Method, system and equipment for evaluating power supply capability of power system and medium

Abstract

The invention discloses a power supply capability evaluation method, system and device of a power system and a medium. The method comprises the following steps: reading parameters of a training sample; generating a sample matrix; executing repetitive load flow calculation based on the sample matrix; constructing a BASPCE model according to the sample matrix and a corresponding repeated load flow calculation result; reading randomly input parameters; generating a random input matrix; directly obtaining a repetitive power flow result based on the BASPCE; and calculating the power supply capability of the novel power system according to the repetitive power flow result. According to the method, the defect that the calculation amount of power supply capacity evaluation is large is overcome, the whole process of repeated power flow is approximated by utilizing the characteristics of the sparse orthogonal polynomial, the complexity of the power supply capacity evaluation process is remarkably reduced through the proposed BASPCE model, meanwhile, the accuracy of the result is not affected, and the method is suitable for large-scale popularization and application. And the power supply capability of a novel power system containing renewable energy sources such as wind and light can be efficiently and accurately evaluated.

Description

Technical Field

[0001] This invention relates to the field of power system power supply capacity assessment technology, and in particular to a method, system, equipment and medium for power system power supply capacity assessment. Background Technology

[0002] With the continuous expansion of the power grid, the uncertainties brought about by load growth and the large-scale grid connection of renewable energy have increased the randomness of the power system. Accurately calculating the effective load supply capacity of the power system is crucial, and sufficient load supply capacity is the foundation for ensuring the safe operation of the power system. Power supply capacity assessment can quantify the flexibility and safety margin of power system load growth, providing a basis for the installation of reactive power compensation devices, grid reconfiguration, and grid planning.

[0003] Several studies have been conducted on methods for calculating the power supply capacity of power systems. Continuous power flow (CPF) lacks flexibility and struggles to handle constraints and control strategies. Optimal power flow (OPF) can account for the various uncertainties of renewable energy sources, but its computational complexity is high. Therefore, to achieve a good balance between accuracy and feasibility in power supply capacity assessment, repetitive power flow (RPF), with its simple calculation process and high flexibility, is a more reasonable choice. Traditional repetitive power flow methods require extensive computer simulations to obtain the final power supply capacity assessment results. However, when many random input variables exist in the system (such as the uncertain output of renewable energy sources widely present in new power systems), it leads to high computational complexity and long computation time, thus hindering the further application of power supply capacity analysis in the field of electrical engineering. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, device, and medium for assessing the power supply capacity of a power system to solve the problems of long calculation time and high complexity of existing methods for assessing the power supply capacity of power systems containing renewable energy.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for evaluating the power supply capacity of a power system, comprising:

[0008] Obtain the training sample configuration parameters, and generate a random sample matrix based on the configuration parameters as the input dataset for the surrogate model;

[0009] The random sample matrix is ​​input into the power system repetitive power flow calculation model, and multiple power flow simulations are performed to obtain the output dataset of the training samples.

[0010] Based on the input and output datasets, an adaptive sparse chaotic polynomial expansion model is constructed, and the expansion coefficients of the model are solved by the sparse regression algorithm. The optimal hyperparameters of the model are determined by the cross-validation algorithm, and the trained adaptive sparse chaotic polynomial expansion model is obtained.

[0011] Read the random input configuration parameters under the actual evaluation scenario, and generate a new random input matrix based on the actual input configuration parameters;

[0012] An adaptive sparse chaotic polynomial expansion model is used to replace the repetitive power flow calculation model of the power system. The new random input matrix is ​​input into the trained adaptive sparse chaotic polynomial expansion model, and the power supply capacity assessment result is directly output.

[0013] As a preferred embodiment of the power system supply capacity assessment method of the present invention, the output dataset for obtaining the training samples includes:

[0014] Obtain the load, wind power, and photovoltaic output parameters of the current sampling point, and use these as initial conditions to perform power flow calculations and determine the reference value of system loss;

[0015] Calculate the first type of power imbalance caused by power prediction error;

[0016] To address the power imbalance caused by power prediction errors, the first power imbalance allocation method is adopted, which distributes the unbalanced power to each balancing node in the multi-balancing node group and performs power flow calculation.

[0017] According to the preset load growth vector and initial step size, the load of all systems is increased synchronously;

[0018] Calculate the second type of power imbalance caused by load growth;

[0019] To address the power imbalance caused by load growth, a second power imbalance allocation method is adopted, which distributes the unbalanced power to each balancing node in the multi-balancing node group and performs power flow calculation.

[0020] Based on system safety constraints and convergence conditions, the system's ultimate load is determined, and the power supply capacity value of the current sampling point is calculated accordingly to obtain the output dataset of the training samples.

[0021] The beneficial effect of this preferred technical solution is that by distinguishing between two types of imbalance sources, namely power prediction error and load growth, and adopting a differentiated multi-balance node allocation strategy, it can more realistically simulate the operating response of the new power system under uncertain disturbances and load evolution, thereby improving the physical rationality and accuracy of power supply capacity assessment.

