Station grid-connected oscillation working condition prediction method based on frequency domain stability analysis, storage medium

Abstract

The present application relates to a method for predicting grid-connected oscillation working conditions of a power station based on frequency domain stability analysis, and a storage medium, which establishes an admittance database and a precise power station frequency domain model, and selects the optimal initial working condition in combination with negative damping characteristics, thereby solving the problem of unreasonable initial working condition selection and large prediction deviation in the prior art from the source, achieving significant improvement in the prediction accuracy of oscillation working conditions, and effectively avoiding the potential safety hazards of the power grid caused by inaccurate prediction; the influence factor of the stability margin on the quantitative parameters is used to lock the key parameter set, thereby eliminating the disadvantages of low efficiency of full parameter traversal and loss of information in dimension reduction without weight in the prior art, significantly reducing the calculation amount and improving the optimization efficiency while ensuring that the prediction accuracy is not affected, meeting the actual needs of real-time prediction and rapid response in engineering field, and significantly improving the engineering practicability of the scheme.

Description

Technical Field

[0001] This invention relates to the field of power system automation, and in particular to a method and storage medium for predicting grid-connected oscillation conditions of power plants based on frequency domain stability analysis. Background Technology

[0002] As the global energy structure transitions towards clean and low-carbon energy, the large-scale grid connection of new energy power plants such as wind power, photovoltaics, and energy storage has become a core trend in the construction of new power systems. The proportion of new energy power generation in the power system continues to rise, gradually becoming an important support for power supply. However, the core equipment of new energy power plants (such as inverters and converters) has characteristics of strong nonlinearity, low inertia, and weak damping. In addition, the grid connection process of these power plants faces complex scenarios such as the coordinated operation of multiple devices, grid load fluctuations, and random changes in environmental factors (wind speed, sunlight), leading to frequent grid connection oscillation problems, which seriously threaten the safe and stable operation of the power grid and the efficient consumption of new energy power.

[0003] Grid connection oscillation, as a typical risk in the grid connection operation of new energy power plants, may cause equipment overcurrent, overvoltage, grid disconnection and other faults, and may even spread to the entire regional power grid, causing large-scale power outages. Therefore, accurate and efficient prediction of grid connection oscillation conditions of power plants, early identification of oscillation risks and location of risk roots have become key technical requirements to ensure the safe and stable operation of new power systems.

[0004] Currently, various technologies for predicting grid-connected oscillation conditions of renewable energy power plants have emerged in the industry. However, these technologies still have many shortcomings and have failed to effectively address the core pain points in practical engineering applications. They are also insufficient to meet the demand for refined and efficient prediction under the background of high proportion of renewable energy grid connection. The specific limitations are as follows:

[0005] First, the coverage of oscillation conditions is incomplete, and the problem of risk omission is prominent. Existing prediction methods mostly focus on single types of oscillations or preset typical risk scenarios, without fully considering the coupled effects of complex factors such as the coordinated operation of multiple devices in new energy power plants, grid parameter fluctuations, and random load changes. They cannot cover all potential combinations of oscillation conditions under multiple devices and multiple operating conditions, and are prone to omitting key risk scenarios, resulting in insufficient comprehensiveness of prediction results and failing to provide comprehensive support for grid safety and prevention.

[0006] Secondly, the optimization efficiency is low, and its practicality in engineering is poor. The influencing factors of grid-connected oscillation conditions in new energy power plants are numerous, involving multiple dimensions such as equipment parameters, grid parameters, and operating condition parameters. These parameters have a high dimensionality, and existing prediction methods mostly employ full parameter traversal or unweighted dimensionality reduction for optimization and prediction. Full parameter traversal leads to a surge in computational load and low optimization efficiency, while unweighted dimensionality reduction easily loses key parameter information, affecting prediction accuracy. Neither method can simultaneously achieve both optimization efficiency and prediction accuracy, making it difficult to meet the actual needs of real-time prediction and rapid response in engineering projects.

[0007] Finally, the prediction results are incomplete, making it difficult to pinpoint the root cause of the risk. Most existing prediction technologies can only output basic prediction results of oscillation risk, such as oscillation warning signals and risk levels, without establishing a complete correlation link between "oscillation results - operating parameters - impact mechanisms." They cannot clarify the quantitative relationship between different operating parameters and oscillation results, nor can they locate the core operating parameters that cause oscillations. As a result, maintenance personnel can only know that there is an oscillation risk, but cannot accurately determine the root cause of the risk, making it difficult to take targeted prevention and control measures and fundamentally reduce the probability of oscillation accidents.

[0008] In summary, existing technologies for predicting grid-connected oscillation conditions at renewable energy power plants suffer from core drawbacks such as incomplete coverage of oscillation conditions, low optimization efficiency, and incomplete prediction results. These shortcomings make it difficult to meet the requirements for the safe and stable operation of new power systems under the background of high-proportion renewable energy grid connection. Therefore, developing a method for predicting grid-connected oscillation conditions at power plants that can solve the above-mentioned pain points and take into account the comprehensiveness, efficiency, and completeness of prediction has become an urgent technical challenge to be solved in the field of power systems and their automation. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a method for predicting grid-connected oscillation conditions of power stations based on frequency domain stability analysis, comprising:

[0010] S1: Establish an admittance database and construct a station frequency domain model;

[0011] S2: Based on the negative damping degree and frequency domain model of each device in the admittance database, select the operating condition with the highest and most stable negative damping degree as the current initial operating condition;

[0012] S3: Based on the frequency domain model, calculate the influence factors of each parameter of each device on the stability margin under the current initial operating conditions, and determine the set of key parameters based on the influence factors;

[0013] S4: In the set of key parameters, calculate the stability margin of each key parameter under different adjustment strategies to obtain candidate adjustment strategies with decreasing stability margin; among the candidate adjustment strategies, select the adjustment strategy with the largest decrease in stability margin as the current adjustment strategy.

[0014] S5: According to the current adjustment strategy, continuously adjust each parameter in the key parameter set with the adjustment step size, and determine whether each updated working condition is stable. If it is, no action is taken; otherwise, output the oscillation working condition until the boundary of any key parameter is reached. Delete the current adjustment strategy from the candidate adjustment strategies, return to step S4, and reselect the adjustment strategy with the largest stability margin descent gradient as the current adjustment strategy. Continue until all candidate adjustment strategies are traversed.

[0015] Further, step S2 includes:

[0016] S21: Calculate the negative damping area of ​​each device based on the admittance data of each device in the admittance database;

[0017] S22: Based on the negative damping zone area of ​​each device, and taking into account the capacity weight of each device in the station, the negative damping degree of each working condition is determined by weighting.

[0018] S23: Arrange the operating conditions in descending order according to the degree of negative damping of each operating condition; determine whether the current operating condition is stable in order; if stable, take the current operating condition as the current initial operating condition; if oscillating, output the current operating condition as the oscillating operating condition, and update the next operating condition of the current operating condition in the descending order to the current operating condition, and perform stability judgment again; until the operating condition with the maximum degree of negative damping and stability is obtained as the current initial operating condition.

[0019] Furthermore, step S3, which involves calculating the influence factors of each parameter of each device on the stability margin, includes:

[0020] S31: Calculate the participation factor of each piece of equipment in the station under the current operating conditions, and determine the global weight of each piece of equipment's impact on the stability margin;

[0021] S32: Under the current operating conditions, adjust only a single parameter of a single device at a time, and determine the local weight of the impact of each parameter adjustment on the stability margin;

[0022] S33: Calculate the influence factor of each parameter of each device on the stability margin based on the global weight of each device and the local weight of each parameter.

