy is generated accordingly;

[0135] The risk assessment function is expressed as: R = ω1·S v1 +ω2·S f1 +…+ω k ·S nk ;

[0136] Among them, S v1 , S f1 , …, S nk are the standardized scores of various stability indicators after DBN evaluation; ω1, ω2, ω k It is the weight corresponding to each stability index, which is determined according to their influence on system stability.

[0137] In S3, the step of determining the risk level of the power grid according to the risk assessment function includes:

[0138] Set the risk level threshold, and set the low risk threshold T l and medium risk threshold T m To classify the risk level, T l and T m It is set based on historical events, system characteristics and acceptable risk level. l and T mIt is dynamically adjusted and updated according to system operation experience and external environment changes; the risk level is expressed as:

[0139] Low risk: R < T l ;

[0140] Medium risk: T l <R<T m ;

[0141] High risk: R ≥ T m ;

[0142] Develop preventive measures based on risk level:

[0143] Low risk, monitor system status, maintain current operations, and prepare for potential risks;

[0144] Medium risk: optimize power generation and load planning, adjust grid operation mode, and reduce risks;

[0145] High risk, take emergency measures, such as shedding load, adjusting generator output, and activating backup equipment;

[0146] S4. Strategy evaluation and optimization, construct advantage functions to evaluate and optimize the control strategies in S3;

[0147] like Figure 3 As shown, the method for constructing the advantage function includes:

[0148] S411. Determine the impact indicators, and determine the expected output or effect of each preventive control strategy based on the strategy generated in S3, such as reducing the probability of voltage collapse, reducing frequency deviation, reducing power generation costs, etc.;

[0149] S412. Quantify the impact of strategies. For each preventive control strategy, quantify its impact on stability and economic indicators; for example, if a strategy reduces the probability of voltage collapse, the strategy will positively affect the stability indicator;

[0150] Stability indicators are key indicators that can reflect the stability of the power system, such as voltage stability, frequency stability, generator load distribution, etc.; converting stability indicators into quantifiable forms, including that voltage stability can be quantified by the percentage of voltage deviation; frequency stability can be quantified by system frequency deviation or frequency change rate;

[0151] Economic indicators are indicators that reflect the economic efficiency of the power system, such as power generation cost, operation cost, maintenance cost, etc. The quantification of economic indicators includes converting economic indicators into monetary units or other quantitative forms, such as cost saving percentage;

[0152] S413. Construct an advantage function, which combines stability and economic indicators. The advantage function is expressed as:

[0153] A=α·R t +(1-α)·R e ;

[0154] R e =λ1·e1+λ2·e2+…;

[0155] α is the weight for weighing the importance of system stability and economy. The selection of weight α should be based on the judgment of the importance of system stability and economy, and needs to be adjusted according to different situations. α is a weight between 0 and 1, which is used to balance the stability and economy indicators; R t is the output of the risk assessment function R; R e Comprehensive economic score; i is the weight of the i-th economic indicator, reflecting the importance of this indicator in the overall economic evaluation, e i is the cost-effectiveness score of the ith economic indicator, expressed in monetary units or other quantitative form, such as a percentage of cost savings.

[0156] like Figure 4 As shown in Figure 2, the methods for evaluating and optimizing control strategies using the constructed advantage function include:

[0157] S421. Quantify the stability index of the advantage function, extract the index related to stability from the constructed advantage function, quantify the stability index, the index related to stability includes voltage stability and frequency stability, and use quantitative methods to measure these indexes, for example, voltage stability can be quantified by the percentage of voltage deviation, and frequency stability can be quantified by system frequency deviation or frequency change rate;

[0158] S422. Design preventive control strategies based on quantitative stability indicators. The preventive control strategies include:

[0159] Adjustment of power generation plan: adjusting the output of generator sets according to system load and voltage stability indicators;

[0160] Load distribution optimization, redistributing grid load to reduce stress points and improve overall stability;

[0161] Activation of backup resources: activating backup generators or adjusting the operation of existing resources when necessary;

[0162] For each control strategy, calculate the change in the advantage function after applying the strategy:

[0163] ΔA=A(C)-A;

[0164] A(C)=α·Rt (C)+(1-α)·R e ;

[0165] A(C) is the system advantage score after applying the control strategy; R t (C) is the output of the risk assessment function R after applying the control strategy, α is the weight that weighs the importance of system stability and economy, and R e Comprehensive economic rating;

[0166] The effect of the control strategy is evaluated according to the value of ΔA. If ΔA>0, it means that the control strategy improves the advantage score of the system, that is, it improves the stability and economy of the system.

