New energy power distribution system node voltage uncertainty quantification method and device, terminal and medium

Through the combination of Bootstrap method and bidirectional long and short-term memory neural network, the voltage-power sensitivity matrix is ​​constructed, which solves the problem of unclear physical significance and insufficient interpretability of the existing voltage uncertainty quantization method, and realizes the accurate quantification of photovoltaic output and Effective analysis of voltage uncertainty improves the stability and prediction accuracy of the distribution system.

CN118944057APending Publication Date: 2024-11-12STATE GRID ZHEJIANG ELECTRIC POWER COMPANY TAIZHOU POWER SUPPLY +1
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202410965092.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-07-12
Filing Date
2024-07-18
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing voltage uncertainty quantification method affects the stable operation of the power distribution system when the physical significance is unclear, insufficient interpretability, and the parameters of the machine learning model are not converged, and the fluctuation characteristics of photovoltaic treatment cannot be fully considered.

Method used

Using the two-way long and short-term memory neural network interval prediction model based on Bootstrap method, the voltage-power sensitivity matrix is ​​constructed, and a linear mapping model from distribution network power uncertainty to node voltage uncertainty is improved to improve interpretability and comprehensive performance indicators The system is compared and analyzed.

Benefits of technology

The non-parametric probability prediction of photovoltaic output is realized, the accuracy and interpretability of the voltage uncertainty model are improved, and it is adapted to different operating scenarios, with higher prediction accuracy and comprehensive performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118944057A_ABST
    Figure CN118944057A_ABST
Patent Text Reader

Abstract

The invention discloses a new energy power distribution system node voltage uncertainty quantification method and device, a terminal and a medium, and the method comprises the steps: building a model for carrying out the deterministic prediction of the photovoltaic output in a new energy power distribution system based on a deep learning model, obtaining a photovoltaic output prediction value, and carrying out the prediction of the photovoltaic output based on the model. A bootstrap method is adopted to realize uncertainty quantification of photovoltaic output; based on voltage-power sensitivity, only considering the influence of photovoltaic active power, constructing a linear mapping relation from photovoltaic output uncertainty to node voltage uncertainty, and solving Gaussian distribution of node voltage deviation; and taking the photovoltaic output predicted value as input, carrying out deterministic power flow operation on the power distribution system to obtain an expected value of the node voltage at the current moment, and superposing the node voltage deviation and the expected value of the voltage to obtain an uncertainty quantification result of the node voltage. According to the method, the uncertainty of the distribution network node voltage is quantified, and compared with a traditional method, the reliability is higher, and the sharpness and the comprehensive performance are better.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network voltage analysis, and in particular to a method, device, terminal and medium for quantifying node voltage uncertainty in a new energy distribution system. Background Art

[0002] Distributed photovoltaic power generation has the characteristics of good environment, flexible operation, easy assembly and grid connection, and has gradually become an important model for power production and clean energy consumption. However, with the increasing penetration rate of new energy such as distributed photovoltaic in low-voltage distribution network, the safety problems such as voltage over-limit and current overload have become more and more serious, seriously affecting the safety and stability of distribution system voltage. Therefore, quantifying the uncertainty of node voltage in new energy distribution system is of great significance for accurately analyzing the temporal and spatial fluctuation characteristics of voltage in the region, and conducting risk assessment and safety warning of distribution system in advance.

[0003] The problem of voltage uncertainty quantification is essentially to establish a mapping relationship between the power and voltage of each node in the distribution network. Many scholars have used probabilistic prediction models to quantify voltage uncertainty. The University of Connecticut team combined the dropout method with an improved graph convolutional neural network to improve the robustness of voltage uncertainty quantification. The University of Mons team used a federated learning model to quantify the voltage uncertainty of the energy community and introduced differential privacy technology in the federated learning framework to achieve privacy protection for community users. However, the existing ideas for voltage uncertainty quantification are to directly build a prediction model from renewable energy output to node voltage, and the prediction method is not interpretable enough. The voltage-power sensitivity matrix can be used to characterize the approximate linear relationship between the change in load and renewable energy output and the change in voltage, and is widely used in distribution system planning, new energy absorption capacity assessment, distributed distribution network voltage control and other fields. How to efficiently obtain the voltage-power sensitivity matrix is ​​also one of the current hot research, but it has not been combined with the problem of voltage uncertainty quantification.

[0004] The main problems with existing methods are: 1) Most existing voltage uncertainty quantification methods directly construct data-driven mapping models from photovoltaic output to node voltage, which have problems such as unclear physical meaning and insufficient interpretability. 2) Existing voltage uncertainty analysis methods will affect the rationality of prediction results when machine learning model parameters have not converged, and will have an impact on the stable operation of the distribution system. 3) The processing of the uncertainty of renewable energy output in the distribution system is relatively simple, such as using direct prediction, model fitting prediction and other methods, which only consider the impact of past information on the current moment, and cannot fully consider the fluctuation characteristics of photovoltaic processing. Summary of the invention

[0005] The present invention provides a method, device, terminal and medium for quantifying node voltage uncertainty in a new energy distribution system. By establishing a bidirectional long short-term memory neural network interval prediction model based on the Bootstrap method, non-parametric probability prediction of new energy output is realized; a linear mapping model from distribution network power uncertainty to node voltage uncertainty is constructed using a voltage-power sensitivity matrix, thereby improving the interpretability of the voltage uncertainty model; an evaluation index system including three different angles of reliability, sharpness and comprehensive performance is constructed, and a more comprehensive comparative analysis of the uncertainty quantification results is carried out.

[0006] In order to achieve the above objectives, in a first aspect, an embodiment of the present invention provides a method for quantifying node voltage uncertainty in a new energy distribution system, comprising the following steps:

[0007] Based on the deep learning model, a model for deterministic prediction of photovoltaic output in the new energy distribution system is constructed to obtain the predicted value of photovoltaic output. Based on the deterministic prediction model, the bootstrap method is used to quantify the uncertainty of photovoltaic output.