[0023] Further, step S32 includes:

[0024] S321: Set the adjustment step size of the parameter, and adjust the individual parameter of a single device in both the positive and negative directions according to the set adjustment step size, while keeping other parameters unchanged, to obtain the positive update condition under positive direction adjustment and the negative update condition under negative direction adjustment.

[0025] S322: Perform stability modal analysis on the positive and negative update conditions respectively to determine the stability margins of the positive and negative update conditions.

[0026] S323: Determine the local weight of the influence of each parameter adjustment on the stability margin based on the ratio of the difference between the stability margin of the positive update condition and the stability margin of the negative update condition to twice the adjustment step size.

[0027] Further, in step S321, the step of setting the adjustment step size of the parameter includes:

[0028] S3211: Determine the adjustment step threshold for each parameter based on its numerical adjustment range;

[0029] S3212: Based on the adjustment step size threshold of each parameter, determine the first step size and the second step size for adjusting each parameter; wherein, the second step size is twice the first step size, and the second step size does not exceed the adjustment step size threshold.

[0030] S3213: Adjust the corresponding parameters with the first step length and the second step length respectively to obtain the stability margin change rate under the first step length adjustment and the stability margin change rate under the second step length adjustment.

[0031] S3214: Determine whether the change rate of stability margin under the second step size adjustment and the change rate of stability margin under the first step size adjustment are less than the set threshold. If so, set the first step size as the adjustment step size of the parameter; otherwise, return to step S3212 and redetermine the first step size and the second step size of each parameter adjustment in the direction of shrinkage.

[0032] Furthermore, step S32 also includes:

[0033] S320: Between the first step size and the second step size, set several equal adjustment step sizes; under each set adjustment step size, execute steps S321-S323 in sequence to obtain the local weight of each parameter under each adjustment step size.

[0034] S324: Based on the local weights of each parameter under each adjustment step, determine the final local weight of each parameter by weighting.

[0035] Furthermore, in step S324, the formula for calculating the final local weight of each parameter using weighted average is as follows:

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] in, For the first The first device The final local weights of each parameter For the first The calculated number of steps is given by the set adjustment step size. The first device Local weights of each parameter, For the first A base coefficient for setting the adjustment step size. For the first A safety factor for setting and adjusting the step size. For the first The contribution coefficient of a set adjustment step size For the first The first device The parameter of the first One adjustment step size For the first The first device The maximum adjustment step size for each parameter For the first The first device The minimum adjustment step size for each parameter.

[0041] Furthermore, step S3, the step of determining the set of key parameters based on the impact factors, includes:

[0042] S34: Arrange the influence factors of each parameter of each device on the stability margin in descending order; and calculate the sum of the influence factors of all parameters of all devices as the global influence factor;

[0043] S35: Select the first M' parameters, construct the proposed key parameter set, and calculate the sum of the influence factors of all parameters in the proposed key parameter set as the local influence factor;

[0044] S36: Calculate whether the ratio of the local impact factor to the global impact factor exceeds the set threshold. If so, use the proposed key parameter set as the final key parameter set; otherwise, let M'+1 be the updated M' and return to step S35.

[0045] Further, step S4 includes:

[0046] S41: Under different adjustment directions, perform small-step numerical adjustments on each key parameter, calculate the corresponding stability margin change, and construct a training dataset;

[0047] S42: Using the numerical changes of each key parameter in the training dataset as input and the changes in stability margin as output, construct a stability margin gradient prediction model to predict the stability margin descent gradient of each key parameter under different adjustment directions.

[0048] S43: Based on the prediction results, determine the adjustment strategy with the largest stability margin descent gradient, and use it as the current adjustment strategy.

[0049] Furthermore, the stable margin gradient prediction model includes: an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence;

[0050] The input layer is used to input the changes in the values ​​of each device parameter in the candidate parameter set;

[0051] The first hidden layer is used to focus on the mapping relationship between the change of a single parameter and the change of the stability margin, extract the linear correlation features and nonlinear sensitivity features of a single parameter, and obtain the original features;

[0052] The second hidden layer is used to mine the synergistic gain and suppression effect of different key parameter adjustment combinations on the stability margin change based on the original features, and to extract the first-order interactive coupling features between parameters.

[0053] The third hidden layer is used to fuse the original features with the first-order coupled features, expand the feature expression dimension, and extract the high-order nonlinear coupled features between key parameters.

[0054] The output layer is used to output the stability margin variation based on the high-order nonlinear coupling characteristics.

[0055] On the other hand, the present invention also provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned methods for predicting grid-connected oscillation conditions of power plants.

[0056] On the other hand, the present invention also provides a computer system, including a memory and a processor; the memory stores program code that can be executed by the processor; the program code is used to execute any of the above-mentioned methods for predicting grid-connected oscillation conditions of power plants.

[0057] This invention relates to a method and storage medium for predicting grid-connected oscillation conditions of power plants based on frequency domain stability analysis. By establishing an admittance database and a precise frequency domain model of the power plant, and combining negative damping characteristics to select the optimal initial operating condition, it fundamentally solves the problems of unreasonable initial operating condition selection and large prediction deviations in existing technologies. This significantly improves the accuracy of oscillation condition prediction and effectively avoids potential power grid safety hazards caused by inaccurate predictions. By quantifying the influence factors of parameters on stability margin and locking the key parameter set, it eliminates the drawbacks of low efficiency in full parameter traversal and loss of information due to unweighted dimensionality reduction in existing technologies. While significantly reducing the amount of computation and improving optimization efficiency, it ensures that the prediction accuracy is not affected. This can meet the actual needs of real-time prediction and rapid response in engineering sites, and significantly improve the engineering practicality of the solution. Attached Figure Description

[0058] Figure 1A flowchart of an embodiment of the method for predicting grid-connected oscillation conditions of power plants according to the present invention;

[0059] Figure 2 A schematic diagram of one embodiment of the operating condition coding for a station;

[0060] Figure 3 This is a schematic diagram of one embodiment of the negative damping interval area. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0062] It should be noted that if the embodiments of the present invention involve directional indications, such as up, down, left, right, front, back, etc., these directional indications are only used to explain the relative positional relationships and movement of the components in a specific posture. If the specific posture changes, the directional indications will also change accordingly. Furthermore, if the embodiments of the present invention involve descriptions such as "first," "second," "S1," "S2," "step one," "step two," etc., these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance, or implicitly indicating the number of technical features indicated or the execution order of the method. Those skilled in the art will understand that anything that does not violate the inventive concept should be included within the scope of protection of the present invention.

[0063] like Figure 1 As shown, this invention provides a method for predicting grid-connected oscillation conditions of power stations based on frequency domain stability analysis, comprising:

[0064] S1: Establish an admittance database and construct a station frequency domain model;

[0065] Specifically, based on the components and connections of the site to be predicted, an admittance database can be established through simulation, and a frequency domain model of the site can be constructed.

[0066] Preferably, step S1 may include:

[0067] S11: For power electronic equipment in the station, acquire admittance data under each operating condition through multi-condition frequency sweep test, establish an admittance database, and construct a "condition-admittance" mapping model.

[0068] Specifically, power electronic equipment, including but not limited to wind turbines, photovoltaic systems, energy storage, SVG, and other equipment with actively adjustable power parameters. More specifically, admittance data, including but not limited to positive-sequence self-admittance, positive-sequence coupled admittance, negative-sequence self-admittance, and negative-sequence coupled admittance under various operating conditions.