[0167] S5. Multi-objective optimization, using the advantage function constructed in S4 as the objective function, and finding the optimal control strategy by using the NSGA-II algorithm;

[0168] like Figure 5 As shown in Figure 2, the method of using the NSGA-II algorithm to find the optimal control strategy includes:

[0169] S51. Initialize the population and randomly generate a set of potential control strategies as the initial population P0. For each solution P in the population, use the advantage function constructed in S4 to calculate each advantage function A(P), expressed as:

[0170] A(P)=α·R t (P)+(1-α)·R e ;

[0171] R t (P) is the stability score of solution P, R e is the economic score, α is the weight that weighs the importance of system stability and economy;

[0172] S52. Non-dominated sorting, for each pair of solutions P in the population i and P j , if P i No better than P on all targets j worse, and better in at least one objective, then P i Non-dominated P j ; All solutions that do not dominate other solutions are divided into a level, usually the first level; then the solutions that are not dominated by them are removed from these solutions, and the remaining solutions constitute the second level, and so on;

[0173] S53. For each non-dominated level S i The strategy in , calculates the crowding distance D(P) to measure the distribution density of solutions within their non-dominated levels;

[0174]

[0175] A max is the maximum value of the optimization function in the entire population; A min is the minimum value of the optimization function in the entire population; A max,s is the non-dominated level S of the solution P i The maximum value of the optimization function in A min,s is the non-dominated level S of the solution P i Minimum value of the optimization function in ;

[0176] S54. Select excellent strategies based on non-dominated sorting and crowding distance to form a new generation of population;

[0177] Select solutions with high non-dominated levels and large crowding distances as parent solutions;

[0178] Select solutions with high non-dominated ranks because they perform better on the objective function; within the same rank, select solutions with large crowding distance to maintain population diversity;

[0179] S55. Perform crossover and mutation operations on the selected strategy to generate a new offspring strategy;

[0180] Randomly select a parent solution, convert each parameter of the parent solution into a binary string, select one or more crossover points, which can be random, and exchange the gene fragments of the parent solution at the crossover point to generate a new offspring solution. The crossover operation can be single-point crossover, multi-point crossover or uniform crossover, etc.;

[0181] Perform mutation operations on the offspring solutions generated by crossover to introduce new genetic mutations. The mutations can be random perturbations, inversions, exchanges, etc. The mutation operation is usually performed with a low probability to avoid destroying the beneficial characteristics generated by crossover.

[0182] Collect the offspring solutions generated by crossover and mutation to form a new offspring population;

[0183] S56. Repeat steps S51 to S54. Each iteration includes generating a new generation of population, calculating crowding distance, selecting excellent strategies, and performing crossover and mutation operations until the quality of the solution reaches the predetermined standard. Finally, the algorithm outputs a set of multiple excellent solutions that achieve the best trade-off between multiple objectives.

[0184] The above description is a preferred embodiment of the invention and is not intended to limit the invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the invention should be included in the protection scope of the invention.

Claims

1. A method for preventing transient stability and emergency coordination of a power system, characterized in that: include: S1. Data collection and real-time monitoring, collect real-time data of the power system; S2. Transient stability assessment: Select features related to transient stability from the data collected in S1, train a deep belief network DBN, and use the trained deep belief network DBN to assess the transient stability of the power system; S3. Risk identification and prevention and control strategy generation: Based on the results of transient stability assessment in S2, a risk assessment function is constructed to determine the risk level of the power grid, and a prevention and control strategy is generated according to the risk level; S4. Strategy evaluation and optimization, construct advantage function to evaluate and optimize the control strategy in S3; S5. Multi-objective optimization, using the advantage function constructed in S4 as the objective function, and finding the optimal control strategy by using the NSGA-II algorithm.

