Differential granularity data driving-based reliability evaluation method for power distribution network containing wind power

Abstract

The invention relates to a reliability evaluation method for a power distribution network containing wind power based on differential granularity data driving, and belongs to the technical field of intelligent power grids. When the reliability of a wind power generation system is evaluated, differential sampling needs to be executed, and a large amount of data needs to be accurately predicted. Existing methods often limit original data within a limited space to improve its statistical value, which may result in too optimistic reliability assessment, failing to consider rare but critical faults, and thus may underestimate risks and affect the robustness of system design and operation. So far, the contradiction between the super-large-scale data acquisition quantity and the data natural characteristics is not solved, and meanwhile, an accurate data distribution model needs to be built urgently so as to carry out comprehensive reliability evaluation on a wind power generation system. A long short-term memory (LSTM) model is used for filling the research blank, and wind power generation is predicted by defining data with different granularities. On the basis of differential granularity data driving, an improved Latin hypercube sampling method is used for dividing sampling intervals, and a Monte Carlo method is combined for sampling wind power generation data. According to the reliability evaluation model, through a flexible sampling architecture of an improved Latin hypercube sampling method, probability distribution representation, interval division and sample layering are enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grid, and relates to a reliability assessment method for a wind power integrated distribution network based on differential granularity data-driven. Background Art

[0002] The wind power generation system is a random and intermittent power generation system, and its active power is difficult to control and predict. Although doubly-fed and permanent magnet direct drive wind turbines have developed rapidly and gradually become the main wind energy equipment, most domestic wind farms are far from the load center. With the increase in the scale of wind farms and the single capacity of wind power generation equipment, the reliability problem of the power system integrated with wind power needs to be solved urgently. The goal of power system reliability assessment is to quantitatively evaluate the reliability of the power grid, so as to guide the planning and operation of the power grid.

[0003] When evaluating the reliability of a wind power generation system, it is necessary to perform differential sampling and accurately predict a large amount of data. Existing methods often limit the original data in a limited space to improve its statistical value, which may lead to an overly optimistic reliability assessment and fail to consider rare but critical faults, thus possibly underestimating risks and affecting the robustness of system design and operation. So far, the contradiction between the ultra-large-scale data acquisition volume and the natural characteristics of data has not been solved, and there is an urgent need to establish an accurate data distribution model to comprehensively evaluate the reliability of the wind power generation system. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a reliability assessment method for a wind power integrated distribution network based on differential granularity data-driven, which can overcome the deficiencies of the prior art and maintain the comprehensiveness of the reliability assessment of the wind power system while improving the sampling efficiency.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A reliability assessment method for a wind power integrated distribution network based on differential granularity data-driven, the method includes two models:

[0007] Model A: Differential Granularity Model

[0008] S1: Initialization: The original wind energy data sequence;

[0009] S2: Data normalization, set the range to [0, 1];

[0010] S3: Calculate the standard deviation of the normalized data in each window through a sliding window to obtain a sliding standard deviation sequence;

[0011] S4: Dynamically adjust the granularity of the time interval according to the calculation result of the sliding standard deviation: when the sliding standard deviation is large, select a smaller granularity; when the sliding standard deviation is small, select a larger granularity. When the sliding standard deviation is in a medium range, select a moderate granularity;

[0012] S5: Provide the normalized wind energy data to the LSTM model as input for model training and prediction.

[0013] Model B: A reliability evaluation model combining the improved Latin hypercube sampling method and the Monte Carlo method

[0014] Select the probability of load shedding and the expected value of power outage as the basis and criterion for evaluating reliability. The calculation formulas are as follows:

[0015] Probability of load shedding (Loss of Load Probability, LOLP):

[0016]

[0017] where N is the number of sampling times, and F LOLP (X i ) is the number of load shedding times.

[0018] Expected value of power outage not supplied (Expected Power Not Supplied, EPNS):

[0019]

[0020] where F EPNS (X i ) is the amount of load shedding.