[0069] More specifically, step S11 preferably includes:

[0070] S111: For power electronic equipment in the power station, adopt data-driven modeling to perform frequency scanning on the power electronic equipment under typical operating conditions (operating condition parameters: output active power, reactive power) to obtain the positive sequence self-admittance, positive sequence coupled admittance, negative sequence self-admittance, and negative sequence coupled admittance of the equipment under typical operating conditions, and establish an admittance database.

[0071] Specifically: when the injected perturbation frequency is When the positive sequence voltage is disturbed, the disturbance frequency at the device port is... and coupling frequency The positive-sequence self-admittance is obtained from the current response. and positive-sequence coupling admittance The expression is:

[0072] ;

[0073] in, This represents the positive-sequence current response at the disturbance frequency. This is a positive-sequence voltage disturbance. The negative sequence current response at the coupling frequency; ; The perturbation frequency; This is the fundamental frequency.

[0074] When the injected perturbation frequency is Negative sequence voltage disturbance, the disturbance frequency at the device port and coupling frequency The negative-sequence self-admittance is obtained from the current response. and negative order coupling admittance The expression is:

[0075] ;

[0076] in, This represents the negative sequence current response at the disturbance frequency. This is a negative sequence voltage disturbance. This represents the positive-sequence current response at the coupling frequency.

[0077] Preferably, establishing an admittance database may also include: preprocessing the collected working condition admittance data, including: outlier removal, data partitioning, etc.; for example: dividing the data into a training set and a test set according to a ratio, with the training set used for subsequent model training and the test set used for subsequent model accuracy verification.

[0078] S112: Based on the admittance database, construct a "condition-admittance" mapping model with condition parameters and frequency as inputs and admittance amplitude and admittance phase as outputs.

[0079] Specifically, the operating parameters and frequency can be selected as inputs, and the admittance amplitude and admittance phase can be selected as outputs. An artificial intelligence algorithm is used to construct an admittance database model for the equipment. Using the admittance database, the model is optimized through cross-validation to ensure the fitting accuracy of the model under different operating conditions. In this way, an "operating condition-admittance" mapping model for the power electronic equipment is constructed.

[0080] More specifically, one can choose to combine the nonlinear correlation characteristics between operating condition parameters and admittance data, select an appropriate model algorithm to construct an "operating condition-admittance" mapping model, and prioritize algorithms that balance accuracy and efficiency (such as BP neural networks, random forests, support vector machines, etc.) to avoid the inefficiency caused by using full parameter traversal. Using the preprocessed training set as input, the operating condition feature parameters are used as the model input layer, and the frequency domain characteristic indicators such as admittance amplitude, phase, and resonant frequency are used as the model output layer for model training. By iteratively adjusting the model hyperparameters (such as the number of hidden layers and learning rate of the neural network, and the number of decision trees of the random forest), the model prediction error is minimized to ensure that the model can accurately capture the correlation between operating condition parameters and admittance frequency domain characteristics.

[0081] S12: Mechanistic modeling is used for passive equipment in the station to construct a mechanism model;

[0082] Specifically, passive devices can be selected, but are not limited to, transmission lines, transformers, and other equipment without actively regulated power parameters. More specifically, for passive devices within a power station, mechanistic modeling is employed. This involves establishing a mathematical expression that accurately describes the admittance frequency response based on the device's physical structure and electromagnetic characteristics, using admittance parameters to precisely describe the component's frequency domain characteristics. For example, transmission lines can include positive-sequence, negative-sequence, and zero-sequence parameters; short-line models can neglect ground capacitance; medium- and long-line models use a Π-type equivalent circuit, considering ground capacitance; long-line models use a distributed parameter model, considering wave propagation effects. Transformers can be modeled using a T-type equivalent circuit to accurately reflect primary and secondary parameters and excitation characteristics.

[0083] S13: Based on the mapping model and mechanism model, and combined with topological relationships, assemble and construct the frequency domain model of the station.

[0084] Specifically, based on the actual situation of the site to be predicted, the topological connection relationship of each node in the site can be established, and the transmission characteristics, coupling effects, resonant frequencies and attenuation laws between nodes can be described in the frequency domain (frequency dimension) to construct the frequency domain model of the site to be predicted.

[0085] Preferably, to differentiate between different operating conditions of the site to be predicted, an operating condition code can be assigned to each operating condition before oscillation prediction, based on the equipment included in the site and the operating parameters of each piece of equipment. This ensures that the correct operating condition code is output during subsequent operating condition updates to correspond to the current operating condition. It is worth noting that this operating condition code is a unique identifier for each operating condition and can adopt any identifier format. For example... Figure 2 As shown, in a preferred embodiment, each power electronic device (including wind turbines, photovoltaics, energy storage, SVG, etc.) in the power station can be divided first, and then the internal operating parameters (active power, reactive power, etc.) of each device can be divided. All operating parameters of all devices can be encoded in an encoding method to obtain a unique identifier for each operating condition. For example, the unique operating condition combination code is formed in the form of "device name" + "operating parameter" + "operating value".

[0086] S2: Based on the negative damping degree and frequency domain model of each device in the admittance database, select the operating condition with the highest and most stable negative damping degree as the current initial operating condition;

[0087] Specifically, in step S2, only two requirements are set for the selection of the current initial operating condition: the highest degree of negative damping and stability. In one embodiment, the negative damping degree of each device can be determined first based on the admittance database. Then, the devices in the admittance database are sorted in descending order of their negative damping degrees, and the operating condition with the highest degree of negative damping is judged sequentially to determine whether it is stable. If it is, the operating condition with the highest degree of damping is selected as the current initial operating condition. If not, i.e., the operating condition with the highest degree of damping oscillates, the operating condition with the second highest degree of negative damping is selected as the current operating condition, and stability is judged again... and so on. If oscillation still occurs, the operating conditions with the third, fourth, ... degrees of negative damping are selected, and stability is judged again, until the operating condition that satisfies both the highest degree of negative damping and stability is found as the current initial operating condition. In another embodiment, the stability of all operating conditions can be judged first, and then the operating condition with the highest degree of negative damping is sought among the stable operating conditions. It can be seen that the key to step S2 is to select the operating condition with the highest degree of negative damping and stability as the current initial operating condition. As for how to select it, the present invention does not impose any limitations.

[0088] Preferably, in step S2, the degree of negative damping of each device is as follows: Figure 3As shown, the negative damping interval area S of each device can be selected, which is the area of ​​the real part of the impedance curve Re(f) less than 0. The larger the area, the higher the degree of negative damping, and the smaller the area, the lower the degree of negative damping. Based on the area of ​​the negative damping interval, the operating condition with the highest degree of negative damping and the most stable condition is determined, which is the current initial operating condition.

[0089] Preferably, step S2 may further consider the impact factors of each device on the overall station, comprehensively consider the negative damping degree of each device in each operating condition, and weigh and determine the current initial operating condition. In a preferred embodiment, step S2 includes:

[0090] S21: Calculate the negative damping area of ​​each device based on the admittance data of each device in the admittance database;

[0091] Specifically, such as Figure 3 As shown, the portion of the impedance real part curve where Re(f) is less than 0 is the negative damping region. Since the degree of negative damping is related to its negative damping frequency range and corresponding numerical value, the larger the area of ​​the negative damping region, the greater the degree of negative damping. Specifically, the area S of the negative damping region, i.e., the area of ​​the admittance real part curve Re(f) less than 0, is calculated using the following formula:

[0092] ;

[0093] in,( , () represents the negative damping range. is the real part of the curve; S is the area of ​​the negative damping interval.