[0008] Based on voltage-power sensitivity, only the influence of photovoltaic active power on node voltage is considered, and a linear mapping relationship from photovoltaic output uncertainty to node voltage uncertainty is constructed to obtain the Gaussian distribution of node voltage deviation. Taking the photovoltaic output prediction value as input, deterministic power flow calculation is carried out on the distribution system to obtain the expected value of the node voltage at the current moment. The node voltage deviation is superimposed on the expected value of the voltage to obtain the uncertainty quantification result of the node voltage.

[0009] Furthermore, the method specifically comprises:

[0010] 1) Obtain the photovoltaic input characteristics X, including temperature and photovoltaic historical output, and obtain the original data set D through deterministic power flow calculation, including node voltage U and voltage-power sensitivity matrix S U-P ;

[0011] 2) Construct a deterministic prediction model for photovoltaic output to obtain the predicted value of photovoltaic output at the current moment Combined with the photovoltaic input feature X as the input feature of the deterministic prediction model, the predicted value of the voltage-power sensitivity matrix is ​​obtained

[0012] 3) Based on the deterministic prediction model of PV output, the bootstrap method is used to quantify the uncertainty of PV output and obtain the expected value and variance of power deviation under a given nominal coverage probability;

[0013] 4) Based on the linear mapping relationship from the uncertainty of photovoltaic output to the uncertainty of node voltage, the Gaussian distribution of node voltage deviation is obtained; at the same time, the photovoltaic output prediction value As input, the deterministic power flow calculation is carried out on the distribution system to obtain the expected value of the node voltage at the current moment.

[0014] 5) According to the obtained node voltage deviation and the expected value of the node voltage, based on the calculation process of the node voltage uncertainty, the uncertainty quantification results of the node voltage under different given confidence levels are obtained.

[0015] Furthermore, the deterministic prediction model adopts a bidirectional long short term memory neural network prediction model (bidirectional long short term memory, Bi-LSTM), and the Bi-LSTM includes two types of long short term memory neural network models, forward and backward. The output of the Bi-LSTM model is also obtained by splicing the outputs from two different directions.

[0016] Furthermore, the bootstrap method for quantifying the uncertainty of photovoltaic output specifically includes:

[0017] 1) Combine the input feature X with the photovoltaic output prediction value Combined as training set Where N is the number of samples in the training set, x i is the eigenvalue vector, t i is the photovoltaic output value;

[0018] 2) From D t Randomly resample K times with replacement to obtain the bootstrap sample set

[0019] 3) Use the lth bootstrap sample set to train the Bi-LSTM model, and the output value is

[0020] 4) Repeat step 3) to obtain B M The estimated values ​​are averaged as the true regression value y(x i )

[0021]

[0022] So we get the variance

[0023]

[0024] 5) Based on the results of the above steps, keep the input variable x iNo change, update the output value to Construct new training samples

[0025] 6) Perform random resampling with replacement using the method in step 2 to train B N A new Bi-LSTM model is used to obtain the variance of the data noise error and the corresponding uncertainty variance of the regression model

[0026]

[0027]

[0028] The total variance of the data noise error is for:

[0029]

[0030] 7) In summary, the uncertainty quantification of photovoltaic output is the overall variance for:

[0031]

[0032] Furthermore, the voltage-power sensitivity is derived by power flow calculation based on the Newton-Raphson algorithm, and the voltage-power sensitivity matrix is ​​quickly constructed, including:

[0033] The power flow correction equation is obtained by sorting:

[0034]

[0035] Where J is the Jacobian matrix; It means to find partial derivatives of variables, ΔP and ΔQ are active and reactive matrices respectively, Δθ and ΔU are node phase angle and voltage amplitude matrices respectively;

[0036] Multiply the inverse matrix J of the Jacobian matrix on both sides of the power flow correction equation -1 , sorted out:

[0037]

[0038] Among them, S is the total sensitivity matrix. The node voltage amplitude calculation equation in the above formula is listed separately to obtain:

[0039]

[0040] In the formula, That is the voltage-active power sensitivity; is the voltage-reactive power sensitivity.

[0041] Furthermore, the linear mapping relationship is specifically:

[0042] For a new energy distribution system with a total of M photovoltaics, the variance of the power error of each photovoltaic is obtained With the expected value μ m , where the variance That is the uncertainty quantification result of photovoltaic output, the expected value μ m This is the predicted value of photovoltaic output; according to the linear additivity of independent normal distribution, substitute it into the linear equation ΔU=S U-P ΔP, construct a linear mapping relationship from PV output uncertainty to node voltage uncertainty, and obtain the variance of the voltage at the corresponding node i With the expected value μ i :

[0043]

[0044] Furthermore, at a given confidence level (1-β)*100%, the voltage uncertainty quantification interval I β =[L β ,U β The upper and lower boundaries of ] are expressed as:

[0045]

[0046] Where: z 1-β / 2 It is the quantile corresponding to the confidence level of (1-β)*100% under the standard Gaussian distribution.

[0047] In a second aspect, a device for quantifying uncertainty of node voltage in a new energy distribution system is provided, for implementing any of the methods described, the device comprising:

[0048] The photovoltaic output probability prediction module uses the Bi-LSTM probability prediction model based on the bootstrap method to divide the variance of the overall error into the variance of the observation data noise error and the variance of the model error, thus realizing the uncertainty quantification of the new energy output;

[0049] The node voltage uncertainty quantification module introduces distribution network flow constraints and uses the voltage-power sensitivity matrix to construct a linear mapping relationship from distribution network photovoltaic output uncertainty to node voltage uncertainty, thereby realizing the quantification of distribution network node voltage uncertainty.

[0050] In a third aspect, a terminal device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements any of the methods for quantifying node voltage uncertainty of a new energy distribution system.