[0022] As a preferred embodiment of the power system supply capacity assessment method of the present invention, wherein: the obtained trained adaptive sparse chaotic polynomial expansion model includes:

[0023] Identify multiple input random variables that affect power supply capacity and their respective probability distributions;

[0024] For each dimension of the input random variable, select a univariate orthogonal polynomial basis function that is orthogonal to it;

[0025] By using tensor product operations, the univariate orthogonal polynomials are combined into multivariate polynomial basis functions;

[0026] A hyperbolic truncation criterion is adopted to generate a candidate set of multivariate basis functions based on the preset maximum expansion order;

[0027] Based on the leave-one-out cross-validation algorithm, the optimal truncation order is adaptively selected;

[0028] Based on the final set of basis functions determined by the optimal truncation order, a mathematical framework for an adaptive sparse chaotic polynomial expansion model is constructed.

[0029] Based on the mathematical framework and the input-output dataset of the training samples, the expansion coefficients of the adaptive sparse chaotic polynomial expansion model are solved by the sparse regression algorithm.

[0030] Substituting the expansion coefficients obtained from the solution into the mathematical framework, we obtain the trained adaptive sparse chaotic polynomial expansion model.

[0031] The beneficial effects of this preferred technical solution are that by adaptively constructing an orthogonal polynomial basis function system based on the probability distribution of the input variables, and by combining hyperbolic truncation and leave-one-out cross-validation to select the optimal model complexity, and then using sparse regression to solve the coefficients, the computational burden in the propagation of high-dimensional uncertainty can be significantly reduced while ensuring high-precision approximation, thereby achieving efficient and automatic construction of the power supply capacity proxy model.

[0032] As a preferred embodiment of the power system power supply capacity assessment method described in this invention, the first type of power imbalance allocation includes determining the power imbalance allocation coefficient based on the proportion of the remaining capacity of each generator in the multi-balance node group.

[0033] The second type of power imbalance allocation involves determining the power imbalance allocation coefficient based on the ratio of the inverse of the power generation cost of each generator in the multi-balance node group.

[0034] As a preferred embodiment of the power system power supply capacity assessment method described in this invention, the adaptive selection of the optimal truncation order specifically involves: traversing different maximum expansion orders, calculating the leave-one-out cross-validation error corresponding to each order, and selecting the maximum expansion order with the smallest error as the optimal truncation order.

[0035] As a preferred embodiment of the power system supply capacity assessment method of the present invention, the step of determining the system limit load based on system security constraints and convergence conditions includes:

[0036] Determine whether the system violates security constraints;

[0037] If safety constraints are violated, load growth will be stopped.

[0038] If no safety constraints are violated, determine whether the current load growth step size is less than the preset convergence threshold.

[0039] If the current load growth step size is less than the preset convergence threshold, then load growth will stop.

[0040] If the current load growth step size is greater than or equal to the preset convergence threshold, the load growth step size will be halved and the load growth and power flow calculation will continue.

[0041] As a preferred embodiment of the power system supply capacity assessment method of the present invention, the calculation of the supply capacity value includes:

[0042] The power supply capacity value of the current sampling point is calculated based on the ratio of the difference between the system load level when load growth stops and the initial load level, relative to the initial load level.

[0043] Secondly, the present invention provides a power system supply capacity assessment system, comprising:

[0044] The training sample generation module is used to obtain training sample configuration parameters and generate a random sample matrix based on the configuration parameters as the input dataset for the surrogate model.

[0045] The power supply capacity label generation module is used to input the random sample matrix into the power system repeated power flow calculation model, perform multiple power flow simulations, and obtain the output dataset of training samples.

[0046] The proxy model training module is used to construct an adaptive sparse chaotic polynomial expansion model based on the input dataset and the output dataset, solve the expansion coefficients of the model through the sparse regression algorithm, and determine the optimal hyperparameters of the model by combining the cross-validation algorithm to obtain the trained adaptive sparse chaotic polynomial expansion model.

[0047] The evaluation scenario input construction module is used to read the random input configuration parameters under the actual evaluation scenario and generate a new random input matrix based on the actual input configuration parameters.

[0048] The power supply capacity rapid assessment module is used to replace the power system repetitive power flow calculation model with an adaptive sparse chaotic polynomial expansion model. The new random input matrix is ​​input into the trained adaptive sparse chaotic polynomial expansion model, and the power supply capacity assessment result is directly output.

[0049] Thirdly, the present invention provides an electronic device, comprising:

[0050] Memory, used to store programs;

[0051] A processor for executing the computer-executable instructions, which, when executed by the processor, implement the steps of the power system power supply capacity assessment method.

[0052] Fourthly, the present invention provides a computer-readable storage medium comprising: when the program is executed by a processor, the steps of implementing the power system power supply capacity assessment method.