[0094] S22: Based on the negative damping zone area of ​​each piece of equipment and considering the capacity weight of each piece of equipment in the station, the negative damping degree for each operating condition is determined by weighted average; the following formula can be used for calculation:

[0095] ;

[0096] in, For the first The capacity weight of each device can be selected as the ratio of the device capacity to the total capacity of the station, i.e., the entire system. For the first Area of ​​the negative damping zone of each device; The degree of negative damping under operating condition t.

[0097] S23: Arrange the operating conditions in descending order according to the degree of negative damping of each operating condition; determine whether the current operating condition is stable in order; if stable, take the current operating condition as the current initial operating condition; if oscillating, output the current operating condition as the oscillating operating condition, and update the next operating condition of the current operating condition in the descending order to the current operating condition, and perform stability judgment again; until the operating condition with the maximum degree of negative damping and stability is obtained as the current initial operating condition.

[0098] Specifically, according to the degree of negative damping under operating conditions Based on the size, sort the various operating conditions in descending order, first selecting... The maximum operating condition is the current operating condition. To distinguish between different operating conditions, a previously set operating condition code can be used as the unique identifier. Under this operating condition code, a stability analysis is performed on the frequency domain model. The "generalized Nyquist criterion based on the determinant" can be used for frequency domain stability analysis. If the frequency domain stability analysis shows a stable state, then this operating condition is taken as the current initial operating condition. If the analysis shows an oscillating state (instability), then the oscillating operating condition code is output, and the next operating condition code is selected according to its size (i.e., the second operating condition code in descending order). Stability judgment is then performed until the first operating condition code that is found to be stable is taken as the current initial operating condition.

[0099] S3: Based on the frequency domain model, calculate the influence factors of each parameter of each device on the stability margin under the current initial operating conditions, and determine the set of key parameters based on the influence factors;

[0100] Specifically, assuming that all equipment has a total of N operating parameters, under the current initial operating conditions, the influence factor of each parameter of each equipment on the stability margin is calculated. Based on the magnitude of the influence factor, the M parameters with the largest influence factors are selected from the N operating parameters to construct a key parameter set.

[0101] Preferably, step S3, which involves calculating the influence factors of each parameter of each device on the stability margin, includes:

[0102] S31: Calculate the participation factor of each piece of equipment in the station under the current operating conditions, and determine the global weight of each piece of equipment's impact on the stability margin.

[0103] Specifically, based on the frequency domain model, frequency domain modal analysis is performed on the current equipment operating condition using the "generalized Nyquist criterion based on eigenvalues". This calculates the stability margin and participation factor of the station under the current equipment operating condition, i.e., the combined coding of this operating condition. The participation factor characterizes the influence of each piece of equipment on the stability margin, and the global weight of each piece of equipment's influence on the stability margin is determined. (Optional) No. The global weights of each device; preferably, the global weights can be normalized so that the range of the global weights is [0,1] and the sum of the global weights of all devices is 1.

[0104] S32: Under the current operating conditions, adjust only a single parameter of a single device at a time, and determine the local weight of the impact of each parameter adjustment on the stability margin.

[0105] Specifically, each time only a single parameter of a single device is adjusted, and the change in stability margin before and after adjustment is used as the local weight of the impact of each parameter adjustment on the stability margin. Preferably, step S32 includes:

[0106] S321: Set the adjustment step size of the parameter, and adjust the individual parameter of a single device in both the positive and negative directions according to the set adjustment step size, while keeping other parameters unchanged, to obtain the positive update condition under positive direction adjustment and the negative update condition under negative direction adjustment.

[0107] S322: Perform stability modal analysis on the positive and negative update conditions respectively to determine the stability margin of the positive and negative update conditions; specifically, the stability margin can also be obtained by frequency domain modal analysis using the "generalized Nyquist criterion based on eigenvalues".

[0108] S323: Determine the local weight of each parameter based on the ratio of the difference between the stability margin of the positive update condition and the stability margin of the negative update condition to twice the adjustment step size.

[0109] Specifically: By using local weights, the influence of each parameter on the stability margin near the current operating point can be reflected. The following formula can be used for calculation:

[0110] ;

[0111] in, For the first The first device One parameter, To adjust the step size, To provide a stability margin for the positive update operating condition, For the stability margin of negative update conditions; For the first The first device Local weights of each parameter;

[0112] Specifically, taking the first Taking a single device as an example, assume that its internal operating parameters include two, namely... =2, specifically: For active power P, Let Q be the reactive power. The corresponding full operating range of this equipment is: active power 0-1 p.u., reactive power -0.3-0.3 pu. Therefore, by adjusting the operating parameters P or Q in small steps, assuming the adjustment step size is set to ±ΔA = ±0.05 pu, while other parameters remain unchanged, then:

[0113] Calculate the absolute value of the partial derivative of the operating condition parameter - active power:

[0114] ;

[0115] Calculate the absolute value of the partial derivative of the operating condition parameter - reactive power:

[0116]

[0117] in, For the first The local weight of the active power of each device can also be... ; For the first The local weight of the reactive power of each device can also be... ; To increase active power The stability margin of the positive update operating condition; To reduce active power The stability margin of the negative update condition; To increase reactive power The stability margin of the positive update operating condition; To reduce reactive power The stability margin of the negative update condition.

[0118] Preferably, step S32 further includes adjusting the local weights. Normalization is performed so that the weight values ​​of the local weights are in the range of [0,1], and the sum of the weights of the parameters under the same equipment is 1.

[0119] ;

[0120] in, For the first The first device Local weights after parameter normalization; For the first The number of parameters for each device; for example, taking the first device as an example. For example, the parameters of a device include two components: active power P and reactive power Q.

[0121] ;

[0122] ;

[0123] in, For the first Local weights of the active power of each device. For the first Local weights of reactive power of individual devices For the first The local weights of the active power of each device after normalization. For the first The local weights of the reactive power of each device after normalization.

[0124] More specifically, to prevent the results from varying significantly due to an unreasonable step size ΔA (too small and ineffective and too large and unpredictable), step S321, which sets the step size for adjusting the parameters, includes:

[0125] S3211: Determine the adjustment step threshold for each parameter based on its numerical adjustment range; specifically, firstly, based on the operating condition numerical adjustment range of different parameters, the adjustment step is preferably no greater than 10% of its own adjustment range length. For example, the adjustment step threshold is set to 0.1;

[0126] S3212: Based on the adjustment step size threshold of each parameter, determine the first step size and the second step size for adjusting each parameter; wherein, the second step size is twice the first step size, and the second step size does not exceed the adjustment step size threshold.

[0127] Specifically, the second step size can be selected as the maximum adjustment step size threshold, which is 0.1; the first step size is half of the second step size, which is 0.05.

[0128] S3213: Adjust the corresponding parameters with the first step length and the second step length respectively to obtain the rate of change of stability margin under the first step length adjustment. rate of change of stability margin under second step size adjustment Specifically, the following formula can be used for calculation:

[0129] ;

[0130] ;

[0131] in, This represents the stability margin when the operating conditions are not adjusted. This represents the stability margin under the first long adjustment. For the first step, For the first The first device One parameter, This is the stability margin under the second step size adjustment. This is the second step length.

[0132] S3214: Determine whether the change rate of stability margin under the second step size adjustment and the change rate of stability margin under the first step size adjustment are less than the set threshold; if so, set the first step size as the adjustment step size of the parameter; otherwise, return to step S3212 and redetermine the first step size and the second step size of each parameter adjustment in the direction of shrinkage.