[0051] In addition, a computer-readable storage medium is provided, the computer-readable storage medium comprising a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the methods for quantifying node voltage uncertainty of a new energy distribution system.

[0052] Beneficial effects:

[0053] Compared with the prior art, the present invention discloses a method, device, terminal and medium for quantifying node voltage uncertainty of a renewable energy distribution system, establishes a Bi-LSTM probability prediction model based on the Bootstrap method, divides the variance of the overall error into the variance of the observed data noise error and the variance of the model error, and realizes accurate quantification of the uncertainty of renewable energy output; introduces distribution network topology constraints, and uses the voltage-power sensitivity matrix to construct a linear mapping relationship from distribution network power uncertainty to node voltage uncertainty, thereby avoiding the traditional voltage uncertainty quantification method's neglect of the inherent physical characteristics of the distribution network model.

[0054] Therefore, the present invention can quantify the uncertainty of photovoltaic output based on the non-parametric probability prediction method, which can improve the prediction accuracy and adapt to different operating scenarios. Considering the distribution network flow constraints, the interpretability of the voltage uncertainty model is improved. And by establishing a comprehensive evaluation index system, it is verified that the proposed method can achieve effective quantification of the voltage uncertainty of the distribution network node. Compared with the traditional method, it has higher prediction accuracy, better sharpness index and comprehensive performance, and the research has a strong integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of a method for quantifying node voltage uncertainty of a new energy distribution system provided by an embodiment of the present invention;

[0056] Figure 2 It is a specific flow chart of a method for quantifying node voltage uncertainty of a new energy distribution system provided by an embodiment of the present invention;

[0057] Figure 3 It is a structural schematic diagram of a device for quantifying node voltage uncertainty in a new energy distribution system provided by an embodiment of the present invention;

[0058] Figure 4 It is an equivalent circuit diagram of a radial power distribution system provided by an embodiment of the present invention;

[0059] Figure 5 is a structural diagram of a test system provided by an embodiment of the present invention;

[0060] Figure 6 is a node voltage uncertainty quantification result provided by an embodiment of the present invention;

[0061] Figure 7 It is a comparison of evaluation indicators of various models under different nominal coverage probabilities provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] It should be noted that the terms "comprises" and "specifically" and any variations of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0064] The present invention realizes the quantification of voltage uncertainty of distribution network nodes and has higher reliability, better sharpness and comprehensive performance than traditional methods.

[0065] Based on the above advantages, a method for quantifying node voltage uncertainty in a new energy distribution system includes the following steps:

[0066] Based on the deep learning model, a model for deterministic prediction of photovoltaic output in the new energy distribution system is constructed to obtain the predicted value of photovoltaic output. Based on the deterministic prediction model, the bootstrap method is used to quantify the uncertainty of photovoltaic output.

[0067] Based on voltage-power sensitivity, only the influence of photovoltaic active power on node voltage is considered, and a linear mapping relationship from photovoltaic output uncertainty to node voltage uncertainty is constructed to obtain the Gaussian distribution of node voltage deviation. Taking the photovoltaic output prediction value as input, deterministic power flow calculation is carried out on the distribution system to obtain the expected value of the node voltage at the current moment. The node voltage deviation is superimposed on the expected value of the voltage to obtain the uncertainty quantification result of the node voltage.

[0068] Example:

[0069] For details, please refer to Figure 1 , Figure 1 1 is a flow chart of a method for quantifying node voltage uncertainty of a new energy distribution system provided by an embodiment of the present invention, comprising steps S11 to S12:

[0070] S11: Based on the deep learning model, a model for deterministic prediction of photovoltaic output in the renewable energy distribution system is constructed to obtain the predicted value of photovoltaic output. Based on the deterministic prediction model, the bootstrap method is used to quantify the uncertainty of photovoltaic output.

[0071] S12: Based on voltage-power sensitivity, only the impact of photovoltaic active power on node voltage is considered, and a linear mapping relationship from photovoltaic output uncertainty to node voltage uncertainty is constructed to obtain the Gaussian distribution of node voltage deviation; with the photovoltaic output prediction value as input, deterministic power flow calculation is carried out on the distribution system to obtain the expected value of the node voltage at the current moment, and the node voltage deviation is superimposed on the expected value of the voltage to obtain the uncertainty quantification result of the node voltage.

[0072] It should be noted that probabilistic prediction technology:

[0073] Probabilistic prediction technology can provide the probability distribution information of the prediction object under certain conditions. Compared with traditional deterministic prediction, it can provide richer data information and realize the effective quantification of the uncertainty of photovoltaic output. It is more in line with objective needs than classical deterministic prediction. There are two main ways to implement the probability prediction of photovoltaic output: one is to first obtain the probability distribution result of the photovoltaic output prediction error, and then superimpose it with the deterministic prediction result to obtain the complete probability distribution of the prediction object; the other is to directly model the probability distribution of the prediction object without relying on the analysis results of the statistical characteristics of the prediction error. According to whether it is necessary to give a priori parameter distribution hypothesis of the prediction result, it can be divided into parametric method and non-parametric method. Compared with parametric method, non-parametric method does not rely on any prior knowledge, but is based on the statistical characteristics of uncertainty factors characterized by historical data. Therefore, it can more accurately represent the upper and lower bounds of new energy output, multi-peak distribution and other characteristics, and its model has better generalization ability. The existing non-parametric probability prediction methods mainly include quantile regression (QR), Gaussian process regression (GPR), and Bayesian neural networks (BNN).

[0074] Bootstrap method:

[0075] Bootstrap is a statistical inference method based on resampling. It does not require assumptions about the overall distribution. Instead, it performs equal-probability random sampling with replacement on the overall sample and estimates the distribution characteristics of the population through multiple independent sub-samples to achieve prediction uncertainty simulation. It has been widely used in parameter estimation, hypothesis testing, probability prediction and other fields.