[0053] The beneficial effects of this invention are as follows: This invention selects corresponding orthogonal polynomial basis functions for the specific probability distributions of input random variables such as load, wind power, and photovoltaics, and constructs a candidate basis function set using tensor product and hyperbolic truncation. Then, it adaptively determines the optimal expansion order using leave-one-out cross-validation, achieving automatic optimization of the surrogate model structure. This effectively suppresses the "curse of dimensionality" while ensuring approximation accuracy and improving the model's generalization ability. Furthermore, by solving the chaotic polynomial expansion coefficients using a sparse regression algorithm, only basis function terms that significantly affect the output response are retained, further reducing model complexity and improving computational efficiency and robustness. Finally, by constructing an adaptive sparse chaotic polynomial expansion model (BASPCE) to replace traditional repetitive power flow calculations, the power supply capacity value is directly and quickly output during the evaluation phase, realizing a shift from time-consuming deterministic simulation to efficient probabilistic evaluation, significantly improving the speed and scalability of power supply capacity evaluation for new power systems under multiple uncertainties. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0055] Figure 1 This is a schematic diagram of the basic process of a power system power supply capacity assessment method provided in one embodiment of the present invention. Detailed Implementation

[0056] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0057] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for assessing the power supply capacity of a power system is provided, comprising:

[0058] S100: Obtain the training sample configuration parameters, and generate a random sample matrix based on the configuration parameters as the input dataset for the surrogate model;

[0059] S200: Input the random sample matrix into the power system repetitive power flow calculation model, perform multiple power flow simulations, and obtain the output dataset of the training samples;

[0060] S300: Based on the input dataset and output dataset, construct an adaptive sparse chaotic polynomial expansion model, solve the expansion coefficients of the model through a sparse regression algorithm, and determine the optimal hyperparameters of the model by combining a cross-validation algorithm to obtain the trained adaptive sparse chaotic polynomial expansion model.

[0061] S400: Read the random input configuration parameters under the actual evaluation scenario, and generate a new random input matrix based on the actual input configuration parameters;

[0062] S500: An adaptive sparse chaotic polynomial expansion model is used to replace the repetitive power flow calculation model of the power system. The new random input matrix is ​​input into the trained adaptive sparse chaotic polynomial expansion model, and the power supply capacity evaluation result is directly output.

[0063] It should be noted that existing power supply capacity assessment methods for power systems containing renewable energy face multiple challenges during operation. These include the strong randomness and volatility of renewable energy output from sources such as wind and solar power, leading to high uncertainty in the system input state. Traditional deterministic assessment methods struggle to accurately characterize the probability distribution and risk level of power supply capacity. Furthermore, system power flow distribution and power balance are coupled by various uncertainties. If Monte Carlo methods combined with repeated power flow simulations are used for probabilistic assessment, tens of thousands of power flow calculations are required, resulting in enormous computational time and failing to meet online or near-real-time assessment needs. Simultaneously, when simulating power imbalances caused by load growth and renewable energy fluctuations, a single balancing node or fixed allocation strategy is often used, failing to distinguish between the different physical mechanisms of prediction error adjustment and load evolution response. This leads to power flow solutions deviating from actual scheduling behavior, affecting the physical rationality of the assessment results. Moreover, high-dimensional random input spaces are prone to the "curse of dimensionality," causing traditional chaotic polynomial expansion methods to fail due to the exponential increase in the number of basis functions, making it difficult to balance accuracy and efficiency and limiting the feasibility of large-scale probabilistic assessments of complex systems.

[0064] Therefore, existing methods for assessing the power supply capacity of power systems containing renewable energy sources suffer from problems such as long calculation times and high complexity in repetitive power flow calculations. This invention addresses these issues through steps S100-S500: reading parameters from training samples; generating a sample matrix; performing repetitive power flow calculations based on the sample matrix; constructing a BASPCE model based on the sample matrix and the corresponding repetitive power flow calculation results; reading randomly input parameters; generating a random input matrix; directly obtaining the repetitive power flow results based on BASPCE; and calculating the power supply capacity of the new power system based on the repetitive power flow results. This invention overcomes the computational burden of power supply capacity assessment by utilizing the characteristics of sparse orthogonal polynomials to approximate the entire repetitive power flow process. The proposed BASPCE model significantly reduces the complexity of the power supply capacity assessment process without affecting the accuracy of the results, thus contributing to the efficient and accurate assessment of the power supply capacity of new power systems containing renewable energy sources such as wind and solar.

[0065] Example 2, this is an embodiment of the present invention, which provides a method for evaluating the power supply capacity of a power system based on the previous embodiment, including:

[0066] In this embodiment, the training sample configuration parameters include dimension m, the type of probability distribution, and the parameters of the probability distribution function. The probability distribution function includes uniform distribution, normal distribution, Gamma distribution, and Beta distribution. The sample values ​​include predicted output for load levels, wind power, and photovoltaic power.