[0133] In this embodiment, a preferred embodiment for dynamically correcting the adjustment step size is given. First, an adjustment step size threshold is determined, and then a first step size and a second step size are determined, preferably taking... =0.05、 =0.1; then calculate the stability margin change rate (i.e., stability margin sensitivity under step size adjustment) for each step size adjustment, and determine whether the change in the stability margin change rate under the first step size and the second step size (twice the first step size) adjustment is less than the set threshold; specifically, this change can be expressed as, but is not limited to, the difference between the stability margin change rate under the second step size adjustment and the stability margin change rate under the first step size adjustment, the difference between the ratio percentage and 100%, etc.; in the example, less than 5%, that is, the change is not significant, which indicates that the current first step size is reasonable, and this reasonable first step size can be set as the adjustment step size of the parameter; if the change in the stability margin change rate under the first step size and the second step size (twice the first step size) adjustment is not less than the set threshold, it indicates that the first step size is too large and the value changes too quickly. The first step size and the second step size should be re-determined in the direction of reduction, and the verification should be repeated until the preset condition that the change rate is within the set range is met.

[0134] More preferably, since different parameter adjustment step sizes will affect the value of local weights to some extent, in order to obtain more accurate and comprehensive local weights, step S32 further includes:

[0135] S320: Between the first step size and the second step size, set several equal adjustment step sizes; under each set adjustment step size, execute steps S321-S323 in sequence to obtain the local weight of each parameter under each adjustment step size.

[0136] No. The first device Taking the local weights of each parameter as an example, under multiple set step sizes, executing steps S321-S323 will yield the set. Where D is the number of adjustment steps; For the first The calculated number of steps is given by the set adjustment step size. The first device Local weights of each parameter;

[0137] S324: Based on the local weights of each parameter under each adjustment step, determine the final local weight of each parameter by weighting.

[0138] In a preferred embodiment, the formula for calculating the final local weight of each parameter using weighted average is as follows:

[0139] ;

[0140] ;

[0141] ;

[0142] ;

[0143] in, For the first The first device The final local weights of each parameter For the first The calculated number of steps is given by the set adjustment step size. The first device Local weights of each parameter, For the first A base coefficient for setting the adjustment step size. For the first A safety factor for setting and adjusting the step size. For the first The contribution coefficient of a set adjustment step size For the first The first device The parameter of the first One adjustment step size For the first The first device The maximum adjustment step size for each parameter For the first The first device The minimum adjustment step size for each parameter; the maximum and minimum adjustment step sizes can be arbitrarily set according to the actual situation of the equipment.

[0144] S33: Based on the global weights of each device and the local weights of each parameter, calculate the influence factor of each parameter of each device on the stability margin; the following formula can be used for calculation:

[0145] ;

[0146] in, For the first The first device The influence factors of each parameter on the stability margin are determined by combining the global weight of the equipment and the local weight of the parameter, comprehensively reflecting the degree of contribution of the parameter adjustment to the change in stability margin. This is the weighting balancing adjustment coefficient, its function is to dynamically correct... and The weighting of the influence varies depending on the specific circumstances; alternatively, it can be included in the formula. delete.

[0147] Preferably, in step S3, the step of determining the key parameter set based on the impact factor has been previously described in a specific embodiment whereby the key parameter set is constructed by sorting the N operating parameters in descending order based on the impact factor and selecting the M candidate parameters with the largest impact factor. The specific number of M can be arbitrarily set.

[0148] In a preferred embodiment, the step of determining the key parameter set based on the impact factor may optionally include:

[0149] S34: Arrange the influence factors of each parameter of each device on the stability margin in descending order; and calculate the sum of the influence factors of all parameters of all devices as the global influence factor;

[0150] S35: Select the first M' parameters, construct the proposed key parameter set, and calculate the sum of the influence factors of all parameters in the proposed key parameter set as the local influence factor;

[0151] S36: Calculate whether the ratio of the local impact factor to the global impact factor exceeds the set threshold. If so, use the proposed key parameter set as the final key parameter set; otherwise, let M'+1 be the updated M' and return to step S35.

[0152] This embodiment provides a preferred method for determining the key parameter set based on impact factors. The initial value of M' and the threshold for setting the ratio of local impact factors to global impact factors can be arbitrarily set according to the actual situation. The key to this embodiment is to first limit the number of parameters to M' to construct a proposed key parameter set; then check whether the quality of the M' parameters in the proposed key parameter set is high, that is, whether the ratio of the sum of the impact factors of these selected key parameters—local impact factors—to the global impact factors exceeds the set threshold. If so, it indicates that the selected key parameter set meets the standard; if not, it is necessary to add more parameters to the key parameter set, that is, select the first M'+1 parameters in descending order to construct a new proposed parameter set, and re-determine the ratio until the first M parameters are selected. The general requirement is: assuming there are a total of N operating condition parameters, then M must at least satisfy the condition that the sum of the impact factors of the M high-weight operating condition parameters (local impact factors) is higher than a certain proportion of the sum of the impact factors of all operating condition parameters (global impact factors). This certain proportion, the threshold for setting the ratio of local impact factors to global impact factors, can be arbitrarily set according to the specific situation of the new energy power plant. For power stations with a relatively simple type of new energy equipment, the preferred ratio is 85%~90%. For mixed power stations (such as wind, solar, and energy storage), the preferred ratio is 90%~98%. If it is necessary to quickly find oscillation conditions while accepting a certain sacrifice in coverage, the ratio can be adjusted to 80%~85%.

[0153] S4: In the set of key parameters, calculate the stability margin of each key parameter under different adjustment strategies to obtain candidate adjustment strategies with decreasing stability margin; among the candidate adjustment strategies, select the adjustment strategy with the largest decrease in stability margin as the current adjustment strategy.

[0154] Specifically, in the set of key parameters, there are many parameters, and each parameter has different adjustment directions (increasing / decreasing the parameter) and adjustment step sizes. Then, we can calculate the stability margin of each key parameter under different adjustment strategies in turn, and construct candidate adjustment strategies for all adjustment strategies that decrease the stability margin. Then, among the candidate adjustment strategies, we select the adjustment strategy with the largest decrease in stability margin gradient, that is, the fastest decreasing trend, as the current adjustment strategy. This is the simplest and most direct way.

[0155] However, due to the complexity of the equipment and the large number of parameters at the site, as well as the numerous adjustment directions and step sizes for each parameter, listing all the different adjustment strategies for each key parameter and calculating the stability margin under each strategy would result in an enormous computational burden. Therefore, in a preferred embodiment, step S4 specifically involves:

[0156] S41: Under different adjustment directions, perform small-step numerical adjustments on each key parameter, calculate the corresponding stability margin change, and construct a training dataset;

[0157] Specifically, this small step size adjustment is relative to the adjustment range of each key parameter. It only calculates the stability margin change within a small range of about 5%-10% of the adjustment range, rather than calculating all of them, in order to reduce the amount of calculation.

[0158] S42: Using the numerical changes of each key parameter in the training dataset as input and the changes in stability margin as output, construct a stability margin gradient prediction model to predict the stability margin descent gradient of each key parameter under different adjustment directions (parameter increase / decrease).

[0159] S43: Based on the prediction results, determine the adjustment strategy with the largest stability margin descent gradient, and use it as the current adjustment strategy.

[0160] Specifically, gradient boosting regression trees, fully connected neural networks, etc., can be used to construct a stability margin gradient prediction model. First, by making numerous small-step numerical adjustments to the operating parameters, the corresponding stability margin changes are calculated to obtain a training dataset. A pre-defined stability margin gradient prediction model is then trained to obtain the trained stability margin gradient prediction model. Next, the current adjustment strategy is input into the model to predict the stability margin descent gradient of each device parameter under different adjustment directions (parameter increase / decrease). The adjustment strategy with the largest stability margin descent gradient (including which device, which parameter, adjustment direction, and adjustment step size) is selected as the optimal current adjustment strategy.