[0076] Specifically, in step S11, the bidirectional long short-term memory neural network prediction model includes two types of long short-term memory neural network models: forward and backward. The long short-term memory neural network model is a variant of the recurrent neural network (RNN), which better captures the long-distance dependency features in the time series data by introducing a gating structure, and avoids the gradient explosion and gradient vanishing problems caused by the conventional RNN method.

[0077] The basic structural unit of the long short-term memory neural network model includes an input gate i t 、Forget gate t , output gate o t and cell memory unit c t Etc. Among them, the forget gate is responsible for controlling the previous cell memory unit c t-1 The degree of information retention in the input gate controls the current input x t After candidate memory unit g t How to pass it to the memory cell c of the current time slice t If the forget gate is always 1 and the input gate is always 0, then c t-1 The information will be saved and passed on. This design can effectively deal with the gradient decay problem of the RNN network and improve the ability to capture long-distance dependencies in the sequence.

[0078] The information update process of the long short-term memory neural network is:

[0079]

[0080] In the formula, x t is the input value of the current unit; σ represents the activation function. The forget gate, input gate, and output gate use the Sigmoid function as the activation function to control the corresponding element value between 0 and 1, which meets the gate control requirements; when generating candidate memories, the hyperbolic tangent function tanh is used as the activation function to ensure that the candidate memory element value is between -1 and 1; g t is the state quantity of the candidate memory unit; b j (j=f,g,t,o) is the bias term; W j (j=f,g,t,o) is the corresponding weight matrix; h t is the output value of the hidden layer; (·) represents the inner product operation.

[0081] It should be noted that each weight matrix and bias term can be trained and learned by the gradient descent method. However, the LSTM model only considers the one-way transmission of information and cannot consider the impact of future data on the current moment. In order to fully consider the fluctuation characteristics of photovoltaic output, a bidirectional long short-term memory neural network prediction model (Bi-LSTM) is constructed to supplement the update equation of the hidden layer state quantity. At time t, the forward model can be considered at the same time. Backward Model For h t Impact:

[0082]

[0083] The output of the Bi-LSTM model is also obtained by splicing outputs from two different directions, which can simultaneously consider the impact of past information and future information on the current moment, further improving the efficiency and performance of feature extraction.

[0084] More specifically, the Bi-LSTM model based on the Bootstrap method realizes the uncertainty quantification of photovoltaic output, and the variance of the model error is obtained respectively. and the variance of the data noise error Assuming that the two error components are statistically independent, the variance of the total forecast error is It can be expressed as the sum of the two, and superimposed with the predicted value to obtain a prediction interval that meets a given confidence level. The specific steps include:

[0085] 1) Combine the input feature X with the photovoltaic output prediction value Combined as training set Where N is the number of samples in the training set, x i is the eigenvalue vector, t i is the photovoltaic output value;

[0086] 2) From D t Randomly resample K times with replacement to obtain the bootstrap sample set

[0087] 3) Use the lth bootstrap sample set to train the Bi-LSTM model, and the output value is

[0088] 4) Repeat step 3) to obtain B M The estimated values ​​are averaged as the true regression value y(x i )

[0089]

[0090] So we get the variance

[0091]

[0092] 5) Based on the results of the above steps, keep the input variable x i No change, update the output value to Construct new training samples

[0093] 6) Perform random resampling with replacement using the method in step 2 to train B N A new Bi-LSTM model is used to obtain the variance of the data noise error and the uncertainty variance of the regression model

[0094]

[0095]

[0096] The total variance of the data noise error is It can be calculated as:

[0097]

[0098] 7) In summary, the uncertainty quantification of photovoltaic output is the overall variance for:

[0099]

[0100] Specifically, in step S12, the voltage-power sensitivity specifically refers to the influence of the change of active and reactive power of the node on the voltage amplitude of each node. When the topology structure and line parameters of the distribution network are known, the voltage-power sensitivity can be derived by performing power flow calculation based on the Newton-Raphson algorithm. The derivation process is as follows:

[0101] See also Figure 4 , Figure 4 is an equivalent line diagram of a radial power distribution system provided by an embodiment of the present invention. ij ,Q ij ,G ij ,B ij ,I ij Respectively represent the active and reactive power, conductance, susceptance, and current on line ij; Y i0 =G i0 +jB i0 represents the ground admittance of node i, G i0 With B i0 They represent the conductance and susceptance of node i to ground respectively.

[0102] First, the AC power flow expression in the polar coordinate system of the distribution system is constructed as follows:

[0103]

[0104] Where: U i represents the voltage at node i; θ ij =θ i -θ j represents the phase difference between nodes i and j. The node voltage loss can be derived as:

[0105]

[0106] Where: R ij ,X ij are the equivalent resistance and reactance of the line respectively. It can be seen that the voltage difference of the distribution network node is strongly correlated with the power transmitted by the line and the line parameters. In the high-voltage transmission line, the resistance is much smaller than the reactance, that is, R ij <<X ij , the effect of active power on voltage can often be ignored. However, in the distribution network, the lines are thin and the resistance is obvious, so the effect of active power on voltage drop cannot be ignored.

[0107] It should be noted that if Figure 4 As shown in the figure, the network topology of the distribution system is usually radial, and the nodes connected to the transmission network are regarded as balanced nodes. In the distribution network that only considers the power load, the voltage gradually decreases toward the user side. When the node is connected to photovoltaic power, the photovoltaic output will meet the load requirements of some users, and power reverse transmission may even occur during the noon period. At this time, the node will output power to the outside, which will cause a certain increase in the node voltage.