[0067] In this embodiment of the application, a random sample matrix is ​​generated. The matrix is ​​m-dimensional, with each dimension containing There are 100 training samples. Each sample in this matrix is ​​generated according to a certain probability distribution function in S100.

[0068] In this embodiment of the application, the output dataset of the training samples obtained in step S200 includes:

[0069] Obtain the load, wind power, and photovoltaic output parameters at the current sampling point, and use these as initial conditions to perform power flow calculations and determine system losses. Reference value;

[0070] In this embodiment of the application, the random sample generation method includes using Monte Carlo sampling to independently extract samples from a specified probability distribution (such as normal, Gamma, or Beta distribution) of variables such as load, wind power, and photovoltaics, and generate a training sample matrix.

[0071] In an optional implementation, the random sample generation method in step S100 can further divide the cumulative distribution function of each dimension of the input random variable into several equal parts based on the specific probability distribution of each dimension. Intervals are used to randomly select a sample from each interval and map it back to the original space, generating a more uniformly distributed sample. * Sample matrix This serves as the input dataset for the surrogate model to improve training efficiency and coverage.

[0072] In an optional implementation, the random sample generation method in step S100 can further be tailored to the specific probability distribution of each input random variable. First, a pseudo-random point set of m-dimensional Sobol sequences with low dissimilarity is generated. Then, it is mapped to the corresponding distribution's sample space through probability integral transformation, forming a highly uniform distribution. * Sample matrix The dataset is used as input to the surrogate model to accelerate convergence and improve the efficiency of high-dimensional sampling.

[0073] Calculate the first type of power imbalance caused by power prediction error;

[0074] To address the power imbalance caused by power prediction errors, the first type of power imbalance allocation is adopted. The unbalanced power is distributed to each balancing node in the multi-balancing node group, and power flow calculation is performed.

[0075] According to the preset load growth vector and initial step size, the load of all systems is increased synchronously;

[0076] Calculate the second type of power imbalance caused by load growth;

[0077] To address the power imbalance caused by load growth, a second type of power imbalance allocation is adopted. The unbalanced power is distributed to each balancing node in the multi-balancing node group, and power flow calculation is performed.

[0078] Based on system safety constraints and convergence conditions, the system's ultimate load is determined, and the power supply capacity value of the current sampling point is calculated accordingly to obtain the output dataset of the training samples.

[0079] In this embodiment of the application, the power supply capability value (ALSC) Defined as:

[0080]

[0081] in, busbar Increased load, busbar The basic load.

[0082] Where N is the number of buses; The active power of the generator at bus i; The active power of the load on bus i; The voltage amplitude of bus i; The phase angle difference between bus i and bus j and These are the real and imaginary parts of the nodal admittance matrix, respectively.

[0083] In this embodiment of the application, the calculation steps for solving the power supply capability value using RPF in step S200 are as follows:

[0084] (1) Perform power flow calculations under the average value of all random input variables to obtain system losses. The reference value, the first type of power imbalance distribution is expressed as:

[0085]

[0086] in, Represents the power output of all conventional generators; This represents the power output of all renewable energy generators. Power imbalance is determined based on the proportion of remaining capacity of each generator. Allocation ratio among idle busbars .

[0087] (2) Based on the current time The load, wind power, and photovoltaic sampling points are given. The system has N buses, multiple balancing node groups B, and traditional generating bus groups. Renewable energy generation bus group N, first part of the power imbalance for:

[0088]

[0089] in, Represents the power output of all conventional generators; This represents the power output of all renewable energy generators. Power imbalance is determined based on the proportion of remaining capacity of each generator. Allocation ratio among idle busbars Represented as:

[0090]

[0091] in, Let be the remaining power of the generator on the i-th bus.

[0092] (3) The active power equation of the equilibrium node is transformed into:

[0093]

[0094] Perform power flow calculations.

[0095] (4) Implement load increase, The second part is power imbalance for:

[0096]

[0097] in, It is the load quantity. Elements in the load growth vector b.

[0098] (5) Determine the power imbalance based on the ratio of the reciprocals of the costs of each generator. Allocation ratio among idle busbars for

[0099] ; ;

[0100] in, It is the cost of the generator at bus i.

[0101] (6) The active power equation at the equilibrium node is transformed into:

[0102]

[0103] Perform power flow calculations.

[0104] (7) Check whether the following constraints have been violated ( (For the power flow equation)

[0105]

[0106] If a constraint violation exists, then Then, subtract the corresponding allocated power from the generator at the equilibrium node, and proceed to calculate ALSC. Otherwise, perform a check to see if h is less than the convergence condition.

[0107] (8) Check if h is less than the convergence condition. Yes, proceed with the calculation of the next time step. ALSC. Otherwise, halve step h, i.e.: Then, the power imbalance is determined based on the ratio of the inverse of the cost of each generator. The allocation ratio among idle busbars.

[0108] (9) Calculate ALSC;

[0109] (10) Switch to the first part of the power imbalance and calculate the next moment. ALSC.