[0161] The preferred stable margin gradient prediction model includes: an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence.

[0162] The input layer is used to input the changes in the values ​​of each device parameter in the candidate parameter set;

[0163] The first hidden layer is used to focus on the mapping relationship between the change of a single parameter and the change of the stability margin, extract the linear correlation features and nonlinear sensitivity features of a single parameter, and obtain the original features;

[0164] The second hidden layer is used to mine the synergistic gain and suppression effect of different key parameter adjustment combinations on the stability margin change based on the original features, and to extract the first-order interactive coupling features between parameters.

[0165] The third hidden layer is used to fuse the original features with the first-order coupled features, expand the feature expression dimension, and extract the high-order nonlinear coupled features between key parameters.

[0166] The output layer is used to output the stability margin variation based on the high-order nonlinear coupling characteristics.

[0167] More preferably, the stability margin gradient prediction model employs a lightweight fully connected neural network (MLP), where each layer is fully connected without convolution, pooling, or attention mechanisms, balancing model complexity with real-time engineering requirements. Specifically:

[0168] Input layer: Includes neurons with the same number of key parameters as the set of key parameters. Each neuron corresponds to the effective step size change of a single key parameter. It is used to receive the standardized parameter adjustment input and provide basic data for feature extraction.

[0169] The first hidden layer consists of 24 ReLU activation function neurons, fully connected to all neurons in the input layer and also fully connected to neurons in the second hidden layer. A Dropout layer (dropout rate 0.15) is configured, and L1 regularization (coefficient 1e-4) is introduced to extract linear correlation features and nonlinear sensitive features of individual parameters. This focuses on the direct mapping law between the parameter's own adjustment amount and the change in stability margin, suppresses sparse feature redundancy, and provides high-quality original features for deep feature extraction.

[0170] The second hidden layer consists of 16 LeakyReLU activation function neurons (negative slope 0.01), fully connected to all neurons in the first hidden layer and also fully connected to neurons in the third hidden layer; it is paired with a Batch Normalization (BN) layer; the LeakyReLU activation function is used to capture weak interaction signals, and the BN layer alleviates the gradient vanishing problem, ensuring the stability of the feature distribution; in order to explore the synergistic gain and suppression effects of different parameter adjustment combinations on the stability margin changes, the first-order interaction coupling features between parameters are extracted.

[0171] The third hidden layer consists of 8 GELU activation function neurons, fully connected to all neurons in the second hidden layer and also fully connected to neurons in the output layer; a Dropout layer (dropout rate 0.1) is configured to extract high-order nonlinear coupling features, fuse the basic features of the first two layers with the first-order coupling features and expand the feature expression dimension, accurately fit the complex mapping relationship between parameter adjustment and stability margin changes through the characteristics of Gaussian error linear units, filter high-frequency noise interference, and simultaneously complete feature dimension compression and core feature condensation, eliminate redundant feature information, output a simplified feature set strongly correlated with the stability margin gradient, balance model complexity and prediction performance, and provide accurate feature input for the output layer;

[0172] Output layer: includes one linear activation function neuron, which is fully connected to all neurons in the third hidden layer; the front end is connected to a BN layer to stabilize the output distribution, and the final output corresponds to the stability margin change of the adjustment combination.

[0173] More preferably, the model training uses the AdamW optimizer (weight decay coefficient 1e-3), the learning rate is adaptively adjusted through a cosine annealing strategy (initially 1e-3, minimum 5e-4), the batch size is set to 32~64, the maximum number of training iterations is 800, and early stopping (training stops if the validation set loss does not decrease after 15 consecutive iterations) and gradient pruning (pruning threshold 1.0) are combined to further improve the model's generalization ability, prediction accuracy and training stability, ensuring that the prediction time for a single sample is ≤10ms, meeting the real-time requirements of grid-connected oscillation control of new energy power plants.

[0174] S5: According to the current adjustment strategy, continuously adjust each parameter in the key parameter set with a certain step size, and judge whether each updated working condition is stable. If it is, no processing is done; otherwise, output the oscillation working condition until the boundary of any key parameter is reached. Return to step S4, delete the current adjustment strategy from the candidate adjustment strategies, and reselect the adjustment strategy with the largest stability margin decrease gradient as the current adjustment strategy. Continue until all candidate adjustment strategies are traversed.

[0175] Specifically, according to the current adjustment strategy, the operating parameters of each device in the site are modified under the current initial operating conditions to obtain new operating condition codes. Frequency domain stability analysis is performed using the "generalized Nyquist criterion based on determinants" to calculate the site's stability margin under the new operating conditions. The stability of the new operating conditions is determined; if stable, no action is taken; otherwise, the current operating condition code is output as the oscillation condition. Since the operating condition parameters have boundaries, the operating conditions are continuously adjusted according to the strategy within the numerical limits, recording all oscillation operating condition codes until a boundary of a parameter under the adjustment strategy is reached. This indicates that the adjustment strategy has reached its endpoint and cannot be further adjusted. At this point, the process returns to step S4, the current adjustment strategy is removed from the candidate adjustment strategies, and a new strategy with the fastest decrease in stability margin is obtained and iterated. Finally, all decreasing strategies are traversed. If an oscillation operating condition code is recorded, all operating condition codes are output; if no oscillation operating condition is found, the site is considered to have no oscillation risk.

[0176] A more preferred method further includes: outputting oscillation condition codes and recording stability margins. When oscillations occur at the station, the calculated weights can be used to pinpoint which equipment and which operating parameter has the greatest impact on the oscillations, and a stability enhancement scheme can be provided accordingly.

[0177] In summary, this invention provides a method for predicting grid-connected oscillation conditions of power plants based on frequency domain stability analysis. By establishing an admittance database and a precise frequency domain model of the power plant, and combining negative damping characteristics to select the optimal initial operating condition, it fundamentally solves the problems of unreasonable initial operating condition selection and large prediction deviations in existing technologies. This significantly improves the prediction accuracy of oscillation conditions and effectively avoids potential power grid safety hazards caused by inaccurate predictions. By quantifying the influence factors of parameters on stability margin and locking the key parameter set, it eliminates the drawbacks of low efficiency in full parameter traversal and loss of information due to unweighted dimensionality reduction in existing technologies. While significantly reducing the amount of computation and improving optimization efficiency, it ensures that the prediction accuracy is not affected. This method can meet the actual needs of real-time prediction and rapid response in engineering sites, significantly improving the engineering practicality of the solution.

[0178] To address the shortcomings of existing technologies in covering incomplete oscillation conditions and significant risk omissions, this solution iteratively traverses all candidate regulation strategies, simulating changes in operating conditions under different regulation methods for key parameters. It comprehensively covers potential oscillation conditions in complex scenarios such as multi-device collaboration, parameter fluctuations, and random environmental changes, thoroughly resolving the pain point of missing key risk scenarios and providing comprehensive technical support for power grid safety and control. Furthermore, the solution not only accurately outputs oscillation conditions but also clearly identifies the core parameters, key regulation methods, and parameter boundaries that trigger oscillations, establishing a complete correlation link between "oscillation results - operating parameters - impact mechanisms." This solves the problems of incomplete prediction results and difficulty in locating the root causes of risks in existing technologies, helping maintenance personnel quickly develop targeted prevention and control measures, fundamentally reducing the incidence of oscillation accidents, and avoiding serious faults such as equipment overcurrent, overvoltage, grid disconnection, and regional power grid outages.