[0108] In order to further study the influence of power on voltage, the power flow correction equation is derived as follows:

[0109]

[0110] Arranged as:

[0111]

[0112] Where: J is the Jacobian matrix; Indicates partial derivative of variables, ΔP and ΔQ are active and reactive matrices, Δθ and ΔU are node phase angle and voltage amplitude matrices, respectively. The expressions of each element of the Jacobian matrix are:

[0113] 1) When i=j:

[0114]

[0115] 2) When i≠j:

[0116]

[0117] Multiply the inverse matrix J of the Jacobian matrix on both sides of the power flow correction equation -1 , we can get:

[0118]

[0119] Among them, S is the total sensitivity matrix. The node voltage amplitude calculation equation in the above formula is listed separately to obtain:

[0120]

[0121] In the formula, That is the voltage-active power sensitivity; is the voltage-reactive power sensitivity. From the sensitivity calculation process, we can see that active power, reactive power, node voltage and line parameters will affect the voltage-power sensitivity of the distribution system.

[0122] More specifically, the calculation process of the node voltage uncertainty is:

[0123] For a new energy distribution system with M photovoltaics, the variance of the power error of each photovoltaic can be obtained by constructing a probability prediction model With the expected value μ m , where the variance That is the uncertainty quantification result of photovoltaic output, the expected value μ m This is the predicted value of photovoltaic output; according to the linear additivity of independent normal distribution, substitute it into the linear equation ΔU=S U-P ΔP, construct a linear mapping relationship from PV output uncertainty to node voltage uncertainty, and obtain the variance of the voltage at the corresponding node i With the expected value μ i for:

[0124]

[0125] At a given confidence level (1-β)*100%, the voltage uncertainty quantification interval I β =[L β ,U β The upper and lower boundaries of ] are expressed as:

[0126]

[0127] Where: z 1-β / 2 It is the quantile corresponding to the confidence level of (1-β)*100% under the standard Gaussian distribution.

[0128] like Figure 2 According to a specific embodiment of the present invention, the overall process of quantifying the node voltage uncertainty may be:

[0129] 1) Obtain the voltage-power sensitivity prediction model in advance and construct a linear mapping relationship from PV output uncertainty to node voltage uncertainty; obtain PV input features X, including temperature, PV historical output, etc., and obtain the original data set D through deterministic power flow calculation, including node voltage U and voltage-sensitivity matrix S U-P .

[0130] 2) Construct a deterministic photovoltaic output prediction model to obtain the photovoltaic output prediction value at the current moment Combined with the photovoltaic input feature X as the input feature of the deterministic prediction model, the predicted value of the voltage-power sensitivity matrix is ​​obtained

[0131] 3) Based on the deterministic prediction model of PV output, the bootstrap method is used to quantify the uncertainty of PV output, and the expected value and variance of power deviation under a given nominal coverage probability are obtained. At this point, the direct data-driven steps have been completed.

[0132] 4) Based on the linear mapping relationship from the uncertainty of photovoltaic output to the uncertainty of node voltage, the Gaussian distribution of node voltage deviation is obtained; at the same time, the photovoltaic output prediction As input, the deterministic power flow calculation is carried out on the distribution system to obtain the expected value of the node voltage at the current moment.

[0133] 5) According to the obtained node voltage deviation and the expected value of the node voltage, node voltage uncertainty quantification results under different given confidence levels are obtained based on the calculation process of the node voltage uncertainty.

[0134] See also Figure 5 , Figure 5 It is a test system structure diagram provided by an embodiment of the present invention. In a specific embodiment, the present invention selects the modified IEEE-33 node distribution system as the research object. The system parameters and line parameters of the distribution system are taken from the standard example set of Matpower toolbox of Matlab software. The total active power demand of the system is 3.715MW. In order to make the voltage fluctuation relatively obvious, a total of six large-capacity photovoltaic power generation systems are connected. The access locations and installed capacities are shown in Table 1.

[0135] Table 1 Photovoltaic equipment configuration

[0136]

[0137] The Adam optimizer is used as the deterministic prediction model for photovoltaic output and voltage-power sensitivity. Other parameter settings are shown in Table 2. The prediction of voltage-power sensitivity only needs to consider nodes equipped with photovoltaics.

[0138] Table 2 Deterministic prediction model parameters

[0139]

[0140] The accuracy of deterministic prediction is analyzed by introducing the root mean square error (RMSE). The mathematical expression of RMSE is:

[0141]

[0142] Where: y t and and are the actual value and predicted result of the tth sampling point of the test sample, respectively. test is the number of test samples. In the deterministic prediction effect evaluation, the smaller the RMSE value, the higher the prediction accuracy of the prediction model. The extreme learning machine (ELM) single hidden layer neural network prediction model is used as a comparative example, and the prediction errors under different lead times are recorded as shown in Table 3. Among them, the output data dimensions of the voltage-power sensitivity prediction are relatively large, and the maximum root mean square error is selected for analysis.

[0143] Table 3 Accuracy analysis of deterministic prediction models

[0144]

[0145] It can be found from Table 3 that the deterministic prediction results of the six photovoltaic power generation systems have comparable accuracy, proving that the output of photovoltaic power stations in the same region is affected by similar climatic conditions and their output characteristics are similar. Compared with the extreme learning machine method, the Bi-LSTM method has a smaller root mean square error, proving that this method has higher prediction accuracy.

[0146] See also Figure 6 , Figure 6 It is the node voltage uncertainty quantification result provided by the embodiment of the present invention.

[0147] In order to avoid wasting performance resources and make the display results more intuitive, the redundant data set with 0 photovoltaic output is removed. The red line in the figure is the voltage reference value, and the blue areas with different transparency represent the prediction results of different confidence intervals. The darkest and narrowest color is the 10% confidence interval, and the confidence intervals are 20%, 30% to 90% in order. Figure 6 It can be seen that the prediction interval basically covers the true value, and the fluctuation of the interval can accurately reflect the changing trend of the true value.