[0110] In this embodiment of the application, the second type of power imbalance allocation in step S200 includes determining the power imbalance allocation coefficient based on the ratio of the reciprocals of the power generation costs of each generator in the multi-balance node group. .

[0111] In this embodiment of the application, step S200, which determines the system's ultimate load based on system security constraints and convergence conditions, includes:

[0112] Determine whether the system violates security constraints;

[0113] If safety constraints are violated, load growth will be stopped.

[0114] If no safety constraints are violated, determine whether the current load growth step size is less than the preset convergence threshold.

[0115] If the current load growth step size is less than the preset convergence threshold, then load growth will stop.

[0116] If the current load growth step size is greater than or equal to the preset convergence threshold, the load growth step size will be halved and the load growth and power flow calculation will continue.

[0117] In this embodiment of the application, the calculation of the power supply capacity value in step S200 includes:

[0118] The power supply capacity value of the current sampling point is calculated based on the ratio of the difference between the system load level when load growth stops and the initial load level, relative to the initial load level.

[0119] In this embodiment, the surrogate model construction method includes selecting the corresponding univariate orthogonal polynomial basis functions based on the probability distribution of the input random variables, generating a candidate basis function set through tensor product and hyperbolic truncation, adaptively determining the optimal truncation order using leave-one-out cross-validation, and solving the expansion coefficients by combining a sparse regression algorithm, and finally constructing a high-precision, sparse adaptive chaotic polynomial expansion surrogate model.

[0120] In an optional implementation, the surrogate model construction method in step S300 can also be based on the input-output dataset of the training samples, select an appropriate covariance function (such as a Gaussian kernel or a Matérn kernel), solve the model hyperparameters by the maximum likelihood estimation method, and construct an interpolation model to achieve accurate prediction of power supply capacity, thereby forming a Kriging surrogate model that replaces repeated power flow calculation.

[0121] In an optional implementation, the proxy model construction method in step S300 can also construct a neural network model with a multi-layer fully connected structure based on the input-output dataset of the training samples. By optimizing the loss function through backpropagation algorithm and gradient descent, the network weights and biases are trained, thereby learning the nonlinear mapping relationship between the input random variables (load, wind power, photovoltaic output) and the power supply capacity output, forming a deep proxy model that can replace repeated power flow calculations.

[0122] In this embodiment of the application, step S300, which yields the trained adaptive sparse chaotic polynomial expansion model, includes:

[0123] Identify multiple input random variables that affect power supply capacity and their respective probability distributions;

[0124] For each dimension of the input random variable, select a univariate orthogonal polynomial basis function that is orthogonal to it;

[0125] By using tensor product operations, the univariate orthogonal polynomials are combined into multivariate polynomial basis functions;

[0126] A hyperbolic truncation criterion is adopted to generate a candidate set of multivariate basis functions based on the preset maximum expansion order;

[0127] Based on the leave-one-out cross-validation algorithm, the optimal truncation order is adaptively selected;

[0128] Based on the final set of basis functions determined by the optimal truncation order, a mathematical framework for constructing an adaptive sparse chaotic polynomial expansion model is established.

[0129] Based on the mathematical framework and the input-output dataset of the training samples, the expansion coefficients of the adaptive sparse chaotic polynomial expansion model are solved by the sparse regression algorithm.

[0130] Substituting the expansion coefficients obtained from the solution into the mathematical framework, we obtain the trained adaptive sparse chaotic polynomial expansion model.

[0131] In this embodiment, the optimal truncation order selection method includes constructing a corresponding chaotic polynomial basis function set for each candidate order and solving the model coefficients for a preset maximum expansion order range, calculating the prediction error of each model using leave-one-out cross-validation, and finally selecting the order with the smallest prediction error as the optimal truncation order to achieve an adaptive balance between model complexity and generalization ability.

[0132] In an optional implementation, the optimal truncation order selection method in step S300 can also construct chaotic polynomial expansion models for different candidate truncation orders, calculate the generalized cross-validation scores of each model using training data, and select the order with the smallest GCV score as the optimal truncation order to efficiently evaluate the model fit and avoid repeated subset validation, thereby achieving adaptive selection of model complexity.

[0133] In an optional implementation, the optimal truncation order selection method in step S300 can also construct a corresponding chaotic polynomial basis function set and solve the model coefficients for each candidate maximum expansion order. By calculating the generalized cross-validation (GCV) score of each model, the order with the smallest GCV value is selected as the optimal truncation order, so as to achieve efficient evaluation and adaptive selection of the model's generalization ability while avoiding repeated training.

[0134] In this embodiment of the application, the implementation steps of the sparse regression algorithm using the LASSO method include constructing a mathematical framework for chaotic polynomial expansion based on the set of basis functions with the optimal truncation order, using the input-output dataset of the training samples, and automatically compressing the expansion coefficients of irrelevant basis functions to zero by solving the least squares optimization problem with L1 regularization terms, thereby obtaining a sparse and high-precision surrogate model coefficient solution.