2. The method for preventing and coordinating transient stability of a power system according to claim 1, characterized in that: In S2, the method for evaluating the transient stability of a power system using a deep belief network DBN includes: S21. Feature selection, selecting features related to transient stability from the data collected in S1, the features related to transient stability include: line current, system frequency, generator rotor angular velocity, line load fault type, load fluctuation and reactive power; S22. Construct a deep belief network DBN, using the selected features related to transient stability to construct the deep belief network DBN, the deep belief network DBN is stacked by multiple levels of restricted Boltzmann machines RBM, each RBM is regarded as a bipartite graph, and the bipartite graph includes a visible layer and a hidden layer; A single RBM defines the energy function between the visible layer and the hidden layer, as well as the probability distribution based on this energy function. Given the visible layer input v and the hidden state h, the probability distribution is defined as: The probability of the visible layer given the hidden layer: The probability of the hidden layer given the visible layer: Among them, σ(x) is the Sigmoid function, bi is the bias of the ith node in the visible layer; c j is the bias of the jth node in the hidden layer; W ij and W ji is the weight connecting the visible layer node i and the hidden layer node j; v is the input data of the visible layer, v i is the input vector of the i-th node in the visible layer; h is the hidden state, h i is the activation state of node i in the hidden layer; The input v of the visible layer refers to the node vector of the visible layer, specifically the selected features related to transient stability; The hidden state h refers to the node vector of the hidden layer in the RBM, which represents the potential features or patterns learned from the input data v of the visible layer; DBN is composed of multiple layers of RBM stacked together. Each layer of RBM is built on the hidden layer output of the previous layer of RBM. The mathematical model of the k-th layer RBM is expressed as: h (k) is the hidden state of the kth layer; v (k-1) is the visible state of the k-1th layer, that is, the hidden layer output of the kth RBM; c (k) is the bias vector of the kth layer, is the weight matrix of the kth layer; σ(x) is the Sigmoid function, S23. Train DBN. Train the deep belief network DBN obtained in S22. Train each layer of RBM of the deep belief network DBN independently using the contrast divergence algorithm.

3. The method for preventing and coordinating transient stability of a power system according to claim 2, characterized in that: In S23, the method of using the contrast divergence algorithm to train each layer of RBM of the deep belief network DBN includes: S231. Randomly initialize weight W, visible layer bias b, and hidden layer bias c; S232. The input v of the visible layer is fed into the RBM, and the Sigmoid function is used to calculate the weights and biases of each node h in the hidden layer based on the weights and biases initialized in S231. j The probability of being activated is: P(h ij =1 / v)=σ(c j +∑ i W ji v i ); S233. According to the hidden layer state h obtained by forward propagation, back propagation is performed to reconstruct the visible layer v. The reconstructed visible layer is called "negative" data, which is the output of the model in the current state. The formula is expressed as: P(v i / h)=σ(b j +∑ i W ij h j ); S234. Use the generated "negative" data to perform forward propagation again to update the probability distribution of the hidden layer, which is shown as: is the visible layer data reconstructed by the model based on the hidden state h; is the visible layer data at a given reconstruction The probability of hidden layer node j being activated under S235. Calculate the gradient of weights and biases based on the difference between the forward propagation and the reconstructed hidden layer state. The formula is: ΔW=η( <v i h j > dat a- <v i h j > recon ); Δb i =η( <v i > data - <v i > recon ); Δc j =η( <h j > data - <h j > recon ); Among them, η is the learning rate, which controls the amplitude of parameter update; <> data Represents the statistical expectation under the input v of the actual visible layer, which is estimated by multiple sampling; <> recon Represents the reconstructed visible layer input The statistical expectation under , which is estimated by multiple sampling; S236. Repeat S232 to S235 until the performance of the model on the training data is no longer significantly improved, or the predetermined number of iterations is reached. Each iteration aims to reduce the difference between the data and the model reconstruction and improve the model's ability to represent the data.

4. The method for preventing and coordinating transient stability of a power system according to claim 1, characterized in that: In S3, the risk assessment function is constructed as: R=ω1·s v1 +ω2·s f1 +…+oh k ·s nk ; Among them, s v1 ,s f1 ,…,s nk are the standardized scores of various stability indicators after DBN evaluation; ω1, ω2, ω k are the weights corresponding to each stability index, ω1, ω2, ω k Determined by their impact on system stability.

5. The method for preventing and coordinating transient stability of a power system according to claim 1, characterized in that: In S3, the step of determining the risk level of the power grid according to the risk assessment function includes: Set the risk level threshold, and set the low risk threshold T l and medium risk threshold T m To classify the risk level, T l and T m It is set based on historical events, system characteristics and acceptable risk level. l and T m It is dynamically adjusted and updated according to system operation experience and changes in the external environment; The risk level is expressed as: Low risk: R < T l ; Medium risk: T l <R<T m ; High risk: R ≥ T m ; Generate a preventive control strategy and develop preventive measures based on risk levels: Low risk, monitor system status, maintain current operations, and prepare for potential risks; Medium risk: optimize power generation and load planning, adjust grid operation mode, and reduce risks; High risk, take emergency measures such as shedding load, adjusting generator output and activating backup equipment.