[0148] Unlike deterministic prediction models, which can directly evaluate prediction performance by the error between the actual value and the predicted value of the test sample, it is difficult to obtain the actual probability distribution of the prediction object in practical applications. Therefore, the performance comparison of uncertainty quantification models often requires the use of general evaluation indicators. The present invention establishes an evaluation indicator system including three different angles of reliability, sharpness, and comprehensive performance, and conducts a more comprehensive comparative analysis of the uncertainty quantification results:

[0149] More specifically, the reliability index is the most important evaluation index for uncertainty quantification, representing the accuracy of the uncertainty quantification result. For the interval probability prediction model, the average coverage deviation (ACD) is introduced to characterize this characteristic:

[0150]

[0151] Where: At a given confidence level (1-β)*100%, P EC The actual value of the variable to be predicted falls within the prediction range. The empirical coverage probability (ECP) within x t and t are the predicted input and output at time t respectively; is the nominal quantile level of the upper and lower boundaries of the probability prediction interval; q α is the quantile value corresponding to the nominal quantile level α; is the nominal coverage probability (NCP); T is the number of samples in the training set; ψ{A} is the indicative function, which is 1 when condition A is met and 0 otherwise. AC The effect is highest when it is close to 0, and prediction results with poor reliability will increase the decision-making risk of the system.

[0152] The sharpness index can describe the concentration of uncertainty quantification results. For probability prediction models with probability density function as output, high sharpness performance is manifested as the "peak feature" of probability density function; for probability prediction models in quantile form, high sharpness performance is manifested as smaller quantile value difference under the same nominal quantile level interval; for probability prediction results in output interval form, high sharpness performance is manifested as narrower prediction interval width under the same nominal coverage probability. For interval probability prediction methods, the average interval width is often used. (average width, AW) as a quantitative evaluation indicator:

[0153]

[0154] Probabilistic prediction results with higher sharpness performance have higher application value in actual scenarios, but if the reliability index of the prediction method is very poor, even if the average interval width is very small, it cannot prove the effectiveness of the method. When there is no obvious gap in the reliability index, too wide acuity will make the decision result conservative and sacrifice a certain degree of system operation economy.

[0155] The comprehensive performance score indicators include pinball loss score (PLS), quantile skill score (QSS), continuous ranked probability score (CRPS), Winkler score, etc.

[0156] The pinball loss score is a performance evaluation based on the loss function of the quantile regression model. However, the PLS method can only evaluate the comprehensive performance of the prediction results at a certain quantile level and cannot be applied to performance evaluation at multiple quantile levels. The application scenarios are limited. The quantile performance score is also for the uncertainty quantization results in the form of quantile output (or the uncertainty quantization results in other forms are first quantized into quantiles and then comprehensive performance evaluation is performed). It can comprehensively evaluate the performance at different quantile levels and is an improvement on the pinball loss score method. For the quantile level α i The predicted quantile of (i=1,2,…,N) Quantile Performance Score S Q Defined as:

[0157]

[0158] S Q The larger it is, the better the comprehensive performance of the uncertainty quantification method at different quantile levels is.

[0159] The uncertainty quantification results in the form of probability distribution are usually evaluated by using the continuous graded probability score CRPS. This method uses the mean square value of the deviation between the cumulative probability distribution function result of the predicted variable and the "actual" probability distribution (replaced by the step function of the actual value of the predicted object) as the evaluation index, which can be analogous to the mean square error index in the deterministic prediction evaluation index. For a given prediction object, the prediction cumulative probability function is F(y t ), continuous graded probability score S C Defined as:

[0160]

[0161] Where: F o (y t ) is the observed cumulative probability function. SC The smaller the value, the better the overall performance of the uncertainty model. Ideally, S C is equal to zero.

[0162] For the uncertainty quantification results in the form of prediction intervals, the Winkler score is usually used as a comprehensive performance evaluation indicator. This method can comprehensively consider the reliability and sharpness performance of the prediction model, which is defined as:

[0163]

[0164] The Winkler score is essentially a penalty for data that does not meet the specified interval based on the sharpness index. It is only applicable to comparing prediction intervals with the same quantile interpretation of endpoints under the same nominal coverage probability. W The smaller the value, the better the overall performance of the uncertainty quantification results.

[0165] See also Figure 7 , Figure 7 It is a comparison of evaluation indicators of various models under different nominal coverage probabilities provided by an embodiment of the present invention.

[0166] It should be noted that in order to further verify the performance of the proposed method, the reliability index D is introduced AC , sharpness index AW and Winkler score comprehensive evaluation index S W , and construct four comparative examples:

[0167] 1) Using the same voltage uncertainty quantification idea as in this paper, but using a different probability prediction model:

[0168] ①Persistence method, also a parameterized probability prediction method;

[0169] ② Considering the power flow constraints of the distribution system, the physical Bootstrap-based ELM (P-BELM) is used as the photovoltaic output probability prediction model.

[0170] 2) Adopting the traditional voltage uncertainty quantification approach, without considering the inherent physical model constraints of the distribution network, a direct data-driven model from PV historical output to voltage uncertainty is directly constructed.

[0171] ③ Bootstrap-based extreme learning machine method (BELM).

[0172] ④ The traditional voltage uncertainty quantization idea is also adopted, and the improved quantile regression neural network (IQRNN) method is used.

[0173] Depend on Figure 7 (a) It can be seen that under different nominal coverage probabilities, the reliability index of the comparative example BELM and IQRNN based on direct voltage probability prediction is poor, which proves that considering the distribution network flow constraints on the basis of photovoltaic probability prediction can effectively improve the accuracy of voltage uncertainty quantification; Figure 7 (b) It can be seen that both the Persistence prediction model and the IQRNN method will have prediction results with too wide intervals, among which the Persistence prediction model always has the worst sharpness performance, while the improved quantile regression neural network method has the best sharpness performance at low nominal coverage probabilities, but the sharpness performance deteriorates significantly at nominal coverage probabilities of 60% and above, indicating that the model's ability to capture uncertainty is relatively unstable; the comparative examples P-BELM, BELM and the method proposed in the present invention have similar sharpness performance, which is because the three methods are all based on the Bootstrap method for uncertainty analysis, dividing the overall uncertainty into model uncertainty and data uncertainty, and for the same original data set, the data noise error difference is very small, and the difference in sharpness performance basically comes from the uncertainty quantification model error, among which the deep learning model proposed in the present invention has the smallest prediction model error and relatively good sharpness performance, proving that it has a strong ability to capture uncertain information of the prediction object; Figure 7 (c) It can be seen that compared with the two comparative examples of direct voltage probability prediction, the Winkler score of the proposed method under different nominal coverage probabilities is relatively stable and minimal, and the comprehensive performance of the prediction model P-BELM that introduces voltage-power sensitivity is second only to the model proposed in the present invention, which verifies that the voltage uncertainty quantification method proposed in the present invention has better overall performance.