[0135] In an optional implementation, the sparse regression algorithm in step S300 can also employ a forward distribution regression method, which includes constructing a chaotic polynomial expansion model based on a set of basis functions determined by the optimal truncation order, using the input-output dataset of the training samples, starting from zero coefficients, iteratively selecting the basis functions with the highest correlation to the current residual with a minimal step size for coefficient updates, gradually approximating the sparse solution until convergence, thereby obtaining the expansion coefficients with sparsity.

[0136] In an optional implementation, the sparse regression algorithm in step S300 can also employ the minimum angle regression (LAR) method, which includes constructing a chaotic polynomial expansion model based on the set of basis functions determined by the optimal truncation order. Using the input-output dataset of the training samples, starting from zero coefficients, the algorithm selects the basis function with the highest correlation to the current residual each time, and synchronously updates the expansion coefficients of multiple related basis functions along the direction that is equiangular with the residual, thereby gradually constructing a coefficient path and efficiently solving for sparse and stable chaotic polynomial expansion coefficients.

[0137] In this embodiment of the application, the construction of the adaptive sparse chaotic polynomial expansion model (BASPCE) includes the following steps:

[0138] Assumption It is an m-dimensional independent input random vector. Based on the theory of generalized chaotic polynomial expansion, we can choose appropriate orthogonal polynomials corresponding to each input random variable, and then approximate the model using the expansions of these orthogonal polynomials. The output response, expressed in the generalized chaotic polynomial expansion, is:

[0139]

[0140] in, This represents a set of truncation schemes contained within the m-dimensional natural number sphere. Let α represent the basis functions of a multivariate polynomial, and let α be the basis function used to identify the basis functions. Multiple subscripts, γ α are basis functions The corresponding expansion coefficients. The truncation scheme Θ is used to achieve finite term truncation of the approximation model.

[0141] For multivariate polynomial basis functions It consists of tensor products of basis functions of univariate orthogonal polynomials:

[0142]

[0143] In the formula, Representing variables The corresponding number orthogonal polynomial of order 1.

[0144] For a random variable whose probability density function is known and which follows any distribution... The corresponding univariate orthogonal polynomial can be constructed using the following recurrence relation. :

[0145]

[0146] Will with variables probability density function Orthogonal functions are denoted as ,but and satisfy:

[0147]

[0148] Furthermore, the coefficient and It can be obtained from the following formula:

[0149]

[0150] The expansion terms in the generalized chaotic polynomial expansion are truncated according to a truncation criterion parameter p, so that the total number of non-zero terms in the truncated PCE model expansion is:

[0151]

[0152] Error is cross-validated using the leave-one-out (LOO) method as shown below. To select the most suitable p-value:

[0153]

[0154] in, This represents the new set of samples obtained by removing the l-th sample from the training samples. The PCE model is thus constructed. From the formula for selecting the most suitable p-value, it can be seen that for the error... To perform the calculation, we need to obtain... indivual Model.

[0155] In this embodiment of the application, the adaptive selection of the optimal truncation order in step S300 specifically involves: traversing different maximum expansion orders, calculating the leave-one-out cross-validation error corresponding to each order, and selecting the maximum expansion order with the smallest error as the optimal truncation order.

[0156] In this embodiment, the random input matrix in step S400 includes dimension m, the type of probability distribution, and parameters of the probability distribution function. The probability distribution function includes uniform distribution, normal distribution, Gamma distribution, and Beta distribution. Load and photovoltaic power plant output are normally distributed, while wind farm output data are Beta distributed. The random input includes the load level, and the predicted output of wind and photovoltaic power.

[0157] In this embodiment of the application, the random input matrix is ​​generated in step S400. The matrix is ​​m-dimensional, with each dimension containing There are 10 training samples. Each value in this matrix is ​​generated according to a certain probability distribution function in S400.

[0158] Example 3 is an embodiment of the present invention. This embodiment provides a method for evaluating the power supply capacity of a power system. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through specific implementation methods and implementation effects.

[0159] The specific details of this embodiment are as follows:

[0160] The example is applied to the IEEE 39-bus standard system, where buses 31, 35, 36, 37, 38, and 39 are balancing nodes, and the photovoltaic power plant and wind farm are located at nodes 30 and 34, respectively.

[0161] In step S1, the parameters of the training samples include dimension m, the type of probability distribution, and the parameters of the probability distribution function. The load and photovoltaic power plant output follow a normal distribution, while the wind farm output data follows a Beta distribution. The sample values ​​include the load level, and the predicted output of wind and photovoltaic power.

[0162] In step S2, a random sample matrix is ​​generated. The matrix has 23 dimensions, with each dimension containing =4000 training samples. Each sample in this matrix is ​​generated according to a certain probability distribution function in step S100. The data in this step is only used as model input data in the training samples of the subsequent surrogate model (adaptive chaotic multinomial expansion).