[0021] First, establish a Monte Carlo model. Assume that a power network system has n nodes or buses, and the system state vector is X = (X1, X2,..., X k ,..., X n ). The state formula of the k-th component is:

[0022]

[0023] Its probability distribution is:

[0024]

[0025] Assume the reliability index test function F(X) and the system joint probability distribution function P(X). The expression of P(X) is:

[0026]

[0027] Then the expected value E(F) and variance V(F) of the system sample are as follows:

[0028] E(F) = ∑F(X)P(X) (6)

[0029] V(F) = ∑[F(X) - E(F)] 2 P(X) (7)

[0030] In actual sampling, only the estimated value of the expected value can be obtained and the estimated value of the variance

[0031]

[0032]

[0033] where N is the total number of samplings, and F i (X) is the test function of the i-th sampling.

[0034] When calculating LOLP, the number of load shedding times is marked as F LOLP (X i ):

[0035]

[0036] When calculating EPNS, the system load shedding amount is marked as F EPNS (X i ):

[0037]

[0038] where P Lj is the total load, and P Gi is the total generator capacity.

[0039] For the statistics of LOLP and EPNS in each sub-interval, the estimated value of the expected value is:

[0040]

[0041] The estimated values of the variances of LOLP and EPNS are respectively:

[0042]

[0043] Taking the variance coefficient β as the convergence basis, its expression is:

[0044]

[0045] where is the variance of the estimated value of the sample expected value of.

[0046] The coefficient of variation of LOLP and EPNS are respectively as follows:

[0047]

[0048]

[0049] Furthermore, it is obtained that:

[0050]

[0051] Subsequently, the Latin hypercube sampling model is improved and the Monte Carlo model is optimized. According to the importance sampling law, P*(X) is used to replace P(X) to increase the occurrence probability of power system failure events. The calculation formula of P*(X) is:

[0052]

[0053] Where α is the optimal multiplier. The iterative method is used for solving, and it ends when α of two adjacent times meets the accuracy. The calculation formula is:

[0054]

[0055] Then for P * (X), F*(X LOLP ) and F*(X EPNS ) are:

[0056]

[0057] Where is the average value of the failure probabilities of all components, N1 is the set of the number of components in the failure state, n1 is the number of components in the failure state in the sampling, N0 is the set of the number of components in the working state, and n0 is the number of components in the working state in the sampling.

[0058] For the statistics of each sub-interval, the expected estimated value is:

[0059]

[0060] The variance estimated values of LOLP and EPNS are respectively:

[0061] Description of the Drawings

[0062] In order to make the objectives, technical solutions and beneficial effects of the present invention clearer, the following drawings are provided for the description of the present invention:

[0063] Figure 1It is a flow chart for the reliability assessment of a wind power integrated distribution network driven by differential granularity data.

[0064] Figure 2 It is a flow chart predicted by the LSTM algorithm. Specific implementation manners

[0065] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than limiting the protection scope of the present invention.

[0066] In order to implement the above invention, the present invention provides a prediction model and a process for the reliability assessment of a wind power integrated distribution network.

[0067] 1. LSTM prediction model

[0068] LSTM is an advanced recurrent neural network designed to overcome the vanishing gradient problem faced by traditional recurrent neural networks when dealing with long-term dependencies. LSTM achieves this through its unique memory cell structure, which enables it to store and selectively forget information over time. The storage unit contains information updated or retained based on three key gates:

[0069] (1) The input gate controls how much current input is added to the storage unit. It considers the current input and the previous hidden state to generate a value between 0 and 1, thereby determining how much new information should be stored.

[0070] (2) The forget gate regulates which parts of the memory unit should be discarded or retained. It decides the importance of the previous memory based on the current input and the previous hidden state, determining whether the information is retained or forgotten.

[0071] (3) The output gate controls how much of the content of the storage unit affects the output of the LSTM unit. It calculates a value that determines the relevance of the stored information to the current output.

[0072] The interaction of these gates allows LSTM to manage long-term dependencies by preventing irrelevant information from interfering with the learning process, making it suitable for time series prediction tasks such as wind power generation.

[0073] 2. Reliability assessment process

[0074] The key to the reliability assessment method proposed by the present invention lies in combining model A and model B. The specific process is as follows:

[0075] Step 1: Initialize parameters and read in the original data of the wind farm system;

[0076] Step 2: Process the original wind power data set according to model A to obtain the granularity set G;

[0077] Step 3: Use G in Step 2 as input to predict wind power output based on the LSTM model;

[0078] Step 4: Sample the state of the distribution network system with wind power based on the improved Latin hypercube sampling to optimize the Monte Carlo method;

[0079] Step 5: Conduct a state analysis of the system and use the optimal load shedding model to shed load;

[0080] Step 6: Statistically analyze the changes in relevant parameters such as the number of load shedding times of the system and calculate the reliability index.