[0174] More specifically, the performance comparison of the uncertainty quantification results of node voltages by different models under 1 hour in advance and 90% nominal coverage probability is shown in Table 4:

[0175] Table 4 Comparison of uncertainty quantification performance under 1 hour lead time and 90% nominal coverage probability

[0176]

[0177] As can be seen from the table, the three index scores of the method proposed in the present invention are all optimal, which proves the stability of the probability prediction interval of the proposed method and the more excellent voltage uncertainty quantification ability of the new energy distribution system. Among them, the Persistence model as the benchmark model has the worst prediction performance and is difficult to provide reliable probability prediction results. The improved quantile regression neural network method combines the quantile regression method with deep learning, and the prediction reliability index is second only to the method proposed in the present invention, but the network model training time of this method is much longer than that of the method proposed in the present invention. The comparative examples P-BELM and BELM both use the Bootstrap-based extreme learning machine prediction model, which also makes the sharpness performance of the two comparable, but the P-BELM method that introduces voltage-power sensitivity has a greater improvement in the reliability of uncertainty quantification compared to direct prediction, and its comprehensive performance is second only to the method proposed in the present invention, which proves the scientific nature of the proposed voltage uncertainty quantification method based on voltage-power sensitivity mapping.

[0178] Figure 3 1 is a schematic diagram of a device for quantifying node voltage uncertainty in a new energy distribution system provided by an embodiment of the present invention, the device comprising:

[0179] The photovoltaic output probability prediction module uses the Bi-LSTM probability prediction model based on the Bootstrap method, which divides the variance of the overall error into the variance of the observation data noise error and the variance of the model error, thus achieving accurate quantification of the uncertainty of renewable energy output.

[0180] The node voltage uncertainty quantification module introduces distribution network flow constraints and uses the voltage-power sensitivity matrix to construct a linear mapping relationship from distribution network power uncertainty to node voltage uncertainty, thereby achieving effective quantification of distribution network node voltage uncertainty.

[0181] A device for quantifying uncertainty in node voltage of a new energy distribution system provided by an embodiment of the present invention can implement all processes of the method for quantifying uncertainty in node voltage of a new energy distribution system of the above-mentioned embodiment. The functions of each module in the device and the technical effects achieved are respectively the same as the functions of the method for quantifying uncertainty in node voltage of a new energy distribution system of the above-mentioned embodiment, and will not be repeated here.

[0182] An embodiment of the present invention provides a terminal device, the terminal device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned embodiment of the method for quantifying the uncertainty of node voltage in a new energy distribution system are implemented. Alternatively, when the processor executes the computer program, the functions of each module in the above-mentioned embodiment of the device for quantifying the uncertainty of node voltage in a new energy distribution system are implemented.

[0183] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a terminal device and does not constitute a limitation on the terminal device. The terminal device may include more or fewer components than shown in the diagram, or may combine certain components, or different components. For example, the terminal device may also include an input / output device, a network access device, a bus, etc.

[0184] The processor may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0185] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card, at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0186] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0187] An embodiment of the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the node voltage uncertainty quantification method of the new energy distribution system as described in the above embodiment.

[0188] In summary, the embodiment of the present invention discloses a method, device, terminal and medium for quantifying the uncertainty of node voltage in a new energy distribution system, establishes a Bi-LSTM probability prediction model based on the Bootstrap method, divides the variance of the overall error into the variance of the observed data noise error and the variance of the model error, and realizes accurate quantification of the uncertainty of the new energy output; introduces the distribution network topology constraint, and uses the voltage-power sensitivity matrix to construct a linear mapping relationship from the distribution network power uncertainty to the node voltage uncertainty, avoiding the traditional voltage uncertainty quantification method from neglecting the inherent physical characteristics of the distribution network model. Therefore, the embodiment of the present invention can quantify the uncertainty of source and load output based on the non-parametric probability prediction method, which can improve the accuracy of the prediction and adjust it to different operating scenarios; considering the distribution network flow constraints, the interpretability of the voltage uncertainty model is improved; finally, by establishing a comprehensive evaluation index system, it is verified that the proposed method can realize the effective quantification of the uncertainty of the distribution network node voltage, and has higher prediction accuracy, better sharpness index and comprehensive performance than the traditional method, and the research has a strong integrity.

[0189] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for quantifying node voltage uncertainty in a new energy distribution system, characterized in that: The following steps are involved: Based on the deep learning model, a model for deterministic prediction of photovoltaic output in the new energy distribution system is constructed to obtain the predicted value of photovoltaic output. Based on the deterministic prediction model, the bootstrap method is used to quantify the uncertainty of photovoltaic output. Based on voltage-power sensitivity, only the influence of photovoltaic active power on node voltage is considered, and a linear mapping relationship from photovoltaic output uncertainty to node voltage uncertainty is constructed to obtain the Gaussian distribution of node voltage deviation. Taking the photovoltaic output prediction value as input, deterministic power flow calculation is carried out on the distribution system to obtain the expected value of the node voltage at the current moment. The node voltage deviation is superimposed on the expected value of the voltage to obtain the uncertainty quantification result of the node voltage.