[0163] In step S3, based on M ED Perform repeated power flow (RPF).

[0164] In step S4, an adaptive sparse chaotic polynomial expansion model (BASPCE) is constructed.

[0165] For a given input random variable, the choice of its corresponding orthogonal polynomial basis function is related to the probability distribution type of this variable. The relationship between several classic input random variable probability distribution types and their corresponding optimal univariate orthogonal polynomial basis functions is shown in Table 1.

[0166] Table 1. Correspondence between variable distribution types and optimal univariate orthogonal polynomial basis functions.

[0167]

[0168] Given a target accuracy error (By default setting of 0), the approach to establishing the BASPCE approximation model can be summarized as follows:

[0169] ① Set initial values ​​according to the hyperbolic truncation scheme And select the corresponding basis function based on the input variables;

[0170] ② Based on the current settings, calculate the expansion coefficients in the PCE model and calculate e at this point. LOO value;

[0171] ③ Compare the error value with the preset threshold e T Compare. If ≤ If the algorithm fails, stop the algorithm and return to the current state. The PCE model with the smallest error value is selected; otherwise, p = p + 1 is set, and the process returns to step ②.

[0172] In summary, the BASPCE method can select the most suitable value of p from a series of possible values ​​based on the optimization condition of minimizing cross-validation error. In other words, the BASPCE method can effectively drive the maximum value of p directly from the experimental design data to obtain the most suitable approximation model (i.e., the surrogate model). In this example, p is solved to be 9.

[0173] In step S5, the randomly input parameters are read. The load and photovoltaic power plant output follow a normal distribution, while the wind farm output data follows a Beta distribution. Sample values ​​include the load level, and the predicted output of wind and photovoltaic power.

[0174] In step S6, a random input matrix M is generated. MCS The matrix has m=23 dimensions, with N elements in each dimension. MCS =8760 random input samples. Each value in this matrix is ​​generated according to a probability distribution function described in step S5. This random input matrix is ​​the input data for practical applications of the adaptive chaotic multinomial expansion model.

[0175] In step S7, BASPCE is used to replace the repetitive power flow, and the random input matrix M is... MCS The output of the BASPCE model is used as the power system supply capacity assessment result, serving as the input to the BASPCE model.

[0176] Example 4 is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a power system power supply capacity assessment system.

[0177] It should be noted that the technical solution of the power system power supply capacity assessment system is based on the same concept as the technical solution of the power system power supply capacity assessment method described above. For details not described in detail in the technical solution of the power system power supply capacity assessment system in this embodiment, please refer to the description of the technical solution of the power system power supply capacity assessment method described above.

[0178] This embodiment provides a power system supply capacity assessment system, comprising:

[0179] The training sample generation module is used to obtain training sample configuration parameters and generate a random sample matrix based on the configuration parameters as the input dataset for the surrogate model.

[0180] The power supply capacity label generation module is used to input the random sample matrix into the power system repeated power flow calculation model, perform multiple power flow simulations, and obtain the output dataset of training samples.

[0181] The proxy model training module is used to construct an adaptive sparse chaotic polynomial expansion model based on the input dataset and the output dataset, solve the expansion coefficients of the model through the sparse regression algorithm, and determine the optimal hyperparameters of the model by combining the cross-validation algorithm to obtain the trained adaptive sparse chaotic polynomial expansion model.

[0182] The evaluation scenario input construction module is used to read the random input configuration parameters under the actual evaluation scenario and generate a new random input matrix based on the actual input configuration parameters.

[0183] The power supply capacity rapid assessment module is used to replace the power system repetitive power flow calculation model with an adaptive sparse chaotic polynomial expansion model. The new random input matrix is ​​input into the trained adaptive sparse chaotic polynomial expansion model, and the power supply capacity assessment result is directly output.

[0184] This embodiment also provides an electronic device applicable to a power system power supply capacity assessment method, including:

[0185] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a power system power supply capacity assessment method as described in the above embodiments.

[0186] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for evaluating the power supply capacity of a power system as proposed in the above embodiments.

[0187] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for evaluating the power supply capacity of a power system proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0188] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0189] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for power system supply capability assessment, characterized by, The method comprises the following steps: obtain training sample configuration parameters, generate a random sample matrix based on the configuration parameters as an input data set of a proxy model; input the random sample matrix into a power system repetitive power flow calculation model, perform multiple power flow simulations, and obtain an output data set of the training sample; based on the input data set and the output data set, construct an adaptive sparse chaotic polynomial expansion model, and solve the expansion coefficients of the model by a sparse regression algorithm, determine the optimal hyperparameters of the model by a cross-validation algorithm, and obtain the trained adaptive sparse chaotic polynomial expansion model; read random input configuration parameters in an actual evaluation scenario, generate a new random input matrix based on the actual input configuration parameters; use the adaptive sparse chaotic polynomial expansion model to replace the power system repetitive power flow calculation model, input the new random input matrix into the trained adaptive sparse chaotic polynomial expansion model, and directly output the power supply capacity evaluation result.