[0179] Furthermore, this solution, based on frequency domain stability analysis, is adapted to the characteristics of strong nonlinearity, low inertia, and weak damping of core equipment in new energy power plants. It can effectively cope with complex grid connection scenarios such as the coordinated operation of multiple equipment, grid load fluctuations, and random changes in environmental factors. By predicting oscillation conditions in advance and quantifying the change law of stability margin, it provides a scientific basis for the optimization of grid connection parameters and stable control of power plants, further improving the stability and controllability of grid connection operation of new energy power plants, helping to efficiently consume new energy power, and providing reliable technical support for the clean and low-carbon transformation of the new power system.

[0180] 1. Significantly improves the accuracy and reliability of oscillation prediction, reducing the risk of misjudgment and omission. This method first establishes an admittance database and constructs an accurate frequency domain model of the power station. It uses the most stable and negatively damped condition as the initial condition, thus avoiding prediction deviations caused by unreasonable initial condition selection. Simultaneously, by quantifying the influence factors of each equipment parameter on stability margin, it accurately identifies the key parameter set. Subsequent adjustment strategy analysis and condition iteration are conducted based on these key parameters, ensuring that the prediction process revolves around the core oscillation causes. This allows the prediction results to truly reflect the actual oscillation evolution law during power station grid connection, providing accurate data support for stable power station operation.

[0181] 2. This method enables early prediction and precise location of oscillation conditions, providing proactive prevention and control for power plant grid-connected stability. It eliminates the need to wait for oscillations to actually occur. By traversing different adjustment strategies for key parameters, it simulates the dynamic changes in stability margin during parameter variations, capturing critical conditions where the stability margin drops to the unstable range in advance—that is, accurately outputting oscillation conditions. Simultaneously, by prioritizing the adjustment strategy with the largest stability margin decrease gradient, it can quickly locate the parameter adjustment methods and parameter boundaries most prone to oscillations, identify the key triggers for oscillations, and help technicians develop targeted prevention and control measures in advance. This avoids safety accidents such as equipment damage and grid interruption caused by oscillation expansion, improving the safety of power plant grid-connected operation.

[0182] 3. Optimize prediction efficiency, reduce prediction costs, and enhance the engineering applicability of the method. This method selects a set of key parameters by influencing factors, avoiding blindly traversing and analyzing all equipment parameters, significantly reducing the computational load and shortening the prediction cycle. Simultaneously, it adopts a hierarchical logic of "initial operating condition optimization - key parameter selection - adjustment strategy iteration," enabling oscillation condition prediction without complex field tests, effectively reducing the manpower and material costs of field testing. Furthermore, the method clearly defines the boundaries of parameter adjustment and the triggering conditions for oscillation conditions, with a clear operation process, facilitating engineering application and adapting to the grid-connected oscillation prediction needs of different types of power plants.

[0183] 4. Enhance the stability and controllability of power plant grid-connected operation and extend equipment lifespan. This method comprehensively traverses candidate regulation strategies, enabling it to fully capture oscillation risks under different parameter regulation methods, avoiding potential oscillation hazards caused by overlooking key regulation strategies. Simultaneously, by predicting oscillation conditions, technicians can optimize parameter regulation schemes in advance, rationally control the adjustment range and speed of key parameters, reduce the impact of oscillations on power plant equipment, lower equipment failure rates, and extend equipment lifespan. Furthermore, based on frequency domain stability analysis, the method can accurately quantify the changing patterns of stability margin, providing a scientific basis for stable control and parameter optimization after power plant grid connection, further enhancing the controllability and stability of power plant grid-connected operation.

[0184] 5. Adaptable to various operating conditions and equipment in power plant grid connection scenarios, with strong versatility. This method first constructs an admittance database, covering the admittance parameters and negative damping characteristics of each device in the power plant, enabling it to adapt to power plants with different equipment compositions and different grid connection operating conditions. Simultaneously, by iteratively updating the adjustment strategy and operating parameters, it can comprehensively cover oscillation conditions under different parameter combinations, eliminating the need for extensive adaptation modifications for specific power plants. This significantly improves the method's versatility and applicability, making it widely applicable to grid connection oscillation prediction in various new energy power plants and integrated energy power plants.

[0185] The beneficial effects of the optimized solution:

[0186] 1. Significantly improved modeling accuracy and more reliable stability analysis. This invention employs data-driven modeling for power electronic equipment, constructing a "condition-admittance" mapping model to accurately fit the frequency domain characteristics of the power electronic equipment; for passive equipment such as transmission lines and transformers, mechanistic modeling is used to ensure the accuracy of admittance characteristics. This fusion of "data-driven + mechanistic modeling" avoids the simplification errors of single mechanistic modeling for complex control characteristics and solves the problem of lack of physical constraints in single data-driven modeling, providing high-precision basic data for subsequent stability analysis.

[0187] 2. Significantly improved optimization efficiency, avoiding the curse of dimensionality. This invention quantifies the comprehensive influence weights of operating condition parameters through global and local sensitivity analysis, reducing high-dimensional operating condition parameters (e.g., 42 parameters from 20 wind turbines + 1 SVG) to 3-8 core candidate parameters, avoiding the exponential increase in computational load caused by full parameter traversal. Simultaneously, based on the gradient prediction model, the strategy with the fastest decrease in stability margin is located, reducing the number of iterations for oscillating operating conditions and significantly shortening training time. This solves the pain points of traditional methods—"inefficient optimization and difficulty in real-time application"—and meets the real-time requirements of engineering scheduling.

[0188] 3. Comprehensive coverage of oscillation risks, leaving no key operating conditions unaddressed. This invention scientifically selects the high-risk initial operating condition with the "maximum negative damping effect," ensuring that optimization begins in the core risk area; combined with strategy traversal, it achieves comprehensive coverage of all potential oscillation conditions within the operating condition constraints. Compared to traditional stochastic optimization methods, this solution can accurately capture oscillation scenarios corresponding to "critical combinations of high-weight parameters," avoiding omissions of key oscillation conditions, while filtering out invalid combinations of low-weight parameters, thus improving the targeting of risk assessment.

[0189] 4. Wide adaptability, compatible with multiple types of renewable energy power plants. The operating condition coding method of this invention can accurately identify any combination of operating conditions, facilitating rapid analysis of equipment operating status and standardized data management. The modeling method and optimization logic do not depend on specific renewable energy equipment types, and can flexibly adapt to complex power plants containing diverse equipment such as wind turbines, photovoltaics, energy storage, and SVG; it can dynamically balance the influence ratio of global and local weights; and it can be flexibly adjusted according to the scale of the power plant, solving the problems of "poor adaptability and difficulty in promotion" of traditional methods. It is suitable for grid connection stability analysis of renewable energy power plants with different capacities and different equipment combinations.

[0190] On the other hand, the present invention also provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned methods for predicting grid-connected oscillation conditions of power plants.

[0191] On the other hand, the present invention also provides a computer system, including a memory and a processor; the memory stores program code that can be executed by the processor; the program code is used to execute any of the above-mentioned methods for predicting the grid-connected oscillation conditions of power stations based on frequency domain stability analysis.

[0192] For example, the program code can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the program code in the computer system.

[0193] The computer system can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer system may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the computer system may also include input / output devices, network access devices, buses, etc.

[0194] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0195] The memory can be an internal storage unit of a computer system, such as a hard disk or RAM. It can also be an external storage device of the computer system, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory can include both internal and external storage units. The memory is used to store the program code and other programs and data required by the computer system. The memory can also be used to temporarily store data that has been output or will be output.