[0081] Step 7: If the end condition is met, output the reliability index; otherwise, return to Step 4.

[0082] Step 8: Evaluate different data sets in the granularity set G based on the sampling efficiency and reliability index, and select the most suitable data granularity as the standard for subsequent evaluation of this wind farm.

[0083] The innovation of the present invention lies first in proposing a differential granularity model to perform prior processing on data, establishing data distribution models with different granularities before reliability assessment, and improving the statistical value; secondly, based on the importance sampling method, the traditional Latin hypercube sampling method is improved, and combined with the Monte Carlo method to establish a more efficient and accurate reliability assessment model.

[0084] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A reliability assessment method for a wind power distribution network driven by differential granularity data, characterized by: The wind energy data is processed by the differential granularity model and predicted using the LSTM model; the reliability of the wind power system is evaluated by combining the improved Latin hypercube sampling model with the Monte Carlo model. The two models are as follows: Model A: Differential Granularity Model S1: Initialize the original wind energy data sequence; S2: data normalization, setting the range to [0,1]; S3: Calculate the standard deviation of the normalized data in each window through the sliding window to obtain a sliding standard deviation sequence; S4: According to the calculation result of the sliding standard deviation, the particle size is dynamically adjusted at time intervals: when the sliding standard deviation is large, a smaller particle size is selected; when the sliding standard deviation is small, a larger particle size is selected. When the sliding standard deviation is in the medium range, a moderate particle size is selected; S5: Provide the normalized wind energy data to the LSTM model as input for model training and prediction. Model B: Reliability assessment model based on improved Latin hypercube sampling method combined with Monte Carlo method The load shedding probability and power outage power expectation are selected as the basis and criteria for evaluating reliability. The calculation formula is as follows: Loss of Load Probability (LOLP): Where N is the number of samples, F LOLP (X i ) is the number of load shedding times. Expected Power Not Supplied (EPNS): Among them, F EPNS (X i ) is the load shedding amount. First, a Monte Carlo sampling model is established. Assume that a power grid system has n nodes or buses, and the system state vector is X = (X1, X2, ..., X k ,...,X n ), the state formula of the kth component is: Its probability distribution is: Assuming the reliability index test function F(X) and the system joint probability distribution function rate P(X), the expression of P(X) is: Then the expectation E(F) and variance V(F) of the system sample are: E(F)=∑F(X)P(X) V(F)=∑[F(X)-E(F)] 2 P(X) In actual sampling, only the expected estimate can be obtained and variance estimates Where N is the total number of samples, F i (X) is the test function of the i-th sampling. When calculating LOLP, the number of load shedding is marked as F. LOLP (X i ): When calculating EPNS, the system load shedding is marked as F EPNS (X i ): Among them, P Lj is the total load, P Gi is the total number of generators. For the LOLP and EPNS statistics in each subinterval, the expected estimate is for: Variance estimates of LOLP and EPNS They are: Taking the variance coefficient β as the convergence basis, its expression is: in, is the sample expected value The variance of . Then the variance coefficients of LOLP and EPNS are: Further conclusion: Subsequently, the Latin hypercube sampling model was improved and the Monte Carlo model was optimized. According to the important sampling law, P*(X) replaces P(X) to increase the probability of events that lead to power system failure. The calculation formula of P*(X) is: Among them, α is the optimal multiplier. The iterative method is used to solve, and the solution ends when two consecutive αs meet the accuracy. The calculation formula is: Then for F*(X) under P*(X) LOLP ) and F*(X EPNS )for: in, is the average value of the failure probability of all components, N1 is the set number of components in failure state, n1 is the number of components in failure state in the sample, N0 is the set number of components in working state, n0 is the number of components in working state in the sample. For each subinterval, the expected estimate is for: Variance estimates of LOLP and EPNS They are:

2. The reliability assessment method for a wind power distribution network driven by differential granularity data according to claim 1 is characterized in that: The data processed by model A is used as the input of model B to evaluate the reliability of the distribution network containing wind power.

Images