2. The power system supply capability assessment method of claim 1, wherein: The output data set of the training sample comprises: obtain the load, wind power and photovoltaic output parameters of the current sampling point, and perform power flow calculation based on the initial conditions to determine the reference value of system loss; calculate the first type of power imbalance caused by power prediction error; for the power imbalance caused by power prediction error, use the first type of power imbalance distribution to distribute the imbalance power to each balancing node in the multi-balancing node group, and perform power flow calculation; according to the preset load growth vector and initial step length, synchronously grow all the loads of the system; calculate the second type of power imbalance caused by load growth; for the power imbalance caused by load growth, use the second type of power imbalance distribution to distribute the imbalance power to each balancing node in the multi-balancing node group, and perform power flow calculation; based on the system safety constraints and convergence conditions, determine the system limit load, and calculate the power supply capacity value of the current sampling point based on the system limit load, to obtain the output data set of the training sample.

3. The power system supply capability assessment method of claim 1 or 2, wherein: The trained adaptive sparse chaotic polynomial expansion model comprises: determine a plurality of input random variables affecting the power supply capacity and their respective probability distributions; for a specific probability distribution of each dimension input random variable, select a one-dimensional orthogonal polynomial basis function orthogonal to the specific probability distribution of each dimension input random variable; combine each one-dimensional orthogonal polynomial into a multi-dimensional polynomial basis function through tensor product operation; generate a candidate multi-dimensional basis function set according to the preset maximum expansion order by using a hyperbolic type truncation criterion; adaptively select the optimal truncation order based on the leave-one-out cross-validation algorithm; based on the final basis function set determined by the optimal truncation order, construct the mathematical framework of the adaptive sparse chaotic polynomial expansion model; based on the mathematical framework and the input-output data set of the training sample, solve each expansion coefficient of the adaptive sparse chaotic polynomial expansion model by a sparse regression algorithm; substitute the solved expansion coefficient into the mathematical framework to obtain the trained adaptive sparse chaotic polynomial expansion model.

4. The power system supply capability assessment method of claim 3, wherein: The first type of power imbalance distribution comprises determining a power imbalance distribution coefficient according to the residual capacity proportion of each generator in the multi-balancing node group; The second power imbalance distribution comprises determining a power imbalance distribution coefficient according to the ratio of the inverse of the power generation cost of each generator in the multi-balance node group.

5. The power system supply capability assessment method of claim 4, wherein: The adaptive selection of the optimal truncation order comprises: traversing different maximum expansion orders, calculating the leave-one-out cross-validation error corresponding to each order, and selecting the maximum expansion order with the minimum error as the optimal truncation order.

6. The power system supply capability assessment method of claim 5, wherein: The determination of the system limit load based on the system security constraints and the convergence condition comprises: judging whether the system violates the security constraints; if the security constraints are violated, stopping the load growth; if the security constraints are not violated, judging whether the current load growth step is less than the preset convergence threshold; if the current load growth step is less than the preset convergence threshold, stopping the load growth; if the current load growth step is greater than or equal to the preset convergence threshold, halving the load growth step and continuing to perform the load growth and power flow calculation.

7. The power system supply capability assessment method of claim 6, wherein: The calculation of the power supply capacity value comprises: calculating the power supply capacity value of the current sampling point according to the ratio of the difference between the system load level when the load growth is stopped and the initial load level to the initial load level.

8. A power system supply capability assessment system applying the method of any of claims 1-7, characterized by It comprises: a training sample generation module configured to obtain training sample configuration parameters, generate a random sample matrix based on the configuration parameters, and use the random sample matrix as an input data set of a proxy model; a power supply capacity label generation module configured to input the random sample matrix into a power system repetitive power flow calculation model, perform multiple power flow simulations, and obtain an output data set of the training sample; a proxy model training module configured to construct an adaptive sparse chaotic polynomial expansion model based on the input data set and the output data set, solve the expansion coefficients of the model through a sparse regression algorithm, determine the optimal hyperparameters of the model through a cross-validation algorithm, and obtain a trained adaptive sparse chaotic polynomial expansion model; an evaluation scenario input construction module configured to read random input configuration parameters in an actual evaluation scenario, and generate a new random input matrix based on the actual input configuration parameters; a power supply capacity rapid evaluation module configured to replace the power system repetitive power flow calculation model with the adaptive sparse chaotic polynomial expansion model, input the new random input matrix into the trained adaptive sparse chaotic polynomial expansion model, and directly output a power supply capacity evaluation result.

9. An electronic device, comprising: It comprises: a memory for storing a program; a processor for loading the program to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, The program is executed by the processor to implement the steps of the method according to any one of claims 1-7. The program is executed by the processor to implement the steps of the method according to any one of claims 1-7.

Images