2. The method for quantifying node voltage uncertainty of a new energy distribution system according to claim 1, characterized in that: The method specifically comprises: 1) Obtain the photovoltaic input characteristics X, including temperature and photovoltaic historical output, and obtain the original data set D through deterministic power flow calculation, including node voltage U and voltage-power sensitivity matrix S U-P ; 2) Construct a deterministic prediction model for photovoltaic output to obtain the predicted value of photovoltaic output at the current moment Combined with the photovoltaic input feature X as the input feature of the deterministic prediction model, the predicted value of the voltage-power sensitivity matrix is ​​obtained 3) Based on the deterministic prediction model of PV output, the bootstrap method is used to quantify the uncertainty of PV output and obtain the expected value and variance of power deviation under a given nominal coverage probability; 4) Based on the linear mapping relationship from the uncertainty of photovoltaic output to the uncertainty of node voltage, the Gaussian distribution of node voltage deviation is obtained; at the same time, the photovoltaic output prediction value As input, the deterministic power flow calculation is carried out on the distribution system to obtain the expected value of the node voltage at the current moment. 5) According to the obtained node voltage deviation and the expected value of the node voltage, based on the calculation process of the node voltage uncertainty, the uncertainty quantification results of the node voltage under different given confidence levels are obtained.

3. The method for quantifying node voltage uncertainty of a new energy distribution system according to claim 2, characterized in that: The deterministic prediction model adopts a bidirectional long short-term memory neural network prediction model (bidirectional long shortterm memory, Bi-LSTM), and the Bi-LSTM includes two types of long short-term memory neural network models, forward and backward. The output of the Bi-LSTM model is also obtained by splicing outputs from two different directions.

4. The method for quantifying node voltage uncertainty of a new energy distribution system according to claim 3, characterized in that: The method of using the bootstrap method to quantify the uncertainty of photovoltaic output specifically includes: 1) Combine the input feature X with the photovoltaic output prediction value Combined as training set Where N is the number of samples in the training set, x i is the eigenvalue vector, t i is the photovoltaic output value; 2) From D t Randomly resample K times with replacement to obtain the bootstrap sample set 3) Use the lth bootstrap sample set to train the Bi-LSTM model, and the output value is 4) Repeat step 3) to obtain B M The estimated values ​​are averaged as the true regression value y(x i ) So we get the variance 5) Based on the results of the above steps, keep the input variable x i No change, update the output value to Construct new training samples 6) Perform random resampling with replacement using the method in step 2 to train B N A new Bi-LSTM model is used to obtain the variance of the data noise error and the corresponding uncertainty variance of the regression model The total variance of the data noise error is for: 7) In summary, the uncertainty quantification of photovoltaic output is the overall variance for:

5. The method for quantifying node voltage uncertainty of a new energy distribution system according to claim 1, characterized in that: The voltage-power sensitivity is derived by the power flow calculation based on the Newton-Raphson algorithm, and the voltage-power sensitivity matrix is ​​quickly constructed, including: The power flow correction equation is obtained by sorting: Where J is the Jacobian matrix; It means to find partial derivatives of variables, ΔP and ΔQ are active and reactive matrices respectively, Δθ and ΔU are node phase angle and voltage amplitude matrices respectively; Multiply the inverse matrix J of the Jacobian matrix on both sides of the power flow correction equation -1 , sorted out: Among them, S is the total sensitivity matrix. The node voltage amplitude calculation equation in the above formula is listed separately to obtain: In the formula, That is the voltage-active power sensitivity; is the voltage-reactive power sensitivity.

6. The method for quantifying node voltage uncertainty of a new energy distribution system according to claim 5, characterized in that: The linear mapping relationship is specifically: For a new energy distribution system with a total of M photovoltaics, the variance of the power error of each photovoltaic is obtained With the expected value μ m , where the variance That is the uncertainty quantification result of photovoltaic output, the expected value μ m This is the predicted value of photovoltaic output; according to the linear additivity of independent normal distribution, substitute it into the linear equation ΔU=S U-P ΔP, construct a linear mapping relationship from PV output uncertainty to node voltage uncertainty, and obtain the variance of the voltage at the corresponding node i With the expected value μ i :

7. The method for quantifying node voltage uncertainty of a new energy distribution system according to claim 6, characterized in that: At a given confidence level (1-β)*100%, the voltage uncertainty quantification interval I β =[L β ,U β The upper and lower boundaries of ] are expressed as: Where: z 1-β / 2 It is the quantile corresponding to the confidence level of (1-β)*100% under the standard Gaussian distribution.

8. A device for quantifying node voltage uncertainty in a new energy distribution system, characterized in that: For implementing the method according to any one of claims 1 to 7, the device comprises: The photovoltaic output probability prediction module uses the Bi-LSTM probability prediction model based on the bootstrap method to divide the variance of the overall error into the variance of the observation data noise error and the variance of the model error, thus realizing the uncertainty quantification of the new energy output; The node voltage uncertainty quantification module introduces distribution network flow constraints and uses the voltage-power sensitivity matrix to construct a linear mapping relationship from distribution network photovoltaic output uncertainty to node voltage uncertainty, thereby realizing the quantification of distribution network node voltage uncertainty.

9. A terminal device, characterized in that: It comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the method for quantifying node voltage uncertainty of the new energy distribution system as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the method for quantifying node voltage uncertainty of a new energy distribution system as described in any one of claims 1 to 7.

Citation Information

Cited By

  • Method and system for monitoring battery health state of uninterruptible power supply

    CN116385956A

  • Method and system for battery state of health monitoring of an uninterruptible power supply

    CN116385956B

  • Voltage traceability method based on data-driven power flow

    CN119627938A

  • Construction method, device and equipment of voltage segmentation and layering calculation architecture and medium

    CN120671301A

  • Photovoltaic power prediction method and system, electronic equipment and medium

    CN121036015A