[0196] The aforementioned computer storage medium and computer system are created based on the aforementioned method for predicting grid-connected oscillation conditions of power stations. Their technical functions and beneficial effects will not be elaborated here. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0197] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for predicting grid-connected oscillation conditions of power stations based on frequency domain stability analysis, characterized in that, include: S1: Establish an admittance database and construct a station frequency domain model; S2: Based on the negative damping degree and frequency domain model of each device in the admittance database, select the operating condition with the highest and most stable negative damping degree as the current initial operating condition; S3: Based on the frequency domain model, calculate the influence factors of each parameter of each device on the stability margin under the current initial operating conditions, and determine the set of key parameters based on the influence factors; S4 includes: S41: Adjusting the numerical values ​​of each key parameter with small steps under different adjustment directions, calculating the corresponding stability margin change, and constructing a training dataset; S42: Using the numerical change of each key parameter in the training dataset as input and the stability margin change as output, constructing a stability margin gradient prediction model to predict the stability margin descent gradient of each key parameter under different adjustment directions; S43: Based on the prediction results, determining the adjustment strategy with the largest stability margin descent gradient as the current adjustment strategy. S5: According to the current adjustment strategy, continuously adjust each parameter in the key parameter set with the adjustment step size, and determine whether each updated working condition is stable. If it is, no action is taken; otherwise, output the oscillation working condition until the boundary of any key parameter is reached. Delete the current adjustment strategy from the candidate adjustment strategies, return to step S4, and reselect the adjustment strategy with the largest stability margin descent gradient as the current adjustment strategy. Continue until all candidate adjustment strategies are traversed.

2. The method for predicting grid-connected oscillation conditions of power stations according to claim 1, characterized in that, Step S2 includes: S21: Calculate the negative damping area of ​​each device based on the admittance data of each device in the admittance database; S22: Based on the negative damping zone area of ​​each device, and taking into account the capacity weight of each device in the station, the negative damping degree of each working condition is determined by weighting. S23: Arrange the operating conditions in descending order according to the degree of negative damping of each operating condition; determine whether the current operating condition is stable in order; if stable, take the current operating condition as the current initial operating condition; if oscillating, output the current operating condition as the oscillating operating condition, and update the next operating condition of the current operating condition in the descending order to the current operating condition, and perform stability judgment again; until the operating condition with the maximum degree of negative damping and stability is obtained as the current initial operating condition.

3. The method for predicting grid-connected oscillation conditions of power stations according to claim 1, characterized in that, Step S3, which involves calculating the influence factors of each parameter of each device on the stability margin, includes: S31: Calculate the participation factor of each piece of equipment in the station under the current operating conditions, and determine the global weight of each piece of equipment's impact on the stability margin; S32: Under the current operating conditions, adjust only a single parameter of a single device at a time, and determine the local weight of the impact of each parameter adjustment on the stability margin; S33: Calculate the influence factor of each parameter of each device on the stability margin based on the global weight of each device and the local weight of each parameter.

4. The method for predicting grid-connected oscillation conditions of power stations according to claim 3, characterized in that, Step S32 includes: S321: Set the adjustment step size of the parameter, and adjust the individual parameter of a single device in both the positive and negative directions according to the set adjustment step size, while keeping other parameters unchanged, to obtain the positive update condition under positive direction adjustment and the negative update condition under negative direction adjustment. S322: Perform stability modal analysis on the positive and negative update conditions respectively to determine the stability margins of the positive and negative update conditions. S323: Determine the local weight of the influence of each parameter adjustment on the stability margin based on the ratio of the difference between the stability margin of the positive update condition and the stability margin of the negative update condition to twice the adjustment step size.

5. The method for predicting grid-connected oscillation conditions of power stations according to claim 4, characterized in that, Step S321, the step of setting the adjustment step size of the parameters, includes: S3211: Determine the adjustment step threshold for each parameter based on its numerical adjustment range; S3212: Based on the adjustment step size threshold of each parameter, determine the first step size and the second step size for adjusting each parameter; wherein, the second step size is twice the first step size, and the second step size does not exceed the adjustment step size threshold. S3213: Adjust the corresponding parameters with the first step length and the second step length respectively to obtain the stability margin change rate under the first step length adjustment and the stability margin change rate under the second step length adjustment. S3214: Determine whether the change rate of stability margin under the second step size adjustment and the change rate of stability margin under the first step size adjustment are less than the set threshold. If so, set the first step size as the adjustment step size of the parameter; otherwise, return to step S3212 and redetermine the first step size and the second step size of each parameter adjustment in the direction of shrinkage.

6. The method for predicting grid-connected oscillation conditions of power stations according to claim 4, characterized in that, Step S32 also includes: S320: Between the first step size and the second step size, set several equal adjustment step sizes; under each set adjustment step size, execute steps S321-S323 in sequence to obtain the local weight of each parameter under each adjustment step size. S324: Based on the local weights of each parameter under each adjustment step, determine the final local weight of each parameter by weighting.

7. The method for predicting grid-connected oscillation conditions of power stations according to claim 4, characterized in that, In step S324, the formula for calculating the final local weight of each parameter using weighted average is as follows: ； ； ； ； in, For the first The first device The final local weights of each parameter For the first The calculated number of steps is given by the set adjustment step size. The first device Local weights of each parameter, For the first A base coefficient for setting the adjustment step size. For the first A safety factor for setting and adjusting the step size. For the first The contribution coefficient of a set adjustment step size For the first The first device The parameter of the first One adjustment step size For the first The first device The maximum adjustment step size for each parameter For the first The first device The minimum adjustment step size for each parameter.

8. The method for predicting grid-connected oscillation conditions of power stations according to claim 1, characterized in that, Step S3, the step of determining the set of key parameters based on the impact factors, includes: S34: Arrange the influence factors of each parameter of each device on the stability margin in descending order; and calculate the sum of the influence factors of all parameters of all devices as the global influence factor; S35: Select the first M' parameters, construct the proposed key parameter set, and calculate the sum of the influence factors of all parameters in the proposed key parameter set as the local influence factor; S36: Calculate whether the ratio of the local impact factor to the global impact factor exceeds the set threshold. If so, use the proposed key parameter set as the final key parameter set; otherwise, let M'+1 be the updated M' and return to step S35.

9. The method for predicting grid-connected oscillation conditions of power stations according to any one of claims 1 to 8, characterized in that, The stable margin gradient prediction model consists of: an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence. The input layer is used to input the changes in the values ​​of each device parameter in the candidate parameter set; The first hidden layer is used to focus on the mapping relationship between the change of a single parameter and the change of the stability margin, extract the linear correlation features and nonlinear sensitivity features of a single parameter, and obtain the original features; The second hidden layer is used to mine the synergistic gain and suppression effect of different key parameter adjustment combinations on the stability margin change based on the original features, and to extract the first-order interactive coupling features between parameters. The third hidden layer is used to fuse the original features with the first-order coupled features, expand the feature representation dimension, and extract the high-order nonlinear coupled features between key parameters. The output layer is used to output the stability margin variation based on the high-order nonlinear coupling characteristics.

10. A computer storage medium, characterized in that, It stores executable program code; the executable program code is used to execute the method for predicting grid-connected oscillation conditions of power plants according to any one of claims 1 to 9.

11. A computer system, characterized in that, It includes a memory and a processor; the memory stores program code that can be executed by the processor; the program code is used to execute the power station grid-connected oscillation prediction method according to any one of claims 1 to 9.

Images