6. The method for preventing and coordinating transient stability of a power system according to claim 1, characterized in that: In S4, the method for constructing the advantage function includes: S411. Determine the impact indicators and determine the expected output or effect of each preventive control strategy based on the strategy generated in S3; S412. Quantify the impact of strategies. For each preventive control strategy, quantify its impact on stability and economic indicators; S413. Construct an advantage function, which combines stability and economic indicators. The advantage function is expressed as: A=α·R t +(1-a)·R e ; R e =λ1·e1+λ2·e2+…; α is the weight for weighing the importance of system stability and economy. The selection of weight α should be based on the judgment of the importance of system stability and economy and adjusted according to different situations. α is a weight between 0 and 1, which is used to balance the stability and economy indicators; R t is the risk assessment function R t Output; R e Comprehensive economic score; i is the weight of the i-th economic indicator, reflecting the importance of this indicator in the overall economic evaluation, e i is the cost-effectiveness score of the ith economic indicator, e i It is a monetary unit or a percentage of cost savings.

7. The method for preventing and coordinating transient stability of a power system according to claim 1, characterized in that: In S4, the method of evaluating and optimizing the control strategy using the constructed advantage function includes: S421. Quantify the stability index of the advantage function, extract the index related to stability from the constructed advantage function, and quantify the stability index; the stability index includes voltage stability and frequency stability, the voltage stability is quantified by the percentage of voltage deviation, and the frequency stability is quantified by the system frequency deviation or the frequency change rate; S422. Design preventive control strategies based on quantitative stability indicators. The preventive control strategies include: Adjustment of power generation plan: adjusting the output of generator sets according to system load and voltage stability indicators; Load distribution optimization, redistributing grid load to reduce stress points and improve overall stability; Activation of backup resources: activating backup generators or adjusting the operation of existing resources when necessary; For each control strategy, calculate the change in the advantage function after applying the strategy: ΔA=A(C)-A; A(C)=α·R t (C)+(1-α)·R e ; A(C) is the system advantage score after applying the control strategy; R t (C) is the output of the risk assessment function R after applying the control strategy, α is the weight that weighs the importance of system stability and economy, and R e Comprehensive economic rating; The effect of the control strategy is evaluated according to the value of ΔA. If ΔA>0, it means that the control strategy improves the advantage score of the system, that is, it improves the stability and economy of the system.

8. The method for preventing and coordinating transient stability of a power system according to claim 1, characterized in that: Methods for finding the optimal control strategy using the NSGA-II algorithm include: S51. Initialize the population and randomly generate a set of potential control strategies as the initial population P0. For each solution P in the population, use S4 to construct the advantage function and calculate each advantage function A(P). A(P) is expressed as: A(P)=α·R t (P)+(1-α)·R e ; R t (P) is the stability score of solution P, R e is the economic score, α is the weight that weighs the importance of system stability and economy; S52. Non-dominated sorting, for each pair of solutions P in the population i and P j , if P i No better than P on all targets j worse, and better in at least one objective, then P i Non-dominated P j ; All solutions that do not dominate other solutions are divided into one level; remove the solutions that are not dominated by them from the obtained solutions, and the remaining solutions constitute the second level, and so on; S53. For each non-dominated level S i The strategy in , calculates the crowding distance D(P) to measure the distribution density of solutions within their non-dominated levels; A max is the maximum value of the optimization function in the entire population; A min is the minimum value of the optimization function in the entire population; A max,s is the non-dominated level S of the solution P i The maximum value of the optimization function in A max,s is the non-dominated level S of the solution P i Minimum value of the optimization function in ; S54. Select excellent strategies based on non-dominated sorting and crowding distance to form a new generation of population; Select solutions with high non-dominated levels and large crowding distances as parent solutions to form a new generation of population; S55. Perform crossover and mutation operations on the selected strategy to generate a new offspring strategy; Randomly select a parent solution, convert each parameter of the parent solution into a binary string, select one or more crossover points, the crossover points are random, exchange the gene fragments of the parent solution at the crossover points, generate a new offspring solution, and the crossover operation is one of single-point crossover, multi-point crossover or uniform crossover; Perform mutation operations on the offspring solutions generated by crossover to introduce new genetic mutations, and collect the offspring solutions generated by crossover and mutation to form a new offspring population; S56. Repeat S51 to S54, each iteration including generating a new generation of population, calculating crowding distance, selecting excellent strategies, performing crossover and mutation operations, until the quality of the solution reaches the predetermined standard